THE INFORMATION VISUALIZER,
`AN INFORMATION WORKSPACE
`
`Stuurt K. Card, George G. Robertson,
`
`Jock D. Mackinlay
`
`Xerox Palo Alto Research Center
`Palo Alto, California
`94304
`(415) 494-4362, Card.PARC@Xerox. COM
`
`of
`
`ABSTRACT
`interface
`the user
`for
`This paper proposes a concept
`information
`called
`an
`information
`retrieval
`systems
`workspace.
`The concept goes beyond the usual notion of
`an information
`retrieval
`system to encompass
`the cost
`structure
`of
`information
`from secondary
`storage
`to
`immediate
`use. As an implementation
`of
`the concept,
`the
`paper
`describes
`an
`experimental
`system,
`called
`the
`Information
`Visualizer,
`and its rationale.
`The system is
`based on (1)
`the use of 3D/Rooms
`for
`increasing
`the
`capacity of
`immediate
`storage avaitable to the user, (2) the
`Cognitive
`Co-processor
`scheduler-based
`user
`interface
`interaction
`architecture
`for coupling
`the user to information
`agents, and (3)
`the use of
`information
`visualization
`for
`interacting with information
`structure.
`
`Information retrieval,
`KEYWORDS:
`interface metaphors,
`desktop metaphor, UI
`information
`visualization,
`animation,
`theory, 3D graphics,
`interactive
`graphics.
`
`INTRODUCTION
`A new paradigm of computing
`use seems to be emerging in
`be applied
`to the storage,
`which
`computational
`aid will
`selection, and use of most sorts of
`information.
`Although
`data bases and information
`retrieval
`systems have been
`around for some time,
`the systems developed
`have relied
`largely on the power of search and indexing
`techniques.
`With
`some important
`exceptions
`(e.g.,
`[1,9,1 1,12])
`few
`systems
`have
`been
`noted
`for
`their
`user
`interfaces.
`Advances
`in
`computer
`technology
`have
`created
`new
`possibilities
`for information
`retrieval
`systems in which user
`interfaces
`could
`play a more central
`role.
`This paper
`proposes a paradigm for such interfaces--the
`information
`An implemented
`experimental
`system,
`the
`workspace.
`Information
`Visualizer,
`is developed as a specific
`instance
`of the paradigm.
`
`of
`
`is
`this material
`all or part
`fee
`without
`to copy
`Permission
`for
`or distributed
`are not made
`the copies
`granted
`provided
`that
`notice
`and the
`direct
`commercial
`advantage,
`the ACM copyright
`title
`the publication
`and its date
`appear,
`and notice
`is given
`of
`that
`copying
`is by permission
`of
`tha Association
`for Computing
`Machinery.
`To copv
`otherwise,
`or
`to republish,
`requires
`a fee
`and/or
`specific
`permission.
`ACM (j.89791.383.3/91
`01991
`
`/()() f)4/()1 81... $ I .50
`
`THE COST STRUCTURE OF INFORMATION
`Information retrieval
`it were a
`has often been studied as if
`self-contained
`problem
`(e.g.,
`the
`library
`automation
`problem).
`Yet
`from the user’s point of view,
`information
`retrieval
`is almost always part of some larger process of
`information
`use. What
`is really needed from the point of
`the user isn’t so much information
`retrieval
`itself, but rather
`the amplification
`of
`information-based
`work processes (or
`other
`uses)--that
`is, methods
`and machines
`that would
`allow people to bring to bear on a task of
`interest more
`information more quickly
`than otherwise possible.
`
`for example, an office worker as shown in Fig. 1.
`Consider,
`is available
`in the desk-side diary,
`through the
`Information
`terminal,
`in the immediate
`files on the desktop,
`computer
`through other people using the telephone,
`in books in the
`bookcase,
`in files in the filing
`cabinet.
`The sources of
`information
`take different
`forms--from paper documents
`to
`machines
`to people,
`but,
`nevertheless,
`each piece
`of
`information
`has a cost
`associated
`with
`finding
`and
`accessing it. Looked at abstractly,
`the office, at a particular
`moment,
`is characterized
`by a cost structure
`over
`the
`information
`in it. What
`is usually meant by an organized
`office is one with a cost structure arranged so as to lower
`the
`cost
`of
`the
`information-based
`work
`processes
`performed within
`it.
`File cabinets, desks,
`filing
`systems,
`and computer-based
`information
`retrieval
`systems can be
`thought of abstractly
`as just means for changing
`this cost
`structure of information.
