18th ANNUAL WORKSHOP Information

18 May (14.00 hrs) - 20 May (17.00 hrs) 2001

Maastricht, The Netherlands

A Multidisciplinary approach of cognition: Modelling, experiments, and applied research in the fields of: - psychology (cognitive, developmental), perception, - artificial intelligence (general aspects), - associative memory and neural networks, neuroscience, - linguistics (also computational), disorders of language, - education and instruction, - cognitive ergonomics, - philosophy, history of concepts.

Fee NLG 70 (Dutch guilders, about 30 Euro for members of the ESSCS, NLG 90 (40 Euro) for others, students NLG 35 (16 Euro).

A block booking of rooms has been made till 15 April 2001, later bookings may be possible, please contact us for more information.

speaking time: 30 minutes, plus 5-10 minutes for discussion

Friday 18 May 2001, 14.00-17.30 (Cognitive Ergonomics)

15.15: Coffee or tea

19.00, Dinner with all participants in Hotel La Colombe

Saturday 19 May, 9.00-12.00

10.15: Coffee or tea

Afternoon: free, sightseeing in Maastricht or environments

Sunday 20 May 2001, 9.00-12.00

Coffee or tea

18th ANNUAL WORKSHOP Submitted Abstracts
(in alfabetical order by author)

Robin Allott


The sound-structure of a word can represent the meaning of the
word in a number of ways:
1. Most straightforwardly the word can sound like its meaning.
Examples: hiss, wail, sigh, cuckoo, tick-tock, ding-dong.
2. A word can generate indirectly a sound representing what the
word means. The names of many animals are in this category with
other words referring to things which produce sounds. Examples:
cat, dog, wolf, wasp, bell, whistle, klaxon, thunder, wind.
3. A word can directly reproduce the action to which it refers.
Examples: spit, suck, chew, yawn, sneer.
4. A word can generate a deictic gesture, that is, a gesture which
involves pointing to what the word refers to. Words for many body
parts and some pronouns are in this category. Examples: head, ear,
eye, you, he.
5. A word can generate an action which represents what the word
refers to. This is similar to deictic words. Many function words fall
in this category as well as many simple action verbs. Examples: this,
that, at, forward, back, hit, throw, take, push, point, pluck, pick, sew,
heavy, light.
6. A word may generate a gesture which outlines what the word
refers to. Examples: arch, edge, heap.
7. A word may generate an action picturing the use of an object
referred to. Example: needle,
8. A word may generate a picturing action plus an associated sound.
Examples: whip, air, fire.
9. A word may generate a gesture which amounts to the 'showing'
of what the word refers to. This is particularly the case for body
parts. Examples: hand, arm, elbow, wrist.
10. A word may generate a deictic gesture plus an action gesture.
Examples: hat, flower.
11. A word may generate a complex containing visual, sound, deictic
or other elements. Examples: snake, rain.
12. A word may generate an internal feeling or action. Words for
emotions seem to come in this category. Examples: bitter, sweet,
sour, sad, angry, cheerful.
13. A word may generate an action or an internal feeling for an
abstract, non-concrete meaning. Examples: remember, understand,
know. The associated actions may metaphorically represent the
meaning e.g. elevate, separate, grasp, spontaneous.
14. A word may generate a movement or position specifically
referring to an aspect of time. Examples: now, then
Animations can be used to demonstrate the relation between word-
structures and gestures. There is also an illustrative list of
words where the sound/meaning relation falls in one or other of the
above categories. The neural basis of word sound/meaning
relationships can be explained in terms of motor control, motor
equivalence and the recently discovered mirror neurons.

