|
Items in cart [1] |
Questions? Call 888-624-8373 |
| Copyright © 2009. National Academy of Sciences. All rights reserved. Terms of Use and Privacy Statement |
Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 97
Colloquium
Visualization for constructing and sharing
geo-scientific concepts
Alan M. MacEachren*, Mark Gahegan, and William Pike
GeoVISTA Center, Department of Geography, Pennsylvania State University, 302 Walker, University Park, PA 16802
Representations of scientific knowledge must reflect the dynamic
nature of knowledge construction and the evolving networks of
relations between scientific concepts. In this article, we describe
initial work toward dynamic, visual methods and tools that sup-
port the construction, communication, revision, and application of
scientific knowledge. Specifically, we focus on tools to capture and
explore the concepts that underlie collaborative science activities,
with examples drawn from the domain of human-environment
interaction. These tools help individual researchers describe the
process of knowledge construction while enabling teams of col-
laborators to synthesize common concepts. Our visualization ap-
proach links geographic visualization techniques with concept-
mapping tools and allows the knowledge structures that result to
be shared through a Web portal that helps scientists work collec-
tively to advance their understanding. Our integration of geovi-
sualization and knowledge representation methods emphasizes
the process through which abstract concepts can be contextualized
by the data, methods, people, and perspectives that produced
them. This contextualization is a critical component of a knowl-
edge structure, without which much of the meaning that guides
the sharing of concepts is lost. By using the tools we describe here,
human-environment scientists are given a visual means to build
concepts from data (individually and collectively) and to connect
these concepts to each other at appropriate levels of abstraction.
cientific knowledge is dynamic. Its continuous evolution is
~ marked by branches that diverge and converge and by
conceptual frameworks that expand until they no longer support
new insights, triggering dramatic reorganizations. In the earth
sciences, perhaps the most poignant example is the theory of
plate tectonics, originating in the early twentieth century with
the work of Alfred Wegener, and eventually causing a massive
reconceptualization of geological knowledge. Wilson (1) offers
insight into this restructuring from a conceptual and philosoph-
ical perspective, and Giere (2) offers insight from a cognitive
perspective. Although most changes in science are not as dra-
matic as those stimulated by the theory of plate tectonics, the
concepts used by geologists, environmental scientists, and ge-
ographers to understand the Earth's complex systems and their
interaction with human activities are nevertheless evolving as
understanding evolves and as the needs of society change.
Information/geographic visualization can play a vital role in
stimulating and communicating the evolution of conceptual
structures. The case of plate tectonics provides a compelling
example of the potential. In this case, visual representations
influenced eventual acceptance of the theory (2~. Specifically,
the visual representations that provided evidence of tectonic
activity interacted with geologists' different conceptualizations
of the problem domain to produce both new concepts and new
explanations for existing data (3~.
Here, we focus on dynamic visual representations of concep-
tual frameworks that support (i) the process of knowledge
construction and the application of that knowledge to scientific
work and (ii) the connections between concepts in the mind and
www.pnas.org/cgi/doi/10.1 073/pnas.03077551 01
their instantiation in data. These visual representations can
provide insight into the similarities and differences among
scientific concepts held by a community of researchers. More-
over, visualization can serve as a vehicle through which groups
of researchers share and refine concepts and even negotiate
common conceptualizations. Our approach integrates geovisu-
alization for data exploration and hypothesis generation, col-
laborative tools that facilitate structured discourse among re-
searchers, and electronic notebooks that store records of
individual and group investigation. By detecting and displaying
similarity and structure in the data, methods, perspectives, and
analysis procedures used by scientists, we are able to synthesize
visual depictions of the core concepts involved in a domain at
several levels of abstraction.
To contextualize our own work, and make the problem
tractable, we focus on applicability of visual knowledge capture
and representation methods for use in the science domain of
human-environment interaction. Specifically, emphasis is on
science work associated with the local human impacts of global
environmental change. Knowledge about human-environment
interaction is contextualized or "situated" by factors such as the
places to which scientists direct their research, their aims, and
their underlying theories. The structure of human-environment
knowledge also depends on the choice of relevant datasets and
methods and the scientist's experience applying them. Vulner-
ability to environmental changes, for example, is assessed (and
possibly even conceptualized) quite differently in Massachusetts
and Arizona. However, at certain levels of abstraction, some
agreement among researchers in different locations about what
constitutes vulnerability is essential for joint work. Geographic/
information visualization can help collaborators construct and
communicate knowledge structures that reflect the multidimen-
sional connections among people, perspectives, data, and con-
cepts at different conceptual scales.
The research we report is part of a science infrastructure
project within which we are building a distributed collaboratory
to support the work of four research sites that make up the
Human Environment Regional Observatory (HERO) network.
This developing collaboratory is an ideal "living laboratory" in
which we can explore the construction of scientific knowledge
and the role of visualization in enabling that construction.
HERO collaborators are developing protocols to guide the
collection of geospatial data for environmental monitoring and
applying these protocols and data to problems such as assessing
the vulnerability of local places to global environmental change
This paper results from the Arthur M. Sackier Colioqulum of the National Academy of
Sciences, "Mapping Knowiec~ge Domains," heic] May 9-11, 2003, at the Arnoic' anc] Mabel
Beckman Center of the National Acaciemies of Sciences and Engineering in Irvine, CA.
Abbreviation: HERO, Human Environment Regional Observatory.
*To whom corresponclence shouic' be ac~cdressecl. E-maii: maceachren~?psu.ec~u.
2004 by The National Academy of Sciences of the USA
PNAS 1 April 6, 2004 1 vol. 101 1 suppl. ~ 1 5279-5286
OCR for page 98
and the relationship between global environmental processes
and local-scale land use/land cover change.