`
`in the office of Fig. 1
`the information
`The cost structure of
`has been arranged with care (Its arrangement was derived
`from a brochure
`of
`a professional
`time management
`company
`[28]):
`A small amount of
`information
`(either
`frequently-needed
`or
`in immediate
`use) is kept where the
`cost of access is low--in
`an Immediate
`Storage
`area,
`principally
`the desk. Voluminous,
`less-used information
`is
`kept
`in a higher-cost,
`larger-capacity
`Secondary Storage
`area. More information
`is available in the library and other
`Tertiary Norage
`areas.
`In addition
`to these simplified
`categories,
`the
`information
`is
`linked
`and
`otherwise
`structured to aid in its retrieval.
`
`181
`
`1
`
`APPLE 1010
`
`

`

`systems, whether
`processing
`information
`general,
`In
`artificial,
`like this office, or natural, biological
`systems,
`like
`the human eye,
`tend to be organized
`to minimize
`the cost
`structure of
`information
`processing.
`General observations
`deriving
`from studies of
`these systems
`can help us to
`formulate
`systems goals for
`the design of user interfaces
`for information
`access. We consider six such observations.
`
`the parts of
`1 [HIERARCHY]. Organizing
`Observation
`a system hierarchically
`often improves
`the quantity of
`information
`processed relative to processing cost.
`[22]
`
`the standard solutions
`is one of
`arrangement
`Hierarchical
`in biological,
`socioeconomic,
`used to achieve efficiencies
`[26,22].
`The eye is a familiar
`and engineering
`systems
`processing hierarchy.
`The office
`example of a information
`in Fig. 1 is an example of an information
`storage caching
`hierarchy.
`
`2 [HIGH COST RATIOS]. The cost of
`Observation
`information
`often
`varies
`radically
`both
`accessing
`the cost of finding
`it and because of
`the
`because of
`cost of assimilating
`it.
`
`ready to hand vs
`information
`The ratio between the cost of
`available may be
`the cost of
`information
`not
`immediately
`large, even orders of magnitude,
`as shown in Table 1. For
`example,
`in a typical
`computer
`system,
`the ratio between
`main memory
`speed and disk access time in a virtual
`memory
`system is 104-105.
`It may take a scholar months
`to discover and collect
`information materials relevant
`for a
`book. The juxtaposition
`of
`these materials, once they have
`been collected
`into a file or perhaps on a desktop, makes
`
`Secondary
`Storage
`
`Fig. 1. An office organized to have an eflicient
`cost structure.
`
`information
`
`182
`
`going back and forth among them relatively
`
`inexpensive.
`
`TABLE 1. COMPUTER MEMORY COST RATIOS
`
`Storage type
`
`Access time
`
`Ratio
`
`Immediate Storage
`
`Seeondary Storage
`Tertiary Storage
`
`RAM
`
`Disk
`
`80 ns
`20 ms
`
`Optical
`
`5s
`
`2.5 X 105
`
`2.5 X 102
`
`Observation
`processing
`reference.
`references
`distributed
`concentrated
`
`OF REFERENCE]. The
`3 KOCALITY
`exhibits
`locality
`of
`of
`information
`a small
`time
`interval,
`That
`is, over
`are
`not
`unl~orrrlly
`to
`information
`throughout
`the corpu.r,
`but
`tend to be
`in a subset, called the working set.
`
`encountered in studies of
`This fact was first systematically
`computer program memory use [8].
`It also holds true if we
`look at the way in which people reference windows
`[6,14].
`
`BEKRENCE
`4
`Observation
`CLUSTERING].
`information
`used
`use defines clusters of
`Information
`The processing
`of
`to perj50rm some task.
`repeatedly
`tends to establish locality
`of reference in
`information
`then jump
`to another
`cluster.
`one cluster,
`Some
`information may participate
`in more than one cluster.
`
`elements in a working
`the information
`think that
`One might
`change
`as information
`processing
`gradually
`set would
`Instead, what actually
`tends to happen is that
`proceeds.
`there is an abrupt
`transition
`to another working
`set of
`information
`elements [20].
`
`INFO/COST].
`[MAX
`5.
`Observation
`Information
`themselves
`to maximize
`(or
`systems tend to adjust
`the
`quantifi
`of
`information
`sometimes minimize)
`processed relative
`to some processing
`cost constraint.
`[22]
`
`the visual
`is the way
`this observation
`of
`An example
`system tends to encode points of greatest curvature
`in an
`image (these carry
`the most
`information
`[22]). Another
`example
`is the minimum work principle
`in conversation
`wherein
`the speaker attempts
`to anticipate
`some of
`the
`hearer’s
`goals
`and
`reply with
`extra
`information
`that
`minimizes
`the joint cost of the exchange.
