1.  Title of submission:

        An introduction to the COGENT Cognitive Modelling Environment 

 2. Name of contact person:

        Dr. Richard Cooper

 3.  Affiliation:

        Department of Psychology
        Birkbeck College

 4. Address including post code/zip code and country:

        Dr. Richard Cooper
        Department of Psychology
        Birkbeck College
        Malet Street
        London, WC1E 7HX
        England

 5. Telephone:  0171 631 6211
    Fax:        0171 631 6312
    Email:      R.Cooper@psychology.bbk.ac.uk

 6. Names and affiliations of additional authors:

        Dr. Peter Yule
        Department of Psychology
        Birkbeck College

 7. Duration of your proposed tutorial:

        1/2 day

 8. Resource needs:

        1 machine for every two attendees +
        1 device for projecting PC output onto a screen +
        1 projector screen

        The machines should be either all Windows PCs or
        all X workstations running SOLARIS or LINUX (our
        slight preference is for PCs). All software will
        be supplied.

        Each participant will receive a copy of the tutorial notes
            (approximately 24 pages)

 9. Abstract (not to exceed 200 words), for inclusion in conference
    announcements and the program:

COGENT is a graphical environment for cognitive modelling. It provides
facilities for developing and running computational models (and computational
experiments) based on a box/arrow notation. The system allows users to 1)
sketch the functional components of their model; 2) flesh out such a sketch by
specifying the computational properties of boxes or writing production-like
rules; and 3) explore the behaviour of the resulting model. COGENT is
currently being used for both cognitive modelling teaching and research. The
tutorial will aim to introduce attendees to the environment, so that they will
be able to make an informed decision about whether COGENT is appropriate for
their teaching or research needs, and, if so, also to begin to develop their
own COGENT models. In order to achieve these aims, an interactive approach
will be taken. Attendees will develop an implementation of the Atkinson and
Shiffrin (1968) Modal Model of memory. They will then explore the effects of
altering model parameters (such as the capacity of the short term store, or
the rate of flow of information from the short term store to long term store)
on the serial recall behaviour of the model. Minimal computational expertise
will be assumed.

10. Abstract biography of presenter(s), (not to exceed 200 words total):

Dr Richard Cooper (14/04/65): B.Math. (Hons, Class 1) University of Newcastle,
Australia, 1986; Ph.D. (Cognitive Science) University of Edinburgh, 1991;
Postdoctoral Research Fellow (Cognitive Modelling, UCL, London, 1990-95);
Lecturer (Department of Psychology, Birkbeck College, London, 1995-present).
Since 1990 RC has been working on methodological and practical aspects of
cognitive modelling. This work lead to the development of the COGENT modelling
environment. RC has presented work relating both to COGENT and to specific
cognitive models at the last four cognitive science conferences.

Dr Peter Yule (31/05/63): M.A. (Philosophy & Psychology, Hons, Class 1)
University of Edinburgh, 1990; M.Sc. (Cognitive Science & Natural Language)
University of Edinburgh, 1991; Ph.D. (Cognitive Science) University of
Edinburgh, 1996; Postdoctoral Research Fellow (Cognitive Modelling, Birkbeck
College, London, 1996-present). Since 1996 PY has worked on the development of
the COGENT modelling environment and its application in the domains of
reasoning and decision making. PY has presented material on this and related
work at the last three cognitive science conferences.

The presenters have been close colleagues since 1996. They are both
experienced in the presentation of tutorial material on COGENT, having given
three previous tutorials on the system.

11. Case for support - Not more than 1500 words arguing the case
    for including this tutorial at Cognitive Science 1999.

COGENT is a graphical environment which is designed to simplify the process of
developing and evaluating computational models of high-level cognitive
processes. Typically, such modelling is performed either in the context of an
architecture, such as ACT-R or SOAR, or directly in an Artificial Intelligence
programming language such as Lisp or Prolog. These approaches each have
significant disadvantages for certain classes of modeller. COGENT is designed
to overcome these difficulties.

Successful use of AI languages requires substantial programming expertise.
Such languages also provide no guidance on the appropriate level of
abstraction for psychological modelling and no "infrastructure" for the
experimentation and evaluation of the resulting model. A consequence of the
first of these failings is that AI languages typically lead to models in which
theory and implementation are difficult to tease apart. A consequence of the
second failing is that the modeller must re-invent the development and
evaluation methodology for each model.

The use of a cognitive architecture to guide and constrain modelling goes some
way toward addressing the difficulties of AI languages, but further potential
problems arise. Cognitive architectures require a commitment from the modeller
to a range of architectural features. These may or may not be implicated in
the behaviour of concern, and may or may not be consistent with the
researcher's beliefs. Concerns about such features may prevent many
researchers from adopting an architecture. Additionally, existing
architectures require significant intellectual investment in order to use them
appropriately.

