Open Problems in the Emergence and Evolution of Linguistic ...

Open Problems in the Emergence and Evolution of
Linguistic Communication: A Road-Map for Research
Chrystopher L. Nehaniv
Adaptive Systems Research Group
University of Hertfordshire
Hatfield Herts AL10 9AB
United Kingdom
C.L.Nehaniv@herts.ac.uk
Abstract
This paper surveys issues around several grand challenge problems for the understanding of the emer-
gence and evolution linguistic communication, and discusses possible approaches. The identified prob-
lems the emergence of (1) advanced use of deixis, gesture, and reference; (2) predication; (3) negation;
(4) syntactic categories; and (5) compositionality.
1 Introduction
searchers could not be mentoned here. The discus-
sion is instead indicative of current research activity
In the last decade or so, there has been an explo-
(and inactivity) as regards a set of fundamental prob-
sion of interest in the modelling and understand-
lems in the area.
ing of language origins. The employment of simu-
We will discuss the following completely or
lation and robotic agent-based, connectionist neural
largely open areas:
network, and evolutionary techniques has provided
new methods for formulating hypotheses, validating
(1) deixis, gesture, and reference;
mechanisms, and selecting between alternative the-
(2) predication;
ories on the emergence of linguistic and language-
(3) negation;
like phenomena in controlled experimental settings
(4) emergence of syntactic categories
that meet the scientific criteria of reproducibility. Re-
(5) compositionality
cent work on the emergence and evolution of human
language and more simple communication systems
has been increasingly interdisciplinary, involving col-
The emergence and modelling of these phenomena
laborations between linguists, philosophers, biolo-
are discussed in the context of embodied, social inter-
gists, cognitive scientists, roboticists, mathematical
action and evolution (cultural or otherwise). Ideally,
and computational modellers – see e.g. research pa-
mechanisms based on sensorimotor and experiential
pers (MacLennan, 1992; Steels, 1995; Hashimoto and
grounding in bottom-up, agent-centered models in-
Ikegami, 1995; Arita and Koyama, 1998; Billard and
volving populations of agents will help yield deep un-
Dautenhahn, 1999; Kirby, 1999; Nehaniv, 2000; Can-
derstanding of the emergence of the above phenom-
gelosi, 2001; Steels, 2003) and interdisciplinary col-
ena.
lections (Wray, 2002; Cangelosi and Parisi, 2002;
One area is conspicuously missing from the above
Christiansen and Kirby, 2003).
list:
This paper surveys some currently open problems
in the emergence and evolution of linguistic commu-
(0) grounding and shared vocabularies
nication that present grand challenges to those work-
ing in constructive aspects of the emergence of com-
and will also be discussed briefly below. This area
munication. In this paper, we address the programme
has not been included in the list of current grand
of demonstrating mechanisms that achieve various
challenges since there has been substantial progress
language-like properties in computational agent and
in it. However grounding and shared vocabular-
robotic models. This is not intended to be an exhaus-
ies will need to be integrated with the answers to
tive survey. Many important research articles and re-
the grand challenge problem areas (1-5) to yield