`
`levels of an
`Lower
`6 [ABSTRACTION],
`Observation
`and organize
`processing
`system simpllfy
`information
`supplying
`higher centers with aggregated
`information,
`forms of information
`through abstraction
`and selective
`omission.
`[22].
`
`is in hand there is a generic
`information
`Even when
`the volume of
`information
`to be processed is
`problem that
`large relative to the abilities
`of
`the user. An answer
`is to
`stage
`processing
`by
`recoding
`the
`information
`in
`progressively more abstract and simpler
`representations.
`
`2
`
`

`

`processing
`by the lower-level
`produced
`The abstractions
`predetermine,
`to a considerable
`extent,
`the patterned
`structures that
`the higher-level
`processing can detect
`[22].
`The higher-level
`processing,
`in turn,
`reduces still
`further
`the quantity of
`information
`by processing it
`into yet more
`abstract and universal
`forms.
`In biological
`systems,
`this
`process
`allows
`the mixing
`of
`information
`generated
`through different
`sensory modalities.
`
`INFORMATION WORKSPACES
`retrieval, narrowly
`If we want
`to move beyond information
`of
`information-
`conceived,
`to address the amplification
`based work processes, we are led to try to develop
`user
`interface
`paradigms
`oriented
`toward managing
`the cost
`structure of
`information-based
`work. This,
`in turn,
`leads us
`to be concerned not
`just with the retrieval
`of
`information
`from a distant source, but also with the accessing of
`that
`information
`once it
`is retrieved
`and in use.
`And
`this
`problem,
`the necessity of
`lowering
`the cost of work by
`providing
`some sort of
`low-cost,
`immediate
`storage for
`accessing objcds
`in use is a common
`problem faced by
`most kinds of work. The common solution is a workspace,
`whether
`it be a woodworking
`shop, a laboratory,
`or an
`office. A workspace is a special environment
`in which the
`cost structwe
`of
`the needed materials
`is tuned
`to the
`requirements
`of the work process using them.
`
`for
`of a sort
`a workspace
`screens also provide
`Computer
`(and, of course, may be
`tasks done with
`the computer
`components
`of
`larger
`workspaces,
`such
`as offices).
`Computer-screens
`as workspaces
`have
`gone
`through
`several stages of evolution
`(Fig. 2). Early workspaces were
`command-based
`scrolling
`teletypes or their CRT equivalent
`and this style survives
`in DOS and UNIX systems today.
`Engelbart’s NLS system [10]
`introduced the notion of direct
`interaction with the text
`in documents
`using a point and
`click editor based on the mouse.
`
`Information
`Workspace
`
`Multple Shared
`Desktop~
`
`INFORMATION
`
`VISUALIZER
`
`+
`ROOMS
`
`4
`
`DATALAND
`
`PROJECTSLBIGSCREEN
`
`Multple Desktops:
`
`SMALLTALI$
`
`Lwge Desktop:
`
`Desktop Metaphoc
`
`SMALLTALK
`
`Point & Click Editors:
`
`Ids
`
`Fig. 2. Evolution
`
`of computer workspaces.
`
`The next stage was the desktop metaphor developed as part
`of
`the Smalltalk
`[15] and Star
`[27]
`systems. By adding
`menus,
`windows,
`and
`icons
`to mouse-based
`direct
`interaction,
`these systems allowed the workspace to contain
`
`183
`
`from
`user processing
`and shifted
`documents
`multiple
`The desktop metaphor
`recall-based
`to recognition-based.
`affected the cost structure of
`information
`by allowing
`low
`cost access to more information
`in the Immediate Storage
`environment.
`Smalltalk Projects introduced
`the notion of
`multiple
`workspaces
`that
`users
`could
`switch
`among,
`allowing
`still more information
`to reside in the immediate
`work area (but at
`the added cost of switching
`and finding
`the right workspace).
`
`desktop
`extended
`step was the large,
`next
`A potential
`The problem with
`introduced
`by Dataland
`[1].
`metaphor,
`the single large workspace (as we found in BigScreen
`[14],
`another attempt at a large desktop)
`is that
`the cost of search
`for
`relevant parts of
`the workspace rapidly
`increases with
`the number of elements in the workspace (unless the space
`itself has meaning
`as in a city map or a grocery
`store).
`Even
`the CCA
`version
`[1]
`of
`the Dataland
`system,
`developed
`for practical
`use, abandoned
`the single
`large
`workspace in favor of multiple workspaces.
`
`desktop
`The Rooms system [4,14] added to the multiple
`objects in
`notion an ability
`to share the same information
`different workspaces, both individually
`and as groups.