In the first section below we review previous research in both
concept representation and the geographic/information visual-
ization methods and tools on which our work builds. Then, we
introduce the HERO problem context, review the concept of
scientific collaboratories, and provide an overview of the HERO
collaboratory. Next, we outline methods we have developed and
implemented to support visually enabled concept building, shar-
ing, and application. In the final section, we discuss future work.
Background
A critical component of science work involves developing.
sharing, and comparing concepts and then applying those con-
cepts to data to generate new knowledge. Within the HERO
project, for example, a core concept is vulnerability of people and
places to environmental change. This concept serves as a focal
point that guides the process of posing research questions,
collecting data, and producing communicable results. Vulnera-
bility is also typical of many concepts in the social and environ-
mental sciences; it is a complex, multifaceted, and context-
dependent concept that has both everyday and scientific
meaning. Despite the lack of apparent consistency in the mean-
ing and application of this concept, the notion of vulnerability is
embraced by many scientists in the environmental-chance re-
search community and is the subject of substantial research
effort. Much of this research is geared toward developing
measures of vulnerability that can be applied to specific places,
but from these diverse local descriptions researchers attempt to
synthesize the likely effects of global environmental change over
large regions, even countries or continents. Moreover. the
concept of vulnerability itself has broadened lately in response
to recent world events; contemporary vulnerability measures
attempt to account for risk from anthropogenic factors and from
naturally occurring phenomena (4~.
Our approach to understanding such concepts and facilitating
their construction and use as a framework for science work draws
on several research domains. Below, we briefly discuss two of
these influences: research on concepts and their representation
and research in information and geographic visualization rele-
vant to visually enabled concept building and sharing.
Concepts and Concept Representation. The first driver for our
research in developing tools to support knowledge construction
is the rapidly evolving domain of concept representation. We
define a "concept" as any abstract information resource that
plays a role in scientific investigation. Scientific concepts do not
necessarily include data or methods (although they may reflect
and be constructed with data and methods). Rather, concepts
may be categories, hypotheses, theories, or other constructions
that help scientists organize their observations about the world.
Thus, a person is not a concept, but the idea of a person is; the
tangible form of an entity is given meaning in the world by the
concepts we attribute to it. As we build tools to help scientists
express and explore the concepts they use to describe the world,
our task is to offer ways to structure and signify knowledge such
that it can be communicated and reused most efficiently. The
simplest form in which a concept might be signified could be a
natural-language term, e.g., "plate tectonics," "grassland," or
"water resources." Using the rules of language, scientists build
more complex expressions of knowledge structures that link
data, methods, theories, and other elements, which might be
shared as journal articles. Our interest is not in such end products
of scientific investigation, but in the process by which scientific
knowledge evolves and in the development of tools to facilitate
knowledge work in science.
The field of knowledge representation is concerned, in part,
with computational aids to communication that reflect semantic
5280 1 www.pnas.org/cgi/doi/10.1073/pnas.0307755101
relationships among concepts. Such representations can enable
computational environments to visualize and reason with con-
cepts by integrating knowledge structures within an individual's
concept space and across multiple users and domains. Natural
language is one medium for representing conceptual informa-
tion, and in later sections we describe a visualization tool that
helps reveal structure in natural discourse. However, it is (at
present) difficult for systems that track the construction of
concepts to reason with knowledge presented linguistically. As a
result, our research relies on computational representations of
knowledge and corresponding visual languages that allow infer-
ences to be drawn more efficiently from complex bodies of
scientific knowledge. These representations, as a complement to
sharing knowledge through natural language, also support shar-
ing and negotiation among scientists about the concepts that
underpin joint work. We propose that the visually enabled
concept representation and sharing methods we are developing
will be particularly useful for asynchronous collaboration.
Typically, knowledge representation systems derive from first-
order logic and its variants; popular examples include Prolog (5),
Loom (6), and more recent developments such as frame logic (7~.
Despite (or perhaps because of) enforced decidability and
consistency that can make knowledge representations effective
for recording conceptual information computationally, most
representational formats suffer from a syntax that is difficult for
humans to create or parse. As a result, we favor notations such
as diagrammatic reasoning tools (8) and conceptual graphs (9)
that readily support visualization and both machine and human
reasoning (it is possible to demonstrate equivalency between
certain conceptual graph structures and predicate calculus or
other logic). The concept visualization tools we describe later in
this article couple a knowledge representation format based in
description logic with these concept graphs; with these tools,
researchers can diagram their thinking and have it stored as a set
of description logic predicates that add to personal and com-
munal knowledge bases.
Knowledge representation languages and the construction of
ontologies that use them to describe features of the world have
garnered substantial attention, not just in the computational
sciences and artificial intelligence (e.g., ref. 10), but in the
environmental and social sciences that are the focus of our
present study (e.g., refs. 11-13~. What is largely missing from this
prior work, however, is consideration of how knowledge is
generated, promulgated, revised, and retired. Knowledge rep-
resentation implementations often focus on recording axioms
about a domain without attempting to situate those axioms in the
context of their creation or use. Environmental and social
scientists grappling with the complexity of human-environment
interaction are situated in a nexus of influences that includes
their experience, their perspectives, and the places they study.
These influences, rather than complicating the pursuit of objec-
tive truths, are fundamental to the nature of science work such
that axiomatic knowledge cannot be cleanly separated from
situated knowledge. Physicists see the world differently from
geographers, not because there are different worlds to see, but
because each community works within a historicity that gives
concepts meaning in an evolving domain. This view of science is
hermeneutic (developing out of ref. 14), embedding findings in
a chain of interpretations, theories, models, methods, and mea-
surements. If we wish to understand where ideas come from and
where they go, we must incorporate references to situatedness in
the representation and communication of scientific knowledge.