---

grounded and shared language-like communication
4. If meanings, spaces of meaning, or syntax in
systems with much more complex types of vocabu-
meaning space do arise, they will be agent-
lary with grounded meaning than what has been so
specific as well.
far achieved.
5. The mappings between signs and meaning are
2 What is Meaning and What is
mediated by interpreted signals between agents,
Language For?
and these mappings are also agent-specific and
depend on the context of the interaction.
We regard linguistic and language-like communica-
tion as the capacity of an agent to influence the world
See (Nehaniv, 1999, 2000) for further discussion of
around by the systematic use of signals mediated by
these points.
their reception by other agents in its environment.
Thus, language is regarded as a means for the agent
Note that none of the above discussion refers to
to ‘manipulate’ the world around for its own benefit,
truth values or truth conditions, which are highly de-
similar to other traits of biological organisms (cf. the
rived properties of human linguistic behaviour (Ne-
discussion of the transition to language from a bio-
haniv, 2000), and that therefore should not be the
logical viewpoint in (Maynard Smith and Szathm´ary,
starting point for an attempt to understand meaning,
1995)). As Wittgenstein (1968) taught us, the mean-
communication, and language. The highly refined
ing of any signalling behaviour, such as in language,
formal tools mathematics and logic – including truth
arises in how it is used by the agent to manipulate
values, predicate logic, context-free grammars, de-
its environment (including other agents) in its interac-
notational semantics, etc. – have allowed scientists
tions with other agents. This can be related to the util-
achieve precision and thus escape from ambiguities
ity to an agent (in a statistical sense) of information
and dependence on context and specific agents. But
in a signalling channel (see Nehaniv (1999); Nehaniv
specific agents and context are inherent to the emer-
et al. (1999, 2002)). According the insights of Peirce
gence of language, while these tools are based on
(1839-1914) [republished in (Peirce, 1995)], the rela-
abstractions and refinements from human language.
tionship between signs and significations is mediated
Any explanation of the emergence of language that
by an interpretant, and the mapping between signs
uses them as primitives to derive the phenomena that
and what they signify is a process that depends on
they are based on thus puts the proverbial cart before
the particular agents involved and on their situated
the horse (Nehaniv, 2000; Milikan, 2004).
contexts. The ideas just presented follow the discus-
This is not to say that these tools and formalisms
sion of Nehaniv (1999, 2000). The Wittgensteinian-
should never be used. In computational modelling
Peircian viewpoint outlined by Parisi et al. (2002) is
this is clearly would not be possible, simply due
similar.
to the use of computers. No simulation or robotic
In particular, these realizations lead a tremendous
study in the emergence and evolution of linguistic
amount freedom in the emergence of language-like
communication has been able to proceed successfully
phenomena that has often been ignored and over-
without simplifying some (or sometimes all) of the
simplified by naively, often unconsciously, applying
above complexity away. If agents are endowed with
constraints on simulation models. This freedom and
some of these language-like capacities, it is impor-
the related lack of constraints is illustrated by several
tant to keep track of which ones. If new phenom-
corollaries. Understanding the emergence of meaning
ena then emerge, one has an argument that the built-
and language requires the generative synthesis of the
in capacities provide scaffolding for the new phe-
phenomena in question beginning with the following
nomena. For instance, the work of Kirby (1999)
facts:
shows that, in populations of agents with the capacity
to use and derive context-free grammars, processes
1. Meaning is always agent-specific.
of self-organization resulting from attempts to learn
grammar based on induction from the evidence of
2. There is no privileged set of pre-existing space
grammar-generated utterances of other agents lead
of possible meaning, containing ideal concepts.
over generations to increasingly compositional gram-
mars. His work does not how it is that context-free
3. There is no unique and no pre-existing syntactic
grammars nor the capacity for compositionality could
structure on possible meanings.
first emerge (since these are given at the start).