Peter J. Braspenning, Gabriel Hopmans and Peter-Paul Kruijsen
Maastricht Univ., Section Communicatons Research & Semiotics


Principal features of today's technological revolution are:
1) Increasing hypertextuality
- for about one and a half century, since the development of the
telegraph, the world has seen the proliferation and accelerating
growth of a global infrastructure of layered, interconnected and
integrated networks,
- over the last half century, since the creation of the first
digital computers, the digitization of as much material as
possible has promoted both the homogenization and standardiza
tion of inputs, which in turn has permitted infinite customi
zation of outputs - over the last half decade, since the
invention of virtual reality, the virtualization and convergence
of the properties of hardware are turning traditional hardware
contents not merely into software but more radically into mindware.
2) Increasing interactivity
- the realisation of more intuitive and user-friendly interactions
between the user and the computer is causing a shift from internal
to external processing of information (where virtual environments
become even more an extension of private mind memories)
- ever more intelligent software is allowing the migration of
psychological processes such as memory and cognition from the
inside of individual minds to the outside world of connected-
knowledge media,
- rapidly evolving multimedia, hypermedia, and virtual reality
make direct-connect mind- machine interactions much more easy
and effective and is in fact preluding our common constant
sensory interaction with dynamic, self-adjusting, intelligence-
based media.
3) Increasing connectedness
- the increase in human interactions - personal, social, and
institutional - through integrated networks is concentrating
and multiplying human mental energy,
- the gradual self-organisation and increasing autonomy of agent
supported databases will provide even more tools for emergent
problem-solving skills,
- hence, the degree of collaboration among individual people's
mind is about to become vastly increased, tutored and focussed
by mediating software and hardware.
The foregoing trends are occurring simultaneously and are inter
dependent indeed. They provide the background for describing
conceptually the new kind of information systems that we will
soon see to appear on the global information stage. In fact,
they have been denoted already as Artificial Social Systems,
because they are mimicking social group collaborating behaviours
and are working in unison with actual social groups of human
The complementarity of trends in innovation, research and
development, manufacturing, marketing and media, organized
labor, and government behaviour appears to take place within
a global process that is largely self-organizing and uses the
principle of connectedness on all system levels involved,
including the connectivity of human minds.
Of all of the possible Artificial Social Systems that will
enter the stage, we will single out one kind of system in
particular, that we have called Artefactual Information Systems
(AIS). We are referring here to information systems that support
human beings in accessing, manipulating and enhancing the
huge amount of cultural memory artefacts that are already
represented or virtually present in digitized forms. While
zooming in on the conceptual architecture of such an AIS,
we will sketch the basic mechanisms that the architecture
should allow for in order to cooperate with human beings in
their interaction modes with cultural artefacts.
Intelligent Information Agents: Agent-Based Information Discovery
and Management on the Internet. Matthias Klusch, 1999.
Connected Intelligence: the arrival of the web society.
Derrick de Kerckhove, 1997.

Abstract coming

G.J. Dalenoort
Dept. of Psychology, University of Groningen, The Netherlands,


Piaget was a functionalist, and he was not the only one. He has
explicitly stated that the findings of the neurosciences could not
have direct relevance for the psychological study of behaviour.
Since the enormous flight of connectionist modelling of perceptive
and cognitive processes, and of the huge increase of direct
measurement of global brain activity by EEG, Petscans, and fMRI,
the opinions have changed.
Unfortunately, the effort in these directions have made that
modelling of the microscopic actual informational processes in
the neural networks making up our brains, has been severely
neglected. Partly this is due to the expectation that connectionist
models were good models for information-processing in the brain,
in spite of the fact that a number of important questions of
cognition and of cognitive development can hardly be expected
to receive answers from the connectionist efforts.
I shall discuss a number of science-theoretical issues of
relating cognition and brain, more specifically how some of the
central concepts of Piaget for the development of cognition can be
related to neural networks.

Dept. of Psychology (E&A), Univ. of Groningen
Grote Kruisstraat 2/1, 9712 TS Groningen, The Netherlands
E-mail: g.j.dalenoort@ppsw.rug.nl
Dieter Gernert
Technical University Munich, Germany


This paper presents five characteristic classes of fundamental
operations, such that at least a significant majority of all
cognitive processes can be described by compositions of
operators taken from these classes. In this sense fundamental
operations are characterized by the terms learning, conclusion,
valuation, revision, tentative (operator classes L, C, V,
R, T). It will be shown how well-known cognitive processes,
like classification, concept formation, pattern recognition,
or problem solving, can be described on this basis. These
operators form an operator algebra of a type already known
in literature; its similarities and dissimilarities when compared
with operator algebras used in quantum theory will be briefly
discussed. A striking feature of general interest lies in the fact
that these operators (except for trivial cases) are non-
commutative. The open question remains which kinds of
cognitive processes can not be represented in the way
proposed here.
Possible applications are related to a novel, very specific
approach to the problem of internal (subjective) time and to a
unified description of processes in which the observer can no
longer be ignored (foundations of quantum theory).
Prof. Dr. Dr. Dieter Gernert
Hardenbergstr. 24, D-80992 Muenchen (Germany)
TU Muenchen
Arcisstr. 21 Tel. : (+49 89) 36078-280
80333 Muenchen
E-mail: t4141ax@mail.lrz-muenchen.de