Further, tools that support scientific knowledge representation
must admit the situated-work practices of their potential users (15~.
Knowledge representations, even those extended as we pro-
pose to include references to aspects of situation, do not by
themselves achieve collaborative knowledge construction.
Rather, knowledge representations must be embedded in tools
MacEachren et a/.
OCR for page 99
that help scientists communicate while preserving the context of
their communication. To this end, many have described human-
computer interaction as a conversation: with oneself, with one's
collaborators, with one's descendants, with a machine (16, 17~.
We trade on this notion of a conversation as a means of helping
researchers uncover the pedigree of shared ideas as they move
from one scientist to another and from one time to another. This
approach complements recent efforts in visualization of argu-
mentation to support science work, discussed below (see ref. 18~.
Situated knowledge representations within collaborative soft-
ware tools ground abstract ideas in a network of "conversations"
across places, times, people, and perspectives. Occasionally,
these conversations are explicit, and later we present results from
visualizing Delphi method discussions as an example; often,
however, they are not, and our work on electronic notebooks (see
below) helps carry out implied conversations between research-
ers across place and time.
Visualizing (Geo)Concepts. The second driver of our research in
developing tools to support knowledge construction is the
combined domains of information/geographic visualization and
diagrammatic reasoning. The information visualization commu-
nity has developed a wide array of information exploration
methods applicable to categorical data that can support inter-
action with scientific concepts. Many of these methods are
designed to support hierarchical organization of information;
examples include the cone tree (19), tree map (20), and the
hyperbolic browser (21~. Recent extensions include work by
Robertson et al. (22) on the representation and exploration of
multiple intersecting hierarchies and by Chen and Kuljis (23) and
Fluit et al. (24) who focus explicitly on representation of knowl-
edge domains. Other visual concept representation methods
adopt a space-partitioning approach that assumes a single level;
examples of these include extensions to Venn diagrams (25) and
mosaic plots (26, 27~. Still other methods focus on spatialization
(28) of information, in general, text documents. Spatialization
involves calculating the relationships among topics or concepts
as distances in attribute space and "mapping" those relationships
into a 2D or 3D space by using dimension reduction and
cartographic representation methods (29, 30~. Among the con-
cept visualization techniques with a spatial metaphor, several
have been developed and successfully used to construct or depict
relationships among electronic resources te.g., Fabrikant and
Buttenfield (31), Havre et al. (32), and Miller et al. (334~.
When concepts involve geospatial components, as is common
in human-environment interaction, developments in geovisual-
ization, information visualization, and exploratory data analysis
that support dynamically linked views, brushing, and focusing
have considerable potential for adaptation to (geo~concept
representation (34-37~. One example is shown in Fig. 1. The
view on the right is a standard choropleth map. It is dynamically
linked to a graph browser (Left) that uses a minimum spanning
tree approach to connecting places (counties, in this case) on the
basis of their distance in multivariate attribute space (a spatial-
ization of this attribute information). (This component is an
extension of an open-source tool by Alex Shapiro called TOUCH-
GRAPH; see www.touchgraph.com.) In the example, the user
clicked on Clearfield County to find other counties that are
similar in attribute space.
The visualization methods noted above focus on helping users
understand complex interrelationships within multivariate, often
hierarchical, datasets. In general, they have not been directed to
initial development (or acquisition) of concepts from individuals
nor to sharing and comparing concepts among these individuals.
However, research on diagrammatic reasoning environments has
used visual techniques to facilitate development of knowledge by
individuals and groups (8~. That research has deep roots in
domains such as legal argumentation, hypertext, and computer-
MacEachren et al.
Clearfield
Clearfield
Fig. 1. Attribute space graph (Left) and linked map (Right). The attribute
graph browser displays a combination of health (cancer mortality and success
in diagnosis), demographic (census), and behavioral risk factors (smoking and
obesity). Selection of Clearfield in the attribute space highlights counties in
both attribute and geographic space that are similar to Clearfield in terms of
all attributes. Most counties similar to Clearfield in this attribute space are
nearby in geographic space.
mediated communication and has begun to produce robust tools
and a rich body of research about how group thinking and
negotiation can be enabled by visualization methods.
One example of diagrammatic reasoning tools with potential
for application to scientific knowledge construction is provided
by BELVEDERE, a software environment that supports the con-
struction of diagrammatic representations of evidential relations
(38~. BELVEDERE enables remote collaboration and provides
learners with shared workspaces for coordinating and recording
their collaboration in scientific inquiry. It includes a visual
representation language through which participants can build
and share scientific arguments. Concepts that can be encoded
include principle, theory, hypothesis, claim, and report; relation-
ships include supports, explains, conflicts, justifies, and under-
cuts; and representations can be private, shared with all, or
shared with a subgroup. Rinner (39) has conducted related work
with place-based group knowledge building. His core idea
involves implementing georeferenced annotation that is linked
to a discussion forum focused on arriving at planning decisions.
The resulting "argumap" is essentially a representation of the
development of group knowledge and (perhaps) consensus. As
outlined below, we are beginning to integrate a range of related
visualization and visually enabled group work perspectives into
tools for scientific knowledge work within the HERO project.