---

3 Symbol Grounding & Shared
known arbitrariness of the sign in regard to its ref-
Vocabularies
erence.
Milikan (2004) has a more general notion of refer-
Different aspects of symbol grounding (Harnad,
ence that relates to utility of information in internal
1990) and the self-organization and maintenance of
states or signalling channels. A more general notion
shared vocabularies are increasingly well-studied
of gesture regards gesture as the signalling of such
and coming to be understood, especially for vocabu-
useful information. This is similar to the viewpoints
laries to identify or name objects (selecting one target
on the meaning of signals in (Nehaniv, 1999, 2000;
of reference from an environment) or label situations
Wittgenstein, 1968).
(MacLennan, 1992; Steels, 1995, 1998; Billard and
The issues discussed in this section evidently relate
Dautenhahn, 1999; Baillie and Nehaniv, 2001; Parisi
closely to the grounding of symbols and the emer-
et al., 2002). Less work has been done on the ground-
gence of shared systems of communication. Despite
ing of shared vocabularies with more complexity, e.g.
progress in these areas, constructive studies linking
in which various parts of speech exist (labelling for
deixis and gesture to these problem areas remain to be
example actions or actions on objects, or with compo-
carried out constructively in robotic and simulation
sitional syntax), although the work of Cangelosi and
models (but see Baillie and Nehaniv (2001); Baillie
collaborators has moved in this direction (e.g. Parisi
et al. (2004) for some first work in this direction).
et al. (2002)).
5 Predication
4 From Deixis, Gesture, and Ma-
For detailed analysis of predication and its complex
nipulation to Reference
structure in human language from the viewpoint of
linguistics, see (Napoli, 1989). In human language,
The items, deixis and gesture, in challenge area (1)
a rudimentary function of noun phrases is to pick out
are clearly related and emergence of reference. Ref-
objects of reference from the environment (possibly
erence is often suggested to be grounded in deixis and
even absent ones). Adjectives constrain the selection
gesture but just how this occurs needs elucidation.
by imposing conditions on which object might be re-
Pointing, deictic gaze, joint attention, and gesture
ferred to.
play important roles in the development of intersub-
One formal view of reference (implicit e.g. in
jectivity and language in humans (cf. Kita (2003)).
(Steels, this volume) and classical box-world natu-
Pointing, since it can be directed at many things
ral language processing systems) is that instances of
and since it directs others’ attention at them, could
lexical items such as a noun (“ball”) or adjective
have provided for a kind “ur-pronominalization” in
(“red”) are understood as predicating properties of
the emergence of linguistic communication. That is,
object variables. Selection of referents is determined
pointing provides for a variable or variables that can
by solving constraints on such predicates over a space
be bound to object and persons in the environment,
of objects in the environment. For example, ball(X)
giving at least of degree shared reference via shared
and red(Y ), restricts the reference to a red ball if
attention.
X must equal Y , as it must in the phrase “the red
Rizzolatti and Arbib (1998) present a hypothe-
ball”. Similarly verbs provide another class of pred-
sis on the emergence of language based on mirror-
icates which might take multiple semantic role argu-
neurons in primates and humans. These neurons in
ments expressed in a given syntactic subcategoriza-
the premotor cortex fire both when carrying out and
tion frame that resolves variable references (Steels,
when seeing an action performed. It is argued that
this volume).
this provides a substrate on which shared meaning
As mentioned above, predicate logic and first-
can arise, as similar affordant gestures (e.g. manip-
order logic are abstractions from the predicate
ulations such as grasping a fruit) are immediately un-
structure of natural language. With the approach
derstood by a conspecific interaction partner. Gestu-
just described, predication itself is a primitive and
ral language is then hypothesized to have developed
therefore does not emerge. However, a transition
and eventually to have given way to vocal language.
from reference to predication is suggested by an
Hurford (2004) acknowledges a possible role for mir-
association that tends to identify referential variables
ror neurons in understanding the possible emergence
in one-place referential predicates (like red(X)), or
of language, but surveys many gaps that remain in
by grammatical rules that force the identification of
such an explanation, such as explaining the well-
variables in the referential predicates.