`It
`also added an overview and other navigational
`aids as well
`These
`as an ability
`to store and retrieve workspaces.
`removed the major disadvantages of multiple
`desktops.
`
`for
`interface
`a user
`for
`The essence of our proposal
`the Rooms multiple
`information
`retrieval
`is to evolve
`desktop metaphor
`into a workspace for
`information
`access-
`-an information
`workspace.
`Unlike
`the conventional
`information
`retrieval
`notion
`of
`simple
`access
`of
`information
`from some
`distal
`storage,
`an information
`of
`workspace
`(1)
`treats
`the
`complete
`cost
`structure
`information,
`integrating
`information
`access from distant,
`secondary or tertiary
`storage with information
`access from
`Immediate
`Storage
`for
`information
`in
`use,
`and
`(2)
`of
`considers
`information
`access part
`a larger
`work
`processes.
`That
`is,
`instead of concentrating
`narrowly
`on
`the control of a search engine,
`the goal
`is to improve
`the
`cost structure of
`information
`access for user work.
`Our
`intention
`is to build on progress
`in information
`retrieval
`studies, but
`to do so by reframing
`the problem as the
`amplification
`of
`information-dependent
`work processes.
`
`is our
`in this paper
`described
`Visualizer
`The Information
`this concept. The Information
`of
`experimental
`embodiment
`Visualizer
`has three major components
`(see Table 2):
`(1)
`3D/Rooms,
`a 3-D vecsion
`of
`the Rooms
`system. This
`component
`effectively
`increases the capacity of
`Immediate
`Storage,
`thus making a more effective
`hierarchical
`storage
`system in accord with Observations
`1-4. (2) The Cognitive
`Co-Processor
`[231, an animation-oriented
`user
`interface
`architecture.
`This
`component
`increases
`the rate of user-
`systcm interaction
`and information
`t.mnsfer
`in accord with
`Observation
`5. And (3)
`information
`visualizations,
`which
`structure-oriented
`browsers
`into
`sets
`of
`serve
`as
`information.
`This
`component
`increases
`the level
`of
`abstraction
`
`3
`
`

`

`TABLE 2. SUMMARY OF THE 13ESIGN RATIONALE
`
`FOR THE INFORMATION
`
`VISUALIZER.
`
`PROBLEM
`
`ANALYSIS/
`OBSERVATIONS
`
`SYSTEM
`GOALS
`
`3D/RooMs
`
`TX ARTIFACT
`COGNTITVE
`CO-PROC-
`ESSOR
`
`lNFORMA-
`TION VISUAL-
`IZATIONS
`
`tid
`nformation
`ccess&
`Iroeessing
`
`1. HIERARCHY
`2. HIGH COST RATIOS
`3. LOCALITY
`OF REF.
`4. REF. CLUSTERING
`
`5. MAX INFO/COST
`
`6. ABSTRACTION
`
`More Immediate Storage
`Larger
`Denser
`
`Cheaper Secondary
`Storage access
`
`systems
`Highly-coupled
`Iterative retrieval
`Faster cycle
`Fewer cycles
`Cognitive
`impend-
`ante match
`
`visualization
`Information
`Linear structure
`Hierarchical
`structure
`Continuous
`data
`Geographical
`data
`
`*
`*
`
`o
`
`o
`0
`o
`
`o
`o
`o
`o
`
`c)
`*
`
`0
`
`*
`0
`*
`
`0
`0
`0
`0
`
`o
`*
`
`e
`
`o
`e
`o
`
`@
`@
`is
`a
`
`abstraction between the user and the available
`in accord with Observation
`6.
`
`information
`
`3D/Rooms
`Storage:
`Immediate
`Increasing
`the office
`cost structure of
`Let us return to the information
`in Fig.
`1.
`Clearly,
`it
`is advantageous
`to put more
`information
`into the cheaper
`Immediate Storage area. The
`problem (for a desktop or a computer desktop metaphor)
`is
`that
`the storage only holds so much. When the storage is
`overloaded
`by overlapping
`information,
`then
`searching
`time
`can raise
`the access cost of
`Immediate
`Storage
`radically
`and
`non-linearly,
`defeating
`the
`purpose
`of
`Immediate Storage and leading to phenomena analogous to
`thrashing [14].
`
`shows that a local
`reference,
`of
`locality
`3,
`Observation
`workspace is viable (and this is why the desktop metaphor
`works).
`Observation
`4,
`reference clustering,
`tells us that
`the working
`set of
`items referenced
`is likely
`to undergo
`sudden shifts to other working
`sets. This is the basis of
`the
`Rooms multiple workspace concept, which allows the user
`to organize his or her work into (partially
`overlapping)
`sets
`Rooms
`of
`items
`and to switch
`among
`them easily.