HERO Collaboratory
A core goal of the HERO project is to develop the technical and
conceptual infrastructure to support long-term scientific re-
search on local and regional human implications of global
environmental change. A central part of our approach to achiev-
ing this goal is to develop a suite of methods and tools that
facilitate synchronous and asynchronous joint work by small
communities of scientists distributed around a network of sites
across the United States. These methods and tools attempt to
merge exploratory geovisualization tasks, during which concepts
are constructed from data, with knowledge representation sys-
tems that capture the structure of relations between concepts,
data, tools, and people.
HERO scientists are engaged in a variety of research pro-
grams, from developing protocols for data collection, through
building theories and models to explicate multiscale processes of
change, to developing policies to mitigate change. The mecha-
nism used to make these methods and tools accessible to
scientists and to enable joint knowledge construction in a
PNAS 1 April 6, 2004 1 vol. 101 1 suppl. ~ 1 5281
OCR for page 100
spatially and temporally distributed context is a scientific col-
laboratory (defined below).
Scientific Collaboratories: An Overview. The challenge of building
national collaboratories was detailed in a 1993 National Re-
search Council report (404. This report characterizes a collabo-
ratory as a "center without walls, in which the nation's research-
ers can perform research without regard to geographical
location interacting with colleagues, accessing instrumenta-
tion, sharing data and computational resources, and accessing
information from digital libraries." Considerable progress has
been made toward the report goals (e.g., refs. 41-45~. Emphasis
thus far, however, has been on collaboratories that facilitate
research in physical or medical sciences and on real-time data
collection or control of experiments. Only limited progress has
been made in application of the collaboratory concept to the
study of human-environment interaction (46) or to fusing
collaboratory concepts with work in collaborative geographic
information systems (47) or collaborative geovisualization (48~;
see ref. 49 for more on map- and geographic information
system-based collaboration. Also, little work has been done on
application of knowledge representation methods, within col-
laboratories, to capture the semantic relationships between all
the resources that a collaboratory may contain. Carroll et al. (50)
and Chao et al. (51) describe the efforts of other science
communities to use emerging knowledge management and
portal technology to support knowledge construction in science.
The science establishment in the United States has recognized
the need for what has been called "mega-collaboration" to
address critical global problems (ref. 52; Zare was chair of the
National Science Board at the time of this publication), and
human vulnerability and responses to global environmental
change is exactly the kind of problem where such megacollabo-
ration is required. As noted by Finholt (44), barriers to inter-
action across distributed research sites will slow the construction
and integration of the knowledge required to resolve challenging
research questions. The goal of a distributed network, such as
that being developed by HERO, is to bridge place and time by
bringing researchers, the visual concept representation and
sharing tools they use and the knowledge they build, to a single
virtual environment.
Electronic Notebooks: A Vehicle for Acquiring, Constructing, and
Sharing Knowledge. One component of the HERO collaboratory
is a Web portal that integrates knowledge representation and
information visualization tools in an electronic implementation
of a traditional scientific notebook. Whereas paper notebooks
were commonly used to record the development of an individ-
ual's ideas, our collaboratory notebooks are designed with the
sharing and collective exploration of scientific information in
mind. The notebook takes the form of an online workspace that
gives investigators access not just to the digital data and tools
they use (e.g., digital libraries, portals such as these are already
becoming common) but to the abstract concepts constructed by
using these data and methods. HERO workspaces provide a
capacity to do more than just encode elements of scientific
conversations that are easily "digitized." They also facilitate
expressing and storing some of the reflection and reasoning that
is usually tacit in the mind of the researcher. Rather than being
stored in the form of a narrative, as might be common in a paper
notebook, this reflection can be described visually through
concept-graphing tools; the notebook system translates the
resulting diagrams into a description logic-based knowledge
representation language for storage and sharing.
Fig. 2 shows the home page of a user's workspace, providing
access to the people he or she collaborates with, tasks that
describe case studies or analysis procedures, concepts that define
categories and ideas, data files used to create or reflect concepts,
5282 1 www.pnas.org/cgi/doi/10.1073/pnas.0307755101
and online tools that can be used to visualize data and concepts.
By using this portal system to describe elements of scientific
investigations, researchers allow their electronic notebook to
capture the evolution of their ideas and those of the communities
of other users. Such a notebook allows common questions to be
answered in new ways, and even some new questions to be asked,
facilitating a dynamic process of concept and method develop-
ment, extension, and application. For instance,
· Who first coined this concept?
· What data have been used to describe this concept?
· What alternative methods have been used to synthesize this
concept?
What concepts contain or are contained by this concept?
Which individuals and groups have applied this concept?
· Do the reported aims of two individuals using the same
concept agree?
Through the portal, HERO team members have access to
personal workspaces that serve as a nexus for their own thinking
and to group and community workspaces where common con-
cepts can be synthesized from individual descriptions. A user can
choose to make the contents of his or her workspace private or
can make them available for crawlers to find in response to other
users' queries. Group workspaces serve to collect points of
agreement (or disagreement) between collaborating scientists
(e.g., ref. 53), perhaps those working in a particular locality (such
as a watershed) or on a specific problem. By contrast, community
workspaces hold discipline-wide concepts that are broadly
shared. In the context of human-environment research, a com-
munity notebook might define nationally or internationally
agreed on concepts leading to shared protocols for vulnerability
assessment or land use change analysis. Through time, concepts
might migrate up and down such a hierarchy as they find or lose
favor with their research community.
Visually Enabled Concept Building, Sharing, and Application
In this section, we present strategies for integration of explor-
atory geovisualization, information visualization, and diagram-
matic reasoning methods and tools to support concept develop-
ment, representation, and sharing. Specifically, we present some
of our early steps toward achieving the separate aims of (i)
building concepts from data and (~ii) describing and communi-
cating the relationships between concepts. We first discuss work
focused on creating concepts to categorize natural features,
deriving categories from data, and applying those categories to
classification tasks. This work demonstrates the application of a
range of integrated visual-computational methods to a subcom-
ponent of the overall problem of concept building and sharing.