---

values of predicates on objects was only a later inven-
Scenario for the Emergence of Predication.
tion and abstraction of humans.
Early on proto-words or gestural signs could have
their referents associated in a general way, non-
specific way merely by co-occurring close together
6 Negation: A Small Research
in time. We elaborate a suggestion on the earliest
Programme
source of predication: it may be a highly derived form
of topic-comment structure, which is itself founded
on association (Nehaniv, 2000). For instance, deic-
It seems little has been done in emergence of nega-
tic gesture serves to select a target of joint attention
tion in constructive evolution of language models.
(topic), and then another gesture or utterance near
A discussion of negation of speach acts and within
to it in time serves to communicate content that was
speech acts occurs in (Searle, 1980/1969). A com-
associated to the topic as comment. Eventually rit-
prehensive book on negation is (Horn, 2001).
ualization of such communicative practice produces
gramaticalization of a topic-comment construction.
Early Scenarios for Negation.
Predication then arises via grammaticalizaton of the
(The material in this subsection is modified from text
special case in which not only an association between
by Donna Jo Napoli (Napoli, pers. comm.).) Early
topic and comment occurs, but the comment gives
predicates used by early humans likely indicated ac-
to the topic a labelling category: “This - food”, a
tions such as “come”, “hide”, “be quiet”, “run”, or
property label “This – bad”, or a semantic action-
referenced objects, such as “food”, “water”. Nega-
role “This – eat”. Thus there is a progression in
tion can operate on nouns as well as on verbs, or
the emergence of predication from association and
other parts of speech, and is, of course, a predicate
topic-comment via ritualization to grammaticalizaton
in itself.1 Letting others know there is nothing in the
of predication.
cave, for example, was probably a pretty important
Ritualization is well-known in animal communi-
early message. So one would expect “nothing” or “no
cation systems (Smith, 1977, 1996; Bradbury and
living thing” to be an early negation.
Vehrencamp, 1998) and one instance of it is grammat-
Non-verbal, facial and manual gestures may have
icalization, a well-recognized process in human lan-
played an important role in early negation. When
guage change (e.g. (Bybee et al., 1994)). A clear path
hunting, when trying to be quiet for any reason,
for research into this open area would be to proceed to
people have always used their faces and hands. We
validate this proposed scenario by building computa-
all recognize the hush gesture. We know to raise
tional or robotic realization and showing whether and
our eyebrows to ask yes/no. This sort of thing is
how the transitions
extremely common around the world. In Australia,
many tribes used to have sign languages just for
association
hunting. (They had sign languages for other things,
→ topic-comment → predication
too – like to use with widows – and for the deaf).
could occur (ideally including grounded referencing).
The first negation was likely either facial or gestural
This should shed light on the details of the emergence
– perhaps a head shake or lowered brows (as in
of predication and the mechanisms required for this to
American Sign Language (Neidle et al., 2001)),
occur.
or protruded lips. Also, early negation was likely
If this could be done, more complex predication
simultaneous with whatever was being negated,
and modification could then be addressed. In more
whether spoken words or other gestures. So shake
complex human language, both predicates and mod-
your head and say “buffalo” - or shake you head
ifers occur. Predicates tend to mark more highly
and say “swim/enter water” or shake your head
salient assertions, while modifiers tend to act in the
and gesture “walk (whatever that gesture might be
background to tune reference via constraints (Ne-
for those peoples) – and you’re getting across the
haniv, 1987).
messages “there are no buffaloes” – “don’t go in the
Let us again remark about the at best low rele-
water” – “don’t walk”. (Scenario and examples due
vance of truth values here. In early language as in
to Napoli (pers. comm.). The author is responsible
animal communication system, the emphasis was of
for any misrepresenations of her views.)
course manipulation of and influence in the environ-
ment via signalling to others (cf. Maynard Smith and
1Or a modifier, where modification is plays a role, e.g. a spec-
Szathm´ary (1995); Milikan (2004); Nehaniv et al.
ifying a constraint on reference within consituent syntactic struc-
(2002)), rather than on propositional assertions. Truth
ture, and is generaly less marked than predication.

---

Computational Scenarios for Emergence of Nega-
a negative signal would have presumably to involve
tion.
the persistence in the network of internal state over
the scope of the negated constituent. Synthetic neural
We now give several ideas for constructivist ap-
imaging techniques like those of Cangelosi and Parisi
proaches to negation:
(2004) could be useful here.
1. It seems straight forward to use inhibition in
artificial neural networks to suppress the behaviour in
7 Syntactic Categories
the presence of a negation signal N. Suppression of
all action could yield compliance (by inaction) with
In artificial neural network connectionist models,
commands such as “don’t touch that”. This could be
Parisi et al. (2002) have shown the grounded emer-
realized to many existing models.
gence of rudimentary nouns and verbs: Nouns, as
2.
A research scenario into the use of more
linguistic signals that co-vary with sensory stim-
specific negation could employ connectionist neu-
uli, and verbs, as linguistic signals that co-vary
ral network models of agents using linguistic sig-
with actions (largely independent of sensory stimuli).
nalling such as those of Cangelosi (2001), which
They have suggested that this could be extended to
can have a noun-verb distinction (see below) that
(proto)adjectives, that select a referent within a noun
they exhibit in language games. We propose that
category using some intrinsic property, and to non-
these be extended by the introduction of tasks into
adjectival modifiers, such as location indicators (e.g.
the language games that sometimes involve nega-
left, right, above), that reflect more temporary prop-
tion: When the new signal N co-occurs with
erties of objects which are not instrinsic to the object
a previously learned linguistic signal S the lan-
but depend on the relationship of object to speakers
guage game task requires choosing a different ob-
and the environment. This remains to be done, as
ject/property (“(proto)noun”/“(proto)adjective”) or
does increasing the complexity of syntactic categories
action (“(proto)verb”), respectively, than would be
the approach can generate (e.g. to verbs with a patient
for the signal S. Tasks without the signal N must also
and recipient role, as “give the apple to Mary”.
be carried out by the agents and require the original
Steels (this volume) also considers the emergence
interpretation of S. That is, the agents could carry
of shared semantic and syntactic frames based on
commands such as “pull cup”, “not-pull [e.g. push]
grammaticalization driven by computational needs of
cup”, “pull not-cup” (i.e. pull an object other than the
disambiguation.
cup), , or even “not-pull not-cup” (e.g. pushing a ball
This issue of emergence of syntactic categories,
would be a correct response). The meaning of the
which are restricted in the types of semantic environ-
negation signal N would be grounded in the language
ments where they can occur (as in the work of Parisi
game tasks these agents have to perform. Demon-
et al. (2002)), and in their signal contexts, and in the
strating that evolving populations of neural network
types of arguments they can take (if any), leads to the
agents could learn this task would establish a connec-
next grand challenge, the achievement of full-blown
tionist basis for specific negation of constituents of
compositional syntax in a grounded communication
simple linguistic utterances. Alternatively, one could
system.
do the same kind of study using agents such as in the
work of Steels (2003).
3. We note that in many human sign languages
8 Compositionality
such as American Sign Language (ASL), the scope
of negation can be given over syntactic subunits by
The emergence of lexical items that take arguments
non-manual gestures. In ASL non-manual marking
(such as transitive verbs that take an noun-phrase as
(furrowing of the eyebrows and side-to-side head-
object) is called compositionality. This has syntac-
shake) may spread over the (c-command) domain of
tic and semantic aspects, and accounts for much of
constituent syntactic node, and moreover such spread
the combinatorial richness of human language. There
is obligatory in the absence of manual marker (Nei-
have also been a growing number of studies on the
dle et al., 2001). (This property agrees well with the
emergence of various aspects of syntax (e.g. (Kirby,
likely simultaneity in the early negation described in
1999; Cangelosi, 2001; Steels, this volume)). While
Napoli’s scenario above.) Thus in constructivist stud-
there has also been some pioneering work on syntac-
ies of the emergence of language, it would be very
tic categories (e.g. Cangelosi and Parisi (2004); Parisi
interesting to investigate scope of negation. For ex-
et al. (2002)), and grounded compositionality (Steels,
ample, in neural network agent models, the use of
1998), many aspects of compositionality in linguistic