`effectively makes the Immediate Stomge kw-ger by having
`multiple
`desktops (at some cost
`increase due to switching).
`It avoids the overload problem,
`by switching
`the screen to
`show only those items in the active Room.
`
`In the
`extends this logic to a 3D workspace.
`3D/Rooms
`classical desktop metaphor and the original Rooms system,
`the view of a Room is fixed.
`In 3D/Rooms,
`the user
`is
`
`He or she
`in the Room.
`given a Dosition and orientation
`&m mo;e
`about
`the Room,
`zoom in to examine
`objects
`closely,
`look behind himself,
`or even walk
`through doors
`The
`3D/Rooms
`workspace
`is
`into
`other
`Rooms.
`sufficiently
`different
`from the desktop metaphor
`that
`it
`requires a new set of building
`blocks: We developed
`a
`walking metaphor algorithm for exploratory movement
`of
`the user [17] and a point of
`interest
`logarithmic movement
`algorithm for very rapid, but precise movement
`relative to
`objects
`of
`interest
`[18].
`Another
`algorithm allows
`3D
`objects
`to be moved rapidly
`using only
`the mouse [18].
`3D/Rooms
`is built upon an animation-based
`user interface
`architecture--the
`Cognitive Co-processor
`[23]. The buttons
`of Rooms have been developed into autonomous
`interactive
`objects,
`Like Rooms,
`3D/Rooms
`contains
`an overview
`allowing
`the
`user
`to
`view
`all
`the
`3D workspaces
`simultaneously
`(Card Plate 1). But
`in 3D/Rooms
`the user
`can aetuall y reach into the Rooms from the overview, move
`about
`in them, and manipulate
`their objects.
`
`(and the associated Cognitive Co-
`The effect of 3D/Rooms
`is to make the Immediate
`Storage
`processor architecture)
`not only larger, but also denser.
`J. J. Gibson has made the
`point
`that perception
`is an active process [13],
`Instead of
`identifying
`what
`is visible
`on the screen with
`the user’s
`perceptual awareness,
`it
`is probably more accurate to think
`of
`the visuat and perceptual
`system as a flying spot scanner
`that
`is updating an internal awareness [22].
`Just as a driver
`is aware of a car behind him glimpsed
`in his mirror a few
`seconds previous,
`the user seems to be able to be aware of
`objects that are “behind”
`him or occluded if he can easily
`
`184
`
`4
`
`

`

`Application
`
`these
`in general
`For example,
`these agents.
`of
`properties
`from those of
`the
`agents operate on time constants different
`user. There are th~e sorts of
`time constants for
`the human
`that we want
`to tune the system to meeh perceptual
`processing,
`immediate response, and unit
`task (Table 3).
`
`..
`T.wk
`Machine
`
`user
`Discourse
`Machine
`
`use,
`
`Task Queue
`
`4
`
`Dwplay Queue
`
`I
`
`Fig. 3. Cognitive Co-proce5sor
`
`interaction
`
`architecture,
`
`them.
`of
`(say within about a second) update his knowledge
`graphics
`Thus interactive
`animation
`and 3D perspective
`tenants
`both allow us to apply Gibson’s active perception
`than
`and to pack the space more densely with information
`would otherwise
`be possible.
`By manipulating
`objects or
`moving in space,
`the user can disambiguate
`images,
`reveal
`hidden
`information,
`or
`zoom in
`for
`detail--rapidly
`accessing more information.
`For example,
`in a companion
`paper
`[24], we describe
`a corporate
`organization
`tree
`requiring
`80 pages on paper
`that has been displayed
`in a
`single 3D/Rooms
`screen.
`
`Coupling with the
`
`Interaction
`Information
`Increasing
`User:
`the Cognitive Co-processor
`for a user to
`is difficult
`it
`Except
`in speciaI circumstances,
`ask an information
`retrieval
`system what he or she wants,
`because the user does not,
`in geneml,
`know what
`is
`available
`and does not know from what
`it has to be
`For
`this
`reason, we have adopted
`an
`differentiated.
`iterative retrieval philosophy,
`in which the user is assumed
`to iterate with the system through several cycles until
`the
`use finds what
`is wanted.
`This is,
`in fact, similar
`to what
`people do when they make requests to reference librarians.
`In order
`to maximize
`the information
`gained as a function
`of
`time cost, we attempt, on the one hand,
`to increase the
`speed of
`this iterative
`cycle
`as much
`as possible
`(for
`example, by using graphics to speed user assimilation),
`and
`on the other
`hand to decrease
`the number
`of
`times
`necessary to go around the cycle (for example,
`by using
`fish-eye [12]
`focus + context displays to speed navigation).