We follow this with an outline of the initial set of methods and
tools developed specifically to support concept building and
sharing among HERO scientists. In the subsequent discussion
section, we detail steps through which our portal approach will
be extended to bridge the currently disconnected fields of
visualization and knowledge sharing and apply them to work in
. in. . .
a specific science c omaln.
From Concepts to Data and Back Again. Whereas it is often useful
to impose a priori concepts on the analysis process, it is equally
important in human-environment science to let concepts
emerge by the combination of data, tools, and other situated
aspects. Also, it is often necessary to mediate prior and emergent
knowledge against each other (when theory and observation are
not in accord). Indeed, it is at this interface that human-
environment researchers regularly confront the dual problems of
incomplete knowledge and incomplete data that characterize
MacEachren et al.
OCR for page 101
i: ~ Tasks ~ ~
~ ::: :: :: ::
~ Start anew task.
: I:
: _
:: :
:~: :~ In f~
it; : ~
::: : ::: :: ::: :
: ~ Concepts
::
~ _
:: ~ ~ :
:~ ~ ~ :: :~: ~ : : _
:: ~_:
:~
~ :~ .^
i: :
:~,.~ - A ~ ~
— !~ _—
i: ~ Places ~~ ~~ ~~' ;~; ~~ ~~ ~
:~ ::::: ::: i::: :::
i: : : : I: ::
::: I:
:: : : : :; : ; : :: ~ : :
~~ ~~ ~~ ~ ~ ~ ~ ~ ~ mv
:: ~ : :: ::: ail: :~ ·.d
~ :~ if:: : i; ~~:~ i: £f`~ "x
::
;:: :: : :: ~ : ::
~~ : tic ~ ~ ~
:: : : ~ : : i: ~ i; : : ~ :: :
Anne HIPPO And The Pennsylvania State university, except as: noted.
F;IPS
Add a new file...
MEL : :
Fig. 2. Interface to knowledge representation
their disciplines.: In this sense, human-environment science is
both a descriptive and a discovery science; a person's under-
standing of concepts both helps to shape and is, in turn, shaped
by interaction with data.
Many practitioners understand well that the creation of con-
cepts is a compromise between their cognitive understanding of
a problem and the emergent properties of the data. Therefore,
concepts both impose structure on data and reveal the structure
already present within the data (54~. Our ultimate aim is to
integrate these top-down and bottom-up approaches to knowl-
edge application and knowledge construction. As noted above,
this integration requires the fusion of two largely disparate
research directions: the encoding and depiction of conceptual
structures, such as situated, dynamic ontologies, representing
what is known, and the support of data exploration and concept
generation to test, refine, or derive conceptual structures, rep-
resenting the discovery of new knowledge. At present, tools to
support these activities are usually separated from each other
with no means of interaction, but in practice activities at either
end of this continuum are not isolated but intimately connected.
As an example, consider the case of land cover and land use
classification. Ontological tools that describe hierarchies of
concepts (such as concepts associated with land use change that
build on the Anderson land cover classification taxonomy) can
tAIthough no shortage of available data exists, these data do not completely describe one's
objects of study. Just as concepts merely refer to more abstract representations in the
mind, data are a proxy for the phenomena they are intended to measure. For instance,
there is no objective measurement for the concept of "vulnerability"; there are only other
phenomena, such as flood frequency or demographics, that may be measured (and even
these, incompletely).
MacEachren et a/.
and construction tools through a web portal.
offer sets of candidate categories from which a computational
classifier might be trained or, conversely, exploring the clustering
of sample points in attribute space might lead one to hypothesize
suitable mental concepts to represent these points.
We have designed and are currently implementing and test-
ing a suite of tools, developed in GeoVISTA Studio (www.
geovistastudio.psu.edu; ref. 55) that connect the top-down pro-
cesses of (i) defining and browsing concepts ontologically, (ii)
selecting specific concepts to use in an analysis exercise, and (iii)
operationalizing the concepts with classifiers with the bottom-up
processes of (iv) data exploration to help formulate concepts
from emergent structures in the data and (v) modification of the
concepts, classifiers, or data used as a result of poor categories
being produced from data (i.e., categories that do not align well
with mental concepts or are not clearly differentiable in the
data). Fig. 3 shows some of these tools, with arrows used to
indicate some of their interactions within the process of geosci-
entific investigation (explained further in the legend to Fig. 3~.
Ontological conceptualizations, as depicted at the upper left
in Fig. 3, are created by individual or groups of researchers using
a concept-graphing tool available through the HERO Web
portal. This tool allows scientists to visually encode knowledge
structures using conceptual graphing techniques. Users of this
tool can produce diagrams to represent the relations between
concepts or the process of an experiment or workflow. The
example shown in Fig. 4 depicts one user's view of the concept
of vulnerability to environmental change. Here, vulnerability is
a product of three "subconcepts": exposure, sensitivity, and
adaptation. Each of these concepts is in turn described by other
concepts. All are linked together by using a set of relationships
with defined semantics that allows the concept graph to be
PNAS 1 April 6, 2004 1 vol. 101 1 suppl. ~ 1 5283
OCR for page 102
ontological conceptualization
P01iCr ~ Ja
0 ~~
_
~3
3tionali~ visual program of analysis sequence
concepts
h ~L;~q g
cam ~—~ ~
\ D ~~
aced rat ~ /
,~ -
C~11
:~ ~ ~ ail:
he
I~ ~
_ ~ ~ ~ ~ 1
~ .! ,,,,~,,,~
resulting map
overlay
Fig. 3. Overview of coordinating bottom-up and top-down approaches to analysis. (a) Concepts to be used in an analysis are extracted from the ontology and
held in an experimental notepad. (b) Design for the experiment is constructed by using the Studio visual programming utility. (c) The data are analyzed for
emergent structures and relationships that can be utilized and for errors and unhelpful attributes that possibly should be removed. (d) The experiment produces
a result set of categories, held intentionally as pieces of a classifier model and extensionally as a map or dataset. (e) Problems with the result can cause the
experimental design to be changed. (c Problems with the result might lead to a reexploration of the data. (g) Problems with the result might cause the user
to modify the concepts being utilized. (h) Modified concepts can be inserted into the ontology, leading to a modified ontology.