---

communication remain completely open for construc-
syntax - compositionality and recursive structure.
tive modellers to begin to explain.
Challenges (2), on predication, and (3), on nega-
Segmentation and pauses in modern human
tion, have been the most neglected by the evolution
speech, e.g. arising from the need to breath or the
of language community. We hope this paper stimu-
temporal nature of cognitive processes, combined
lates discussion on these issues and promote research
with local context have been shown information-
especially into those areas.
theoretically to improve the disambiguation of
The problem of predication (2) is argued to be re-
speech, suggesting that sequential process of smaller
lated to associative processes and to topic-comment
sequential units may help provide the basis for syntax
structures, as precursors. Predicates as the exist to-
in language evolution and language processing Lyon
day in human languages are seen as a highly derived
et al. (2003).
special case of related processes.
Cangelosi (2001) showed the emergence of verbs
Computational scenarios for studying the emer-
for actions that take target objects references in neu-
gence of predication and of negation have been pro-
ral network agents that can manipulate simple objects
posed and discussed in order to encourage the inves-
in the environment in an evolutionary simulation, but
tigation into their hemergence.
non-compositional communicative signals could also
Other immediate work to be done to meet these
evolve.
grand challenges includes: (4) emergence of syn-
Assuming a fixed and syntactically structured
tactic categories needs to be shown without assum-
meaning space, and a capacity to use and learn
ing an underlying categorization on some pre-existing
context-free grammars, Kirby (1999, 2001), as men-
space of meanings in grounded language games. (5)
tioned above, has shown that grammars with high de-
compositionality (and recursion) needs to be shown
grees of compositionality arise and are easier to trans-
to emerge in a setting of grounded meaning without
mit over the course of generations of learning in such
the assumption of an underlying grammatical abil-
agents starting from agents using non-compositional
ity, such as the capacity to learn and use context-free
‘holistic’ grammars (i.e. with a different utterance for
grammars.
each meaning). Extending this work to agent-centred
spaces of meaning grounded in interaction and lan-
guage games remains to be achieved.
Acknowledgements
Steels (this volume) argues that the purpose of
compositional grammar is to reduce the number of
Thanks to Donna Jo Napoli for many ideas relevant to
variables in a decoded meaning structure in order
the early negation and for discussing some thoughts
to hope with compuational complexity in interpre-
on early predication.
tation. He constructs agents in simulation studies
that apply this principle and are able to converge
on shared grammars by reinforcing and modifying
References
syntatic and semantic role-structural frames (to pro-
pogate referental constraints) based on communica-
Takaya Arita and Yuhiji Koyama. Evolution of lin-
tive success and failure. The same structures are used
gusitic diversity in a simple communication sys-
for parsing and for production.
tem. Artificial Life, 4(1):109–124, 1998.
Recursive composition structure is possible if the
expansion of argument can non-trivially include the
Jean Baillie and Chrystopher L. Nehaniv. Deixis
same argument type (as with clauses embedded in
and the development of naming in asynchronously
other clauses). When this occurs, in principal the lan-
interacting connectionist agents. In First Inter-
guage becomes unbounded in size.
national Workshop on Epigenetic Robotics, pages
123–129. Lund University Cognitive Studies, vol.
85, 2001.
9 Conclusions
Jean Baillie, Chrystopher L. Nehaniv, Patrick Quick,
Our list of grand challenge areas identified five chal-
Attila Egri-Nagy, and Sandra Warren. Deixis, in-
lenges beyond symbol grounding and the emergence
teraction topology, and the emergence of naming.
of shared systems of communicating meaning: (1)
In First International Workshop on the Emergence
the role of deixis, gesture, and manipulation in the
and Evolution of Linguistic Communication, pages
grounding and emergence of reference, (2) predica-
33–40. Japanese Society for Artificial Intelligence,
tion, (3) negation, (4) syntactic categories, and (5) full
2004.