Ab de Haan
Cognitive Science, NICI, Nijmegen, The Netherlands


Cognitive ergonomics can be seen as a science-based design of
environments. These technological and other types of
environements, such as computerinterfaces, shopping malls
and communication methods and protocols are meant to
influence human cognition and can thus be seen as a variant
of "cognitive engineering". In that sense cognitive ergonomics
distinguishes itself from AI, that is primarily directed at
influencing machine cognition.
To influence something you need to know it. Therefore
science plays a fundamental role in cognitive ergonomics. A
number of examples shall be used to exemplify this position.
Central in these examples shall be to notion of a task (set) as
the expression of a neural cogfigurations and the role this
notion plays in switching between tasks.
Dr. A de Haan
Cognitive Science, NICI,
P.O. Box 9104, 6500 Nijmegen, The Netherlands
tel: +31 243611938 home: 0575560399
fax: +31 243616066 0650652726
E-mail: dehaan@nici.kun.nl

Geert de Haan
McLuhan Institute, Maastricht University, The Netherlands


An increasing wealth of trusted and well-described information
from museums and libraries is becoming accessible through
the web, albeit not without hindrances. Information is hidden
behind collection-specific 'unfriendly' user interfaces and
collections often utilise domain-specific or incompatible
description methods or metadata sets to describe items.
I-MASS (Information management and Interoperability of
content for distributed systems of high-volume data
repositories through Multi agent systems; see:
is a European IST project aiming to overcome these problems
by designing and building a Virtual Reference Room
(Veltman, 1996)
First, since we cannot expect museums and libraries to adopt
some new and universal description methods, I-MASS will use
multi-agent systems to hide or otherwise to manage the
differences between metadata standards. Secondly, the ability
to overcome the differences between metadata standards is
used to create a single user interface for users to access the
different museum and library collections in a uniform way.
Thirdly, the multi-agent system is further used to create and
employ dynamic user models to provide information that is
adapted to the characteristics and preferences of individual users.
In the presentation we will describe the I-Mass project in a
global way, including the underlying philosophy and the
problem of interoperability of content. Subsequently, we will
discuss the cognitive ergonomic aspects of designing and
evaluating the system with respect to information presentation
and the use of dynamic user models, paying particular
attention to the issue of user initiated versus automatic
adaptation of the system.

Veltman, K.H. (1996). Proceedings of the EVA '96 Conference,
Future Strategies and Visions. Universal Media Searching -
SUMS and SUMMA, National Gallery, London.
Geert de Haan
User Interfacer, Maastricht McLuhan Institute
Grote Gracht 82
mail: P.O. Box 616 6200 MD Maastricht
The Netherlands
tel: +31-43-3882714 fax: +31-43-3252930
E-mail: G.deHaan@mmi.unimaas.nl

Christian R. Huyck and Richard Bowles
Middlesex University, London, UK


Neurons generally fire when they receive activation from other
neurons. However, there is evidence to show that they can
also fire spontaneously, i.e. without any external activation
[Abeles 93] Spontaneous neural firing may have some
physiological reason, but neurons are computational devices.
What are the computational ramifications of spontaneous
neural firing?
Computational research suggests two useful functions:
attenuation of neural waves, and recruitment of neurons to
Cell Assemblies. Spontaneous firing should lead to more
information content in neural waves. This is due to the
dynamics of firing and of learning being influenced by
spontaneous firing. Both of these processes make it less
likely that all of the neurons in a neural wave will not fire.
Fewer neurons firing increases information content.
Spontaneous firing also improves the dynamics of neural
recruitment in CAs. Our experiments show that spontaneous
firing can lead to more robust CAs.
This work is by no means conclusive. However, it fits in with
existing neurophysiological evidence, and shows how spontaneous
firing might lead to better performance in real neural systems.