`
`the user-system interaction, we
`rapid cycles of
`To support
`interface
`interaction manager
`have created a new user
`substrate called the Cognitive Co-processor,
`derived from
`Sheridan’s notion of supervisory
`control
`[25,2]. The idea is
`that
`the
`user
`is
`trying
`to
`control
`possibly multiple
`The
`applications
`running
`as semi-autonomous
`agents.
`Cognitive Co-processor
`(Fig. 3)
`is meant
`to be a sort of
`“impedance matcher” between the cognitive
`and perceptual
`information
`processing
`requirements
`of
`the user and the
`
`185
`
`TABLE 3. HUMAN TIME CONSTANTS FOR TUNING
`
`COGNITIVE CO-PROCESSOR
`
`TIME CONSTANT
`
`VALUE
`
`REFERENCES
`
`Perceptual processing
`
`.1 s
`
`Immediate response
`
`Unit
`
`task
`
`1s
`10 s
`
`[5]
`
`[21]
`
`[5,21]
`
`The Cognitive
`time constant.
`processing
`The perceptual
`is based on a continuously-running
`scheduler
`Co-processor
`loop and double-buffered
`graphics.
`In order
`to maintain
`the illusion
`of animation
`in the world,
`the screen must be
`repainted
`least every
`.1 sec [5].
`The Cognitive
`Co-
`at
`processor
`therefore
`has a Governor
`mechanism
`that
`When
`the cycle
`time
`monitors
`the basic
`cycle
`time.
`becomes too high, cooperating
`rendering
`processes reduce
`the quality of
`rendering
`(e.g.,
`leaving off most of
`the text
`during motion)
`so that
`the cycle speed is increased.
`
`The immediate response time constant. A person can make
`an unprepared
`response to some stimulus within
`about a
`If
`second [21].
`there is more than a second,
`then either
`the
`listening
`party makes a backchannel
`response to indicate
`“uh-huh”)
`or the speaking pmty
`that he his listening
`(e.g.,
`“uh...”)
`to indicate
`he is still
`makes a response
`(e.g.,
`thinking
`of the next speech. These serve to keep the parties
`of
`the interaction
`informed
`that
`they are still engaged in an
`interaction.
`In the Cognitive Co-processor, we attempt
`to
`have agents provide
`status feedback at
`intervals no longer
`than this constant.
`Immediate
`response animations
`(e.g.,
`swinging
`the branches of a 3D tree into view) are designed
`to take about a second.
`If
`the time were much shorter,
`then
`the user would
`lose object constancy
`and would
`have to
`reorient himself.
`If
`they were much longer,
`then the user
`would get bored waiting for the response.
`
`Finally, we seek to make it
`task time constant.
`The unit
`possible for
`the user to complete some elementary
`task act
`within
`10 sw (say, 5-30 SW) [5,21], about
`the pacing of a
`point
`and click
`editor.
`Information
`agents may require
`considerable
`time to complete
`some complicated
`request,
`but
`the user,
`in this paradigm,
`always stays active. He or
`she can begin
`the next
`request
`as soon as sufficient
`information
`has developed from the last or even in parallel
`with it.
`
`Increasing
`Visualization
`According
`proceeds
`information
`
`the Abstraction
`
`of
`
`Information--Information
`
`often
`processing
`information
`6,
`to Observation
`in
`the
`system simplifying
`by
`lower
`levels
`through aggregation,
`abstraction
`and selective
`
`5
`
`

`

`large amounts of raw information
`In this way,
`omission.
`are reduced to volumes within
`the capacity of
`the higher
`centers and the abstractions introduced by the lower centers
`can be further aggregated into patterns.
`
`shows how the
`visualization
`in scientific
`Recent work
`in the process of
`computer
`can serve as an intermediary
`abstraction.
`Large sets of data are reduced to graphic form
`in such a way that human perception
`can detect patterns
`revealing underlying
`structure in the data more readily than
`by a direct anatysis of
`the numbers.
`Information
`in the
`form of
`documents
`also
`has
`structure.
`Information
`visualization
`attempts
`to display
`structural
`relationships
`and context
`that would
`be more difficult
`to detect by
`individual
`retrieval
`requests.
`
`across many
`common
`are
`structures
`abstract
`Some
`is hierarchical
`structure
`information
`sets. One example
`A companion
`paper
`[24]
`(e.g., UNIX
`directories).
`describes structural browsers called the Cone Tree and the
`Cam Tree,
`based on an animated
`3D visualization
`of
`hierarchy.