---

A. Billard and K. Dautenhahn. Experiments in learn-
editors, Proceedings of the AISB’99 Symposium on
ing by imitation - grounding and use of commu-
Imitation in Animals and Artifacts, pages 55–63.
nication in robotic agents. Adaptive Behavior, 7
Society of the Study of Artificial Intelligence and
(3/4), 1999.
the Simulation of Behaviour, 1999.
Jack W. Bradbury and Sandra L. Vehrencamp. Prin-
Simon Kirby. Spontaneous evolution of linguistic
ciples of Animal Communication. Sinauer, 1998.
structure: an iterated learning model of the emer-
gence of regularity and irregularity. IEEE Transac-
Joan Bybee, Revere Perkins, and William Pagliuca.
tions on Evolutionary Computation, 5(2):102–110,
The Evolution of Grammar: Tense, Aspect, and
2001.
Modality in the Languages of the World. University
of Chicago Press, 1994.
Sotaro Kita, editor. Pointing: Where Language, Cul-
ture and Cognition Meet. Lawrence Erlbaum As-
Angelo Cangelosi. Evolution of communication us-
sociates, Inc, 2003.
ing signals, symbols, and words. IEEE Transac-
tions on Evolutionary Computation, 5(2):93–101,
Caroline Lyon, Chrystopher L. Nehaniv, and Bob
2001.
Dickerson. The segmentation of speech and its im-
plications for the emergence of language structure.
Angelo Cangelosi and Domenico Parisi, editors. Sim-
Evolution of Communication, 4(2):161–182, 2003.
ulating the Evolution of Language. Springer Ver-
lag, 2002.
Bruce MacLennan. Synthetic ethology: An approach
to the study of communication. In Artificial Life II.
Angelo Cangelosi and Domenico Parisi. The pro-
Addison Wesley, 1992.
cessing of verbs and nouns in neural networks: In-
sights from synthetic brain imaging. Brain and
John Maynard Smith and E¨ors Szathm´ary. The Major
Language, 89(2):401–408, 2004.
Transitions in Evolution. W.H. Freeman, 1995.
Morten H. Christiansen and Simon Kirby, editors.
Ruth Garrett Milikan. Varieties of Meaning - The
Language and Evolution. Oxford, 2003.
2002 Jean Nicod Lectures. MIT Press, 2004.
Stevan Harnad. The symbol grounding problem.
Donna Jo Napoli. Predication theory: A Case Study
Physica D, 42, 1990.
for Indexing Theory, volume 50. Cambridge Univ.
Press, 1989.
Takashi Hashimoto and Takashi Ikegami. Evolu-
tion of symbolic grammar system. In F. Mor´an,
Donna Jo Napoli. email correspondence, February
A. Moreno, J. J. Merolo, and P. Chac´on, editors,
2005, pers. comm.
Advances in Artificial Life (3rd European Con-
Chrystopher L. Nehaniv. Predication and modifi-
ference on Artificial Life, Granada, Spain, June
cation in Chinese and universal grammar. Hon-
1995), volume Lecture Notes in Artificial Intelli-
ors Thesis in Linguistics, University of Michigan,
gence, 929, 1995.
1987.
Laurence R. Horn. A Natural History of Negation.
Chrystopher L. Nehaniv. Meaning for observers and
CSLI Pubilications (The David Hume Series: Phi-
agents. In IEEE International Symposium on In-
losophy and Cognitive Science Reissues), 2001.
telligent Control/Intelligent Systems and Semiotics
(ISIC/ISAS’99), pages 435–440, 1999.
James Hurford. Language beyond our grasp: What
mirror neurons can, and cannot, do for the evo-
Chrystopher L. Nehaniv. The making of meaning in
lution of language. In Kimbrough Oller and Ul-
societies: Semiotic & information-theoretic back-
rike Griebel, editors, Evolution of Communication
ground to the evolution of communication. In
Systems: A Comparative Approach, pages 297–
Starting from Society, pages 73–84. Society for the
313. MIT Press, Cambridge, MA, 2004. URL
Study of Artificial Intelligence and Adaptive Be-
http://www.ling.ed.ac.uk/∼jim/mirror
haviour, 2000.
kl.s.ps.
Chrystopher L. Nehaniv, Kerstin Dautenhahn, and
Simon Kirby. Learning, bottlenecks, and infinity: a
Martin J. Loomes. Constructive biology and ap-
working model of the evolution of syntactic com-
proaches to temporal grounding in post-reactive
munication. In K. Dautenhahn and C. L. Nehaniv,
robotics.
In Sensor Fusion and Decentralized