COGENT attempts to address the above difficulties by providing the modeller
with an environment designed specifically for the development and evaluation
of cognitive models. The basic system provides the modeller with a sketch pad
on which the model can be drawn as a box and arrow diagram. This level of
description is clearly inadequate as a complete computational specification,
but it provides the psychologist with a familiar notation, thus simplifying
the modelling processes. In order to fully specify a COGENT model the boxes
must be fleshed out, either by specifying computational properties (such as
capacity limitations or decay rates of buffers) or by adding production-like
rules. The specification of such rules is the closest that the COGENT user
comes to writing actual code. Structured editors ensure that the rules are
well-formed.

COGENT also includes an execution environment which allows a model's behaviour
to be explored and evaluated. The environment embodies the notion of a
subject, as well as notions from standard experimental psychology of trial and
block. These provide further hooks to ease the psychologist into cognitive
modelling. It is thus possible, for example, to define subjects with different
memory spans and compare their performance over several trials of the task
under investigation. Behavioural measures may also be graphed or tabulated,
simplifying the interpretation of model output.

It should be clear from the above that COGENT is a valuable tool to the
psychologist who wishes to develop cognitive models but who lacks the
computational skills normally required. COGENT's role, however, extends beyond
this. The system is currently being used as a teaching tool at a number of UK
and continental European universities (primarily Edinburgh, Hertfordshire,
London and Helsinki). It is also being used by a number of UK-based
psychologists and cognitive scientists as a serious research tool. COGENT
models of reasoning, decision making, memory, and problem solving have
recently been developed. Much of this work has been published at cognitive
science conferences over the last two years, and further work is to be
submitted to the 1999 conference.

COGENT has also generated significant interest in the US and beyond. In
particular, a number of Cognitive Science Society members expressed interest
in the system when it was demonstrated at the last two annual conferences of
the society. A paper describing the system has also recently appeared in
Behavior Research Methods, Instruments & Computers (Cooper & Fox, 1998). Given
this, and the level of maturity which the system has now reached, we believe
that a COGENT tutorial would not only be well suited to a cognitive science
audience, but it would also be well-timed for CogSci99.

The proposed tutorial has several aims. At one level the tutorial is intended
to allow participants to evaluate the COGENT environment, and thereby
determine if the environment is appropriate for their needs. These needs may
be related to teaching and/or research. At another level, the tutorial is
designed to give participants sufficient background to begin developing their
own models within the environment. Participants with interests in this area
will be encouraged to discuss their particular modelling concerns with the
tutorial conveners.

We have conducted COGENT tutorials on three previous occasions: the 2nd COGENT
workshop (London, 1997); the 2nd European Conference on Cognitive Modelling
(Nottingham, 1998); and the 3rd COGENT workshop (London, 1998). The
presentation at the proposed Vancouver tutorial will resemble most closely
that of the most recent COGENT workshop presentation. The tutorial will begin
with a brief introduction to the COGENT system, emphasising the purpose of the
system, its strengths and its weaknesses. This will last approximately 20
minutes. The bulk of the tutorial will focus on the implementation of a
well-known cognitive theory: the Modal Model of memory (Atkinson & Shiffrin,
1968), and the subsequent exploration of the behaviour of that model on the
standard serial recall task. In this phase of the tutorial participants will
work in pairs following prepared tutorial notes. The tutorial will end with a
short debriefing session, highlighting the key features of COGENT relevant to
teaching and research (as exemplified by the Model Model of memory), and
providing participants with a chance to feedback their experience with COGENT
and the tutorial to the tutors. The tutors will be available for further
discussion after the tutorial and at the main conference.

The tutorial will make minimal assumptions about the backgrounds of
participants. It will be assumed that participants have a basic understanding
of cognitive psychology and the purpose or rationale of cognitive modelling,
and that they have some experience with using windows-based software. No other
background will be assumed.

The tutorial notes (hard-copies of which will be provided to all participants)
are available at http://cogent.psyc.bbk.ac.uk/wintutorial/index.html. These
notes, which have been debugged at previous tutorial sessions, include a
detailed walk-through of the development of the Modal Model in COGENT,
followed by a section of exercises which explore various possible
modifications to the model (e.g., How does varying the capacity of the short
term store affect behaviour? How does varying the rate of flow of information
from the short term to the long term store affect behaviour?). Experience has
shown that this use of a simple but well known model, together with a series
of exercises relating to that model, works well in a tutorial situation.
Participants work at their own pace. Those with little modelling experience
are able to produce a complete working model within the tutorial period (and
are hence not discouraged). Those with greater cognitive modelling experience
tend to work quickly through the development of the model, and are then able
to spend a significant proportion of the tutorial time on exploring the
properties of the model. The model is open to a range of modifications, and
these modifications maintain the interest of the more experienced modellers.
In addition, the model produces graphical output (serial position curves)
which may quickly and easily be interpreted. The effects of modifications to
the model may therefore quickly be assessed.

References:

Atkinson, R. C. & Shiffrin, R. M. (1968): Human memory: A proposed system and
    its control processes. In K. W. Spence (ed.), The Psychology of Learning
    and Motivation: Advances in research and theory, Volume 2. pp. 89-195. New 
    York: Academic Press.

Cooper, R. & Fox, J. (1998): COGENT: A visual design environment for
    cognitive modelling. Behavior Research Methods, Instruments & Computers.
    30, 553-564.