decomposed into a set of concept definitions stored in descrip-
tion logic. Some nodes on this graph represent data files, and the
links between these nodes and other concepts suggest what
observations might be used to describe abstract concepts.
Visualizing Structure in Scientific Discourse. In this subsection we
return to the idea, introduced above, of natural language as a
knowledge representation. One component of the HERO col-
laboratory available through the Web portal is a tool for
conducting online Delphi method activities. The Delphi method
(56) is a multiparticipant technique for eliciting and refining
expert belief and is used by HERO researchers to synthesize core
concepts involved in phenomena such as vulnerability to envi-
ronmental change. Through Delphi exercises, we enable another
type of scientific "conversation" to be performed. By using
natural language processing techniques, the key themes in
Delphi discussions can be extracted from the content of partic-
ipants' postings; these themes are then compared against a
lexical database that helps organize them into conceptual hier-
archies, which participants can browse to navigate a discussion
or to summarize the important ideas in the science domain under
discussion. Fig. 5 shows a graph browser displaying concept
relationships from a Delphi discussion on vulnerability (the
browser uses the same underlying technology as that in Fig. 1~.
The key themes in the discussion have been automatically
aggregated to higher levels of abstraction. In this case, the
concepts emerged from the text through bottom-up processing,
but are being viewed by this user in a top-down fashion. At the
5284 1 www.pnas.org/cgi/doi/~0.~073/pnas.0307755101
center of the graph are the most general representations of the
concepts associated with vulnerability, as expressed in the dis-
course. Some nodes have been expanded to show increasingly
specific representations of those concepts and can continue
being expanded until the actual terms used in the discussion
appear. This conceptual graph is populated exclusively with
"kind-of" relationships between concepts, yet it demonstrates a
technique useful for extracting concepts from text data. Ulti-
mately, a Delphi concept map could be produced to show a set
of domain concepts extracted from text discussions or journal
articles; these concepts can then be linked to data and geovisu-
alization tools that would help describe them further.
Discussion (Future Work): Visually Enabled Knowledge Work
As noted above, this article provides just a sketch of a compre-
hensive conceptual approach we are developing for enabling and
understanding the process of concept construction in human-
environment science. Further work needs to be directed toward
formalizing this approach, extending methods for concept visu-
alization, integrating visualization with groupware to enable
visual support for group thinking, and applying the results in the
living laboratory of the HERO project. Specific next steps are
detailed below.
Formalizing the Approach. The framework for formalizing our
approach to concept representation is based on extensions to the
DARPA Agent Markup Language + Ontology Inference Layer
MacEachren et a/.
OCR for page 103
~5~\ = ~~ /
and/
0ebx13"
~ -
~ :~
A
~ date I :.';
/ ~n. rabilz~~~~ - ~ r
~ ~ Tedmo~;jh-~ 5
Ices relic ~~ ~
if /
A" (1900~
GIRL- ~ ~v~~~ ~
;jj~~ Rely Ares / I
~c
, : ~ ~
:
Description: Basic framework ~~:~
vulnerability of people and places to
environmental change produced as
a result of ~e;13sk~hi discussion :~: ~
among vulnerability group. ~~:
created Ola Ah~qvist (coma*)
Created Mon' 11 August 2003 *
: on 15:57 ~ :
Bullt - Exposure
from Economic loss
: Dada
Location tmore..~]
:~ ~9~ :
:: ~,0~,~:
# at 7~5
concepts
Resoorce 1:g80
Used by ~~rarsDhl :
t troll in: none
Fig. 4. A concept graph that depicts a HERO researcher's conceptualization of vulnerability. The graph allows concepts, data, and tools to be linked in visual
descriptions of the research process.
(DAML+OIL) markup language, which can be expressed in
XML. These extensions combine a frame-based syntax with
description logic inference rules. Semantic information is stored
in a discrete and portable format that enables collaborators
to share concepts easily through the HERO knowledge portal.
The framework we have implemented thus far supports con-
struction of concepts that refer to one or more ontologies,
linked concept networks in which any concept can become an
attribute of another concept and ontologies that can be
individual or shared. Ongoing work involves coupling the
formalization of concepts and concept structures to visual
tools that support both direct construction of individual
concepts (by individuals) and collaboration among individuals
:
Concept Map of Delphi topic Vulnerability to Environmental Change
as of 09:05 EDT Wed. Feb 12 2003
region 5ubstencc _
[a ~ ~;~
hazard —~ ~ event
thing ~ state hogan activity
physical oh j ecu ~~ _
- ~ / E5CUSSI ~ Action
Dody ~6ter ~
. ~ —~holoaical f encore
Gun
Epoxies
fiord non
J 1~
Fig. 5. Conceptual graph showing top-down view of concepts extracted
from online discussion on vulnerability.
MacEachren et a/.
to develop shared concepts and ontologies of which they are a
part.