---

Control in Robotics Systems II (September 19-
Ludwig Wittgenstein. Philosophical Investigations
20, 1999, Boston, Massachusetts), Proceedings of
(Philosophische Untersuchungen) – German with
SPIE Vol. 3839, pages 156–167, 1999.
English translation by G.E.M. Anscombe. Basil
Blackwell, 3rd edition, 1968.
Chrystopher L. Nehaniv, Daniel Polani, Kerstin Daut-
enhahn, Ren´e te Boekhorst, and Lola Ca namero.
Alison Wray, editor. The Transition to Language. Ox-
Meaningful information, sensor evolution, and the
ford, 2002.
temporal horizon of embodied organisms. In Arti-
ficial Life VIII, pages 345–349. MIT Press, 2002.
Carol Neidle, Judy Kegl, Dawn MacLaughlin, Ben-
jamin Bahan, and Robert G. Lee. The Syntax of
American Sign Language: Functional Categories
and Hierarchical Structure. MIT Press, 2001.
Domenico Parisi, Angelo Cangelosi, and Ilaria Fal-
cetta. Verbs, nouns and simulated language games.
Italian Journal of Linguistics, 14(1):99–114, 2002.
Charles S. Peirce. Collected Papers, Volume 2: Ele-
ments of Logic. Harvard, 1995.
G. Rizzolatti and M. A. Arbib. Language within our
grasp. Trends in Neurosciences, 21(5):188–194,
1998.
John R. Searle. Speech Acts: An Essay in the Philos-
ophy of Language. Cambridge University Press,
1980/1969).
W. John Smith. The Behavior of Communicating: An
Ethological Approach. Harvard, 1977.
W. John Smith. Communication and expectations: A
social process and the operations it depends upon
and influences. In Marc Bekoff and Dale Jamieson,
editors, Readings in Animal Cognition, pages 243–
255. MIT Press, 1996.
Luc Steels. A self-organizing spatial vocabulary. Ar-
tificial Life, 2:315–332, 1995.
Luc Steels.
The origins of syntax in visually
grounded robotic agents. Artificial Intelligence,
103:1–24, 1998.
Luc Steels. Evolving grounded communication for
robots. Trends in Cognitive Science, 7(7):308–312,
2003.
Luc Steels. What triggers the emergence of grammar?
In Angelo Cangelosi and Chrystopher L. Nehaniv,
editors, Seconnd International Symposium on the
Emergence and Evolution of Linguistic Communi-
cation. Society for the Study of Artificial Intelli-
gence and Adaptive Behaviour, this volume.

---