`Another
`example
`is linear
`structure.
`We
`discovered in field observations
`of an architect’s office,
`for
`example,
`that
`time
`of
`creation was one of
`the most
`important
`retrievat attributes of a document
`since it related
`so intimately
`to the work process.
`In another companion
`paper
`[19], we describe
`a structural
`browser,
`called the
`Perspective Wall
`that allows retrieval
`using a visualization
`of linear structures.
`
`to
`data surface similar
`A third example is a 2D continuous
`is a
`much scientific
`data.
`In this case our visualization
`Data Sculpture
`(see Card Plate 2).
`The user can walk
`around
`or zoom into
`this visualization
`containing
`over
`65000
`sampling
`points
`as if
`it were a sculpture
`in a
`museum. The user can also manipulate
`some of
`its viewing
`parameters.
`A fourth example is the spatial structure of a
`building.
`Card Plate
`3 shows a portion
`of an office
`building
`used as a structural browser
`for people. Selecting
`an organization will produce the names and pictures of
`its
`members
`and select
`their
`offices.
`Clicking
`on offices
`retrieves their
`inhabitants.
`
`graphics
`computer
`use interactive
`These visualizations
`dynamically
`changing
`views of
`the
`animation
`to explore
`information
`structures. More visualizations
`are visible
`in
`the Rooms
`overview
`of Card Plate
`1.
`The visualizers
`attempt
`to present abstractions
`of
`large amount
`of data
`tuned to the pattern
`detection
`properties
`of
`the human
`perceptual
`system.
`For example,
`they use color,
`lighting,
`hidden
`shadow,
`transpanmcy,
`surface
`occlusion,
`continuous
`transformation,
`and motion
`cues to induce
`object constancy and 3D pe~pective
`
`and Searching
`Indexing
`here on the user interface paradigm
`We have concentrated
`aspects of
`the Information
`Visualizer.
`The Information
`Visualizer
`is based on an indexing
`and search subsystem
`built
`by other members
`of our group
`[7].
`Briefly,
`this
`subsystem TDB provides stemming and a full
`text
`inverted
`
`as word vectors as
`are represented
`database. Documents
`are requests. This allows us to search for documents given
`a set of descriptors
`or to use documents
`themselves as the
`retrieval
`request
`to find other documents
`in an iterative
`“relevance
`feedback
`In
`one
`retrievat
`paradigm.
`demonstration,
`for example,
`biographies
`of several
`staff
`members who are linguists
`are selected.
`The result
`is that
`on the next
`retrieval
`iteration more linguists
`are retrieved.
`Associative
`retrieval
`based on such linguistic
`searches can
`be
`used
`to
`highlight
`portions
`of
`the
`information
`Thus we can combine
`associative
`and
`visualization.
`structural searches.
`
`DISCUSSION
`is an
`described
`have
`we
`Visualizer
`The
`Information
`used to develop
`a new user
`experimental
`system being
`interface paradigm for
`information
`retrieval,
`one oriented
`toward the amplification
`of
`information-based
`work.
`It
`is
`based on our analysis of seveml aspects of
`information
`use
`that have led us to reframe
`the information
`retrieval
`problem as a problem in the cost
`structuring
`of
`an
`information
`workspace.
`This,
`in turn, has led us to evolve
`the computer
`desktop metaphor
`toward (1) 3D/Rooms
`(to
`manage
`information
`storage
`cost hierarchies),
`(2)
`the
`Cognitive Co-processor
`interaction
`architecture
`(to support
`highly-coupled
`iterative
`interaction with multiple
`agents),
`and (3)
`information
`visualization
`(to increase the level of
`information
`abstraction to the user).
`
`a set of studies in which we have
`This paper continues
`theoretical
`and empirical
`analyses
`attempted
`to integrate
`with practicat
`system design, either
`through the analytical
`characterization
`of
`existing
`designs
`[5,16,17]
`or
`the
`synthetic
`generation
`of new designs based on analytical
`underpinnings
`[4, 14,16, 18,23].
`The
`development
`of
`theoretical methods and practical
`designs in engineering-
`oriented disciplines
`tends to take a different
`course than in
`the natural
`sciences owing
`to the particular
`interplay
`of
`synthetic
`and analytic
`activities.
`This
`course
`can be
`summarized
`as the systems research paradigm:
`(1) Initial
`exploratory
`point
`designs demonstrate
`the possibility
`of
`obtaining
`some performance.
`These may be incrementally
`improved
`through
`cut and try.
`(2) Abstractions
`are
`developed describing
`the structure of regions in the design
`space.