E-mail: C.Huyck@mdx.ac.uk

Inge Lemmens
Section Communications Research & Semiotics,
Faculty of General Sciences, University Maastricht


Studies have shown that the effectiveness of the user-interface
has become increasingly important as computers have become
useful tools for persons not trained in computer science. In
face, the interface is often the most important factor in the
success or failure of any computer system. Traditionally,
interfaces are strongly process-based: the function of the
interface is to guide the user through a sequence of actions by
which some goal is reached. An alternative way to structure
interfaces arises from sociological investigations. The claim
that is made is that structures, regularities, and patterns
of action and behaviour, emerge out of the everyday action
of individuals working according their own common-sense
understanding of how the social world works. When humans use
computer systems to perform some task, they have certain
information about the state of the system. This information
can only be useful when it helps the user while performing
a certain task. It is thus the relationship between the state
of the system and the state of the work which the user is
trying to accomplish that is important. When such a
relationship can be established, then information about the
state of the system can be used to organise ongoing working
activity. We will investigate how the techniques of
computational reflection can aid in establishing the relationship
between the ongoing activity and the state of the system.
References: Dourish, P., Accounting for System Behaviour:
Representation, Reflection, and Resourceful Action. Technical
Report EPC-1995-01, Rank Xerox, 1995.

Section Communications Research & Semiotics,
Faculty of General Sciences, University Maastricht
E-mail: I.Lemmens@CRS.UNIMAAS.NL

Markus Peschl
University of Vienna, Austria


In this paper simulation is presented as a radical alternative
approach introducing a new quality to the process of theory
development. One of the main methodological characteristics of
cognitive science (compared to other disciplines studying
cognition) is the extensive use of simulation models. In the first
part of this paper the foundations as well as implications from
the perspective of epistemology as well as of philosophy of
science will be developed. It will be shown how the method of
simulation becomes an integral part for the process of theory
construction in cognitive science. The second part of this
paper is concerned with the question of identifying the
adequate level of abstraction for computational models of
cognition. The strength of cognitive models with high
explanatory power lies in providing mechanisms on a
conceptual level; i.e., on a level of abstraction which respects
the structure of underlying (physical) mechanisms, but reduces
the empirical details of these mechanisms in such a way that
the resulting model sufficiently approaches the behavioral

Markus Peschl
Dept. for Philosophy of Science and Social Studies of Science
University of Vienna
Sensengasse 8/10, A-1090 Wien, Austria / Europe
Tel: +43 1 4277-47624, Fax: +43 1 4277-9476
* e-mail: Franz-Markus.Peschl@univie.ac.at
* personal homepage/teaching:
* WWW Dept. for Philosophy of Science and Social Studies of Science:
* Austrian Society for Cognitive Science:

Thomas Riga, Alberto Greco,
University of Genoa, Italy
Angelo Cangelosi,
University of Plymouth, UK


A neural network model of categorical perception is
presented, which addresses the symbol-grounding problem, i.e.
the problem of making symbols inherently significant to a
system and not mediated by an external user/interpreter.
Analogue sensorimotor projections are first linked to arbitrary
symbolic names, thereby grounding them. Subsequently,
descriptions of higher-level categories are built by combining
these grounded names, from which the new categories inherit
their grounded significance.
In a recent connectionist model, neural networks were
tested for their ability to transfer grounding from sensori-
motor categories to high order categories learnt via symbol
combination (Cangelosi, Greco & Harnad, 2000). This model
successfully performed the transfer of grounding for small
category sets.
In the present work, we have extended this model to deal
with larger category sets, and to look at different aspects of
the transfer of grounding. Each network has three layers of
units. Connections are modularly organised, by dividing hidden
units into two groups, one for processing shapes, and one for
features. The networks were first trained to classify retinal
images using the backpropagation algorithm. Training stimuli
consisted of four animal shapes (e.g. horses) and four texture
features (e.g. stripes). Subsequently, networks were trained to
name these stimuli (entry-level stage) through direct grounding
in retinal input. During the third training phase (higher-level
stage) the networks acquired new category names (e.g. "zebra")
defined on the sole basis of symbol combination.
During the test phase, novel retinal images (e.g. drawings of
zebras obtained by combining horse shapes and stripe features)
were used in input. The networks were able to correctly
categorise and name images using entry- and higher-level
symbols. This result clearly shows that grounding is
"transferred" from directly grounded names to higher-order
ones (grounding transfer). Moreover, the networks were able
to give the correct sensorimotor response when they received
the name of a higher-level category in input (inverse grounding
These results support the approach to symbol grounding
based on fully connectionist models. The same network
processes both the sensorimotor grounding and the generation
of new categories through symbolic learning. The modular
organisation of hidden units suggests that it is important that
sensorimotor grounding be separated for different classification
features. We are currently working on various extensions of
this model, which deal with more complex hierarchies of
categories, different stimuli and features sets, and various
connectionist architectures.
- Cangelosi A., Greco A. and Harnad S. (2000). From robotic
toil to symbolic theft: Grounding transfer from entry-level to
higher-level categories. Connection Science, 12 (2), 143-162.

