Application by Geoffrey Edwards to participate in the NCGIA Initiative #21
Workshop on Formal Models of Common Sense Geographic Worlds

1) Why participate and what is my contribution (1 page)

I have been working on the links between cognitive science and spatial
information since my attendance at the Buffalo 1988 meeting on Cognitive
and Linguistic Aspecs of Geographical Space. Following this I participated
in the NATO ASI meeting on that topic in Spain in 1990, and at the COSIT
meetings in 1993 and 1995. These meetings led to new collaborations with
G=E9rard Ligozat from LIMSI, in Paris, in 1995, and Bernard Moulin from Lava=
l
University, in late 1994. Through these collaborations, which consist of
using the Vorono=EF data structure to support the building of mental models
and pivot representations on a computer to aid natural language processing,
I have been taking part in AI meetings dealing with Spatial Reasoning and,
most recently, a mixed AI/Cognitive Science meeting on Spatial
Representations.

Since the 1995 COSIT meeting I have been working on the beginnings of a
formal model for a more cognitively-oriented representation of spatial
concepts. The new model appears to be rich in properties and useful within
a wide variety of contexts dealing with spatial information. I expect to
present the current state of this work, including early attempts to build
an object-oriented spatio-temporal data structure based on these ideas in
the Java development environment.


2) Position statement (3 pages)

        Spatial information theory is composed of two separate and parallel
streams of inquiry - formal geometric and set-theoretic approaches on the
one hand, and cognitive science-based approaches on the other. Little
integration between the two approaches has been achieved, despite many
efforts since the ground-breaking work in the late 80's which showed the
links between the two approaches.

        In the geometric approaches, we are still "hung up" on
point-line-polygon primitives and all the operations which follow from
these structures, as well as simple extensions to handle time (events and
intervals) and 3D concepts (Raper, SDH, 1996). These primitives are
convenient because they match closely well established mathematical
concepts.

        In the cognitive science approaches, it has been noted that space
and time are coupled and cannot easily be separated, hierarchies are
characterised by ordering effects, and spatial representations are always
partial (and multiple). However, integrating cognitively "neutral"
primitives such as points, lines and polygons, within such cognitive
approaches is awkward in the absence of more appropriate primitives.

        Some integrated efforts are emerging from the work of AI
researchers coming into the domain with new ideas about indeterminacy and
qualitative data and inferences (logics) which can be made with such data.
These efforts appear to be midway between classical geometry & topology on
the one hand, and cognitive studies of human behaviour on the other.
However, these efforts do not address the problem of developing appropriate
primitives.

        In Geocognostics (Edwards, AAAI Spring Symposium, 1996), I propose
a new set of spatio-temporal primitives which permit us to model
spatio-temporal data so as to have properties similar to those observed for
cognitive data. Geocognostics proposes two new primitives, one called a
view and the other called a trajectory. A trajectory is an ordered sequence
of events which represent decisions about movement and orientation by a
persistent observer within an arbitrary space. These decisions are made
based on information available with views, which are collections of
observations acquired at the trajectory events by the persistent observer.
Observations may be external, based on collecting locally neighbouring
information, or they may be internal, based on memory of more global
conditions or other useful information.

        More complex views may be built out of primitive views by combining
information from the views and their trajectories together. Furthermore, it
is possible to construct "projected trajectories", which consist of
imagined decision paths through an existing view treated as a figural
space.

        Views and trajectories are being structured into a new set of data
structure primitives for exploration of these concepts. It is possible to
construct both traditional vector and raster data sets and points, lines
and polygons out of these primitives. However, to do so requires stripping
out the temporal ordering of the events, and hence losing the behaviour
more appropriate to cognitive data. Other, more interesting, operations
consist of building complex views out of the simpler view-trajectory
structures. These views incorporate a temporal order and constitute partial
spatial representations of the world. By positing the existence of
persistent objects within views, it becomes possible to build other kinds
of representations which can be used to simulate cognitive representations
of the world.

        Views in geocognostics are not limited to visual fields, but could
represent any sensory information or any internal brain state (emotions,
thoughts, memories, etc.). Furthermore, different assumptions about the
world will lead to different kinds of complex representation. Hence, for
example, assuming persistent objects will permit representations of the
"objective world", whereas assuming a link between decisions and emotional
states might lead to representations more appropriate to a dream world.

        These data structures, while interesting, would be of purely
academic interest if it weren't for a number of properties which makes them
more widely useful. First of all, these data structures include a built-in
temporal ordering and represent a fully temporal spatial data structure.
Secondly, they incorporate decisions directly into the primitives, and
hence represent a data structure which supports decisional tools. Thirdly,
the trajectory/view perspective corresponds to the experience of users with
respect to user interfaces of software. This is most clearly seen for the
World Wide Web, which can be easily modeled via a series of decisions which
affect the views one sees. Fourth, the trajectory/view perspective provides
new insights into problems related to spatial statistics and spatial error.
=46rom the point of view of geocognostics, for example, interpolation error
consists of a possible range of decisions about trajectory placement.
=46ifth, the trajectory/view perspective corresponds closely to that
incorporated into natural language and hence the process of "translating"
spatial concepts or scenarios into natural language, and vice versa, is
facilitated. Sixth, the trajectory/view perspective may be used within
"non-spatial" spaces, such as proposed by Kuhn (SDH 1996).

        The Vorono=EF data structure is a fully dynamic trajectory-like
spatial data structure which is particularly suited to the concepts
incorporated into geocognostics. In related work, we have already shown how
the Vorono=EF data structure can be used as a simulated mental model for
natural language processing (Edwards & Moulin, IJCAI, 1995), and as a pivot
structure between a spatial database, natural language route descriptions
and graphical representations (Edwards et al, SDH, 1996).

        Current work within this framework consists of the following activit=
ies:

        (a) Building a low-level spatio-temporal data structure which
implements the ideas expressed in Geocognostics;
        (b) Applying the concepts of Geocognostics to the analysis of
spatial error;
        (c) Making the link between Geocognostics and the work on mental
models and natural language processing using the Vorono=EF model and its
corresponding data structure;
        (d) Building more complex spatial representations by combining
trajectories and views; and
        (e) Studying the relationship between the Web and the
trajectory/view concept in order to build mental models of the Web.

3) CV (5 publications)

        - Edwards, G., 1996, Geocognostics - A New Paradigm for Spatial
Information, AAAI Spring Symposium '96, Stanford, California
        - Edwards, G., and B. Moulin, 1995, Towards A Simulation of Mental
Models using the Vorono=EF Data Structure, IJCAI Workshop on Spatial
Expressions, Montr=E9al, Qu=E9bec.
        - Edwards, G., G. Ligozat, A. Gryl, L. Fraczak, B. Moulin and C.M.
Gold, 1996, A Vorono=EF-based pivot representation of spatial concepts and
its application to route descriptions expressed in natural language, 7th
International Symposium on Spatial Data Handling, Delft, Netherlands, pp.
7B.1-7B.15
        - Edwards, G., 1993, The Vorono=EF model and cultural space:
applications to the social sciences and humanities, Proceedings of COSIT,
Lecture Notes in Computer Science 716, Elba Island.
        - Edwards, G., 1992, Spatial knowledge for image understanding,
NATO ASI on Cognitive and Linguistic Aspects of Geographic Space, Kluwer.