Extending Visualization Methods to Better Support Concept Building,
Representation, and Comparison. One goal here is to develop
visual-computational methods that support comparison of con-
cept maps. These methods will allow scientists to compare their
own representation of a concept with that of other individual
scientists or with the group viewers) derived from Delphi discus-
sions. Such comparisons can reveal points of tension within a
community's view of a domain and help to clarify distinctions
between a novel extension to a concept and the accepted (group)
view. Computational comparison will include graph-similarity
measures (e.g., maximum common subgraph) for evaluating
overlap between multiple ontologies in DAML.
Integrating Visualization with Groupware. A related goal is to draw
on the range of recent developments in methods for visually
enabled group work, diagrammatic reasoning, and argument
visualization and fuse them by exploratory visualization methods
to provide a flexible environment to support group knowledge
building.
Tailor the Methods and Tools for Specific HERO Activities. As a
proof-of-concept test for methods and tools, we will adapt them
for use by HERO team members in individual and community
concept development focused on the concepts of vulnerability,
water resource management, and land use change and the
related concepts from which each is composed.
Each of the objectives above is being advanced through
the work of 12 student researchers who are part of the
HERO Research Experience for Undergraduates Site. These
PNAS 1 April 6, 2004 1 vol. 101 1 suppl. ~ 1 5285
OCR for page 104
students will apply the initial tools to the problem of under-
standing "sensitivity" of local water resources to environmen-
tal change.
1. Wilson, J. T. (1968) Proc. Am. Philos. Soc. 122, 309-320.
2. Giere, R. N. (1988) Explaining Science: A CognitiveApproach (Univ. of Chicago
Press, Chicago).
3. MacEachren, A. M. (1995) How Maps Work: Representation, Visualization and
Design (Guilford Press, New York).
4. Muntz, R. R., Barclay, T., Dozier, J., Faloutsos, C., MacEachren, A. M., Martin,
J. L., Pancake, C. M. & Satyanarayanan, M. (2003) ITRoadmap to a Geospatial
Future: Report of the Committee on Intersections Between Geospatial Information
and Information Technology (Natl. Acad. Press, Washington, DC).
5. Colmerauer, A. & Roussel, P. (1993) SIGPLAN Notices 28, 37-52.
6. MacGregor, R. (1994) in Proceedings of the Twelfth National Conference on
Artificial Intelligence (AAAI-94), eds. Hayes-Roth, B. & Korf, R. E. (Am. Assoc.
for Artificial Intelligence, Seattle), pp. 213-220.
7. Kifer, M., Lausen, G. & Wu, J. (1995) JACM 42, 741-843.
8. Kirschner, P. A., Shum, S. J. B. & Carr, C. S. (2003) VisualizingArgumentation:
Software Tools for Collaborative and Educational Sense-making (Springer,
London).
9. Sowa, J. (2000) Knowledge Representation: Logical, Philosophical, and Compu-
tational Foundations (Brooks/Cole, Pacific Grove, CA).
10. Guarino, N. (1997) Int. J. Hum. Comput. Stud. 46, 293-310.
11. Doel, M. A. (2001) Environ. Plann. D Soc. Space 19, 555-572.
12. Fonseca, F. T., Egenhofer, M. J. & Agouris, P. (2002) Trans. in GIS 6, 231-257.
13. Frank, A. U. (2001) Int. J. Geogr. Inf. Sci. 15, 667-678.
14. Gadamer, H.-G. (2003) Truth and Method (Continuum, New York).
15. Schultze, U. & Boland, R. J. (2000) J. Strategic Inf Syst. 9, 193-212.
16. Nake, F. & Grabowski, S. (2001) Knowledge-Based Syst. 14, 441-447.
17. Winograd, T. & Flores, F. (1986) Understanding Computers and Cognition
(Ablex, Norwood, NJ).
18. Shum, S. B., Uren, V., Li, G., Domingue, J. & Motta, E. (2003) in Visualizing
Argumentation: Software Tools for Collaborative and Educational Sense-Making,
eds. Kirschner, P. A., Shum, S. J. B. & Carr, C. S. (Springer, London), pp.
185-204.
19. Robertson, G. G. (1991) in Proceedings of CHI 91: Conference on Human
Factors in Computing Systems, eds. Robertson, S. P., Olson, G. M. & Olson, J. S.
(Association of Computing Machinery, New York), pp. 189-194.
20. Johnson, B. & Shneiderman, B. (1991) in Proceedings of IEEE Visualization '91,
eds. Nielson, G. M. & Rosenblum, L. J. (IEEE, Piscataway, NJ), pp. 284-291.
21. Lamping, J., Rao, R. & Pirolli, P. (1995) in Proceedings of CHI 95: Human
Factors in Computing Systems, eds. Katz, I. R., Mack, R., Marks, L., Rosson,
M. B. & Nielson, J. (ACM Press, New York), pp. 401-408.
22. Robertson, G., Cameron, K., Czerwinski, M. & Robbins, D. (2002) J. Inf
Visualization 1, 50-65.
23. Chen, C. & Kuljis, J. (2003) J. Am. Soc. Inf Sci. Technol. 54, 435-446.
24. Fluit, C., Horst, H. t. & van der Meer, J. (2002) in On-To-Knowledge:
Content-Driven Knowledge Management Tools Through Evolving Ontologies,
EU-IST Project IST-1999-10132, Report, Dec. 9, 2002, (Commission of the
European Communities, Amsterdam, The Netherlands).