Alberto Greco (greco@disa.unige.it),
DISA, Lab. of Psychology and Cognitive Science,
Via Balbi 4, 16126 Genova, Italy.
Angelo Cangelosi (acangelosi@plymouth.ac.uk),
Centre for Neural and Adaptive Systems,
University of Plymouth, UK.

Christian Stary
University of Linz, Austria


The presentation introduces the concept of adaptation
of software systems and user interfaces in terms of adaptability
and adaptivity. An extensive literature review has shown
that many concepts for manual and automated adaptation
has been proposed and implemented. However, in the
engineering community there is still a lack of common
understanding of adaptation. Thus, the first part of the
talk will try to develop some structure to embed existing
knowledge on adaptation. In the second part some approaches
to model software and users, in order to construct and
run adaptive systems are analysed. It turns out that symbolic
approaches tends to dominate modeling and representations,
however, allowing adaptation of interfaces only to a very
limited extent. Some ideas to overcome current limitations
are discussed.


P.H. de Vries
Dept. of Psychology, University of Groningen, The Netherlands


For the processing of information in the brain a number of
specialized subsystems can be distinguished. For example, the
location of an object and its identity cannot be represented as
a fixed conjunction since it would require an implausibly large
number of conjunctions. The separate representation of
location and identity then requires a dynamic from of binding
so that objects and their locations can be directly coupled. It
will be argued that such a dynamic binding does not require a
general temporal encoding of each memory trace. Instead it
suffices that the brain generates a limited number temporal
firing patterns within a certain context. The hypotheses about
binding and their role in other cognitive processes such as
serial recall and the perception of letters and words will be
discussed on the basis of a conceptual network.
Each node in this network represents a memory trace which
is characterized by a critical threshold: when the activation in
a node exceeds this threshold it will autonomously grow to its
maximal value. Nodes active above the critical threshold are
assumed to be in short-term memory, whereas subthreshold
activity corresponds to priming or automatic processing.
Computer simulations of the network, illustrating the role of
binding and of the critical threshold will be described for the
distinction of the name and location of letters and of the
serial recall of a list of words. A newly conducted experiment
will be described.

P.H. de Vries
Dept. of Psychology, University of Groningen
Grote Kruisstraat 2/1, 9712 TS Groningen, The Netherlands
email: p.h.de.vries@ppsw.rug.nl
tel. +31 50 3636454 (6472), fax. +31 50 3636304

Reinder Wassenaar
Psychology, Groningen


Most theories and models of concepts claim that concepts are
represented as a set of characteristic features, but real
concepts, however, are much more dynamic and complex. An
important question is the representation of these features.
What is their relation to simple perceptual components and
how are the features learned? Studies suggest that this set
can be flexibly learned during a categorization task. It is
important to realize that most of the research in categorization
has been strongly influenced by simple stimuli-sets, e.g. the
so-called 5-4 set, first created by Medin and used by most of
the protagonists of the exemplar models of categorization.
Models usually simulate a particular aspect of the conceptual
system, and even about these models there is no strong consensus.
We should use data from other subfields of cognitive science
to get a better grasp of the complex conceptual system and try
to incorporate aspects like binding, combination of concepts,
and concept-interrelatedness into existing models.

ESSCS european society for the study of cognitive systems