25. Marshall, R. J. (2001) Stat. Med. 20,1077-1088.
26. Hartigan, J. A. & Kleiner, B. (1984) Am. Statistician 38, 32-35.
27. Friendly, M. (1994) J. Am. Stat. Assoc. 89, 190-200.
28. Kuhn, W. & Blumenthal, B. (1996) Spatialization: Spatial Metaphors for User
Interfaces (ACM Press, New York).
29. Wise, J., Thomas, J., Pennock, K., Lantrip, D., Pottier, M., Schur, A. & Crow,
V. (1995) in Proceedings of IEEE Symposium on Information Visualization 1995,
eds. Gershon, N. & Eick, S. (IEEE, Piscataway, NJ), pp. 51-58.
5286 1 www.pnas.org/cgi/doi/10.1073/pnas.0307755101
This work was supported by National Science Foundation Grants
BCS-9978052, BCS-0113030, and BCS-0219025 and by the U.S. Geo-
logical Survey.
30. Skupin, A. (2000) in Proceedings of IEEE Symposium on Information Visual-
ization 2000 (Info Vis 2000), eds. Roth, S. & Keim, D. (IEEE, Piscataway, NJ),
pp. 91-98.
31. Fabrikant, S. I. & Buttenfield, B. P. (2001)Ann.Assoc.Am. Geogr. 91, 263-280.
32. Havre, S., Hetzler, B. & Nowell, L. (2000) in Proceedings of IEEE Symposium
on Information Visualization 2000 (Info Vis 2000), eds. Roth, S. & Keim, D.
(IEEE, Piscataway, NJ), pp. 115-123.
33. Miller, N., Wong, P., Brewster, M. & H., F. (1998) in Proceedings of IEEE
Symposium on Information Visualization 1998, eds. Wills, G. & Dill, J. (IEEE,
Piscataway, NJ), pp. 189-196, 532.
34. Fredrikson, A., North, C., Plaisant, C. & Shneiderman, B. (1999) in Proceedings
of the Workshop on New Paradigms in Information Visualization and Manipu-
lation (NPIVM '99) (ACM Press, New York), pp. 26-34.
35. Andrienko, G. L. & Andrienko, N. V. (1999) Int. J. Geogr. Inf. Sci. 13, 355-374.
36. Gahegan, M., Harrower, M., Rhyne, T.-M. & Wachowicz, M. (2001) Cartogr.
Geogr. Ini Sci. 28, 29-44.
37. MacEachren, A. M., Hardisty, F., Dai, X. P. & Pickle, L. (2003) Commun. ACM
46, 59-60.
38. Suthers, D. (1999) in Proceedings of the 32nd Hawaii International Conference
on System Sciences 1999, eds. El-Renwini, H. & Helal, S. (IEEE Computer
Society Press, Los Alamitos, CA).
39. Rinner, C. (2001) Environ. Plann. B Plann. Design 28, 847-863.
40. Cerf, V. G., Cameron, A. G. W., Lederberg, J., Russell, C. T., Schatz, B. R.,
Shames, P. M. B., Sproull, L. S., Weller, R. A. & Wulf, W. A. (1993) National
Collaboratories: Applying Information Technology for Scientific Research (Natl.
Acad. Press, Washington, DC).
41. Kouzes, R. T., Myers, J. D. & Wulf, W. A. (1996) Computer 29, 40-46.
42. Henline, P. (1998) Interactions May/June, 66-72.
43. Olson, G. M., Atkins, D. E., Clauer, R., Finholt, T. A., Jahanian, F., Killeen,
T. L., Prakash, A. & Weymouth, T. (1998) Interactions May/June, 48-55.
44. Finholt, T. A. (2001) Annul Rev. Ini Sci. Technol. 36, 73-108.
45. Olson, G. M., Malone, T. W. & Smith, J. B. (2001) Coordination Theory and
Collaboration Technology (Lawrence Erlbaum Associates, Mahwah, NJ).
46. Kuhlman, K. M., Soffer, A. & Foresman, T. W. (1997) in Second IEEE Metadata
Conference (IEEE, Piscataway, NJ).
47. Churcher, C. & Churcher, N. (1999) Trans. Geogr. Ini Syst. 3, 23-30.
48. Rhyne, T. M. & Fowler, T. (1998) in ACM SIGGRAPH 98 Course Notes, no.
35 (ACM Press, New York).
49. MacEachren, A. M. (2001) Prog Hum. Geogr. 25, 431-444.
50. Carroll, J., Rosson, M.-B., Dunlap, D. & Isenhour, P. (2003) in Proceedings of
the 36th Hawaii International Conference on System Sciences 2003, ed. Sprague,
R., Jr. (IEEE, Piscataway, NJ), pp. 120-129.
51. Chau, M., Chen, H., Qin, J., Zhou, Y., Sung, W. K., Chen, Y., Qin, Y.,
McDonald, D., Lally, A. & Landon, M. (2002) in Proceedings of the 2nd
ACM/IEEE-CS Joint Conference on Digital Libraries (JCDL '02) (ACM Press,
New York), p. 373.
52. Zare, R. N. (1997) Science 275, 1047.
53. Harvey, F. & Chrisman, N. (1998) Environ. Plann. A 30, 1683-1694.
54. Anderberg, M. R. (1973) Cluster Analysis for Applications (Academic, New
York).
55. Gahegan, M., Takatsuka, M., Wheeler, M. & Hardisty, F. (2002) Comput.
Environ. Urban Syst. 26, 267-292.
56. Turoff, M. & Hiltz, S. (1995) in Gazing into the Oracle: The Delphi Method and
its Application to Social Policy and Public Health, ed. Ziglio, E. (Kingsley,
London).
MacEachren et a/.
Representative terms from entire chapter:
environmental change