`(3) Theories and empirical
`observations
`are used to
`characterize
`sub-regions
`of
`the design
`space,
`showing
`which
`designs
`lead to what performance.
`(4) Finally,
`knowledge
`of
`the design space is codified
`in such a way
`that
`it can be transmitted
`as a body of knowledge
`to other
`people who need to build such systems.
`
`science
`natural
`This sequence reverses the more familiar
`course of
`theory to application.
`But, actually,
`the interplay
`of system synthesis, abstraction,
`and analysis may play out
`in almost any order, and there can be subregions
`of
`the
`design space in very different
`states.
`In the present study
`we
`have
`been
`able
`to
`utilize
`theory
`and
`empirical
`relationships
`established
`from
`previous
`research
`[5,6,8,12,14,20,21
`,22,25,26]
`as well
`general
`theoretical
`observations
`from the literature
`as “tools
`for
`thought”
`to
`
`186
`
`6
`
`

`

`Card Plate 1
`
`Card Plate ?
`
`
`
`
`Card Plate 3
`
`187
`
`7
`
`

`

`even though we have adopted
`the problem,
`conceptualize
`The
`system-building
`methodology.
`an
`exploratory
`explomtory
`system building,
`in turn, brings us to a position
`where we will be able to perform empirical
`use studies and
`Regardless
`of
`the order,
`the
`design
`characterization.
`general need is both for new user interface paradigms
`that
`utilize
`emerging
`technological
`possibilities
`and
`the
`analytical
`and empirical
`foundations
`that
`help
`us to
`understand the merits of
`these designs and the possibilities
`for new ones.
`
`Learning
`
`Friend
`
`21
`
`The
`A.
`Interaction.
`
`REFERENCES
`Lifetime
`The Human Interface.
`1. Bolt, R. A.
`1984.
`Publications, Belmont, California,
`In
`2. Card, S, K. Human factom and artificial
`intelligence.
`P. A. Hancock & M. H. Chignell
`(Eds.),
`Intelligent
`Interfaces:
`Theory, Research and Design.
`Elsevier
`Science Publishers B.V.
`(North-Holland),
`Amsterdam,
`1989.
`Workspaces.
`Information
`3. Card, S. K.
`Conference (Tokyo, September),
`1989.
`virtual-
`4. Card, S. K. & Henderson, A. H, Jr. A multiple,
`workspace
`interface
`to support
`user
`task switching.
`Proceedings
`of
`the CHI+GI
`1987 (Toronto, April
`5-
`7). ACM, New York, 1987, pp. 53-59.
`5. Card, S. K., Moran,
`T. P., & Newell,
`Psychology
`of
`Human-Computer
`Erlbaum, Hillsdale, New Jersey, 1983.
`6. Card, S.; Pavel, M., & Farrell,
`J. Window-based
`computer
`dialogues.
`In Brian Shackel
`(Ed), Human-
`Computer
`Interaction--[nteract
`’84 (September
`4-7).
`North-Holland,
`Amsterdam,
`1984, pp. 239-243.
`7. Cutting, D. & Pedersen, J. Optimization
`for dynamic
`inverted
`index maintenance.
`ACM SIGIR ’90, 13th
`International
`Research
`Conference
`on
`and
`in
`Development
`Information
`Retrieval.
`(Brussels,
`September 5-7), ACM, New York, pp. 405-411.
`IEEE
`8. Denning, P. J. Working
`sets past and present.
`Transactions
`on Software Engineering,
`SE-6(1 1980):
`66-84.
`J. Whitetxmrds:
`J. & Widom,
`9. Donahue,
`ACM
`Transactions
`tool.
`database
`Systems 4(1 1986), 24-41.
`Information
`A
`intellect:
`D. C.
`Augmenting
`human
`10. Engelbart,
`Stanford
`framework
`(AFOSR-3223).
`conceptual
`1962,
`Research Institute, Menlo Park, California,
`11. Fairchild,
`K. M.; Poltrock,
`S. E.; & Fumas, G. W.
`SemNet: Three-dimensional
`graphic representation
`of
`large
`knowledge
`bases.
`In R. Guindon
`(Ed.),
`Cognitive
`Science and its Applications
`for Human-
`Computer
`Interaction.
`Erlbaum, New Jersey, 1988.
`12. Fumas, G. Generalized
`fisheye views.
`In M. Mantei
`and P. Orbeton (Eds.), ACM CHI
`’86 Conference on
`Human
`Factors
`in Computing
`Systems
`(Boston,
`Massachusetts,
`April,
`14-18). ACM SIGCHI,
`New
`York, 1986, 16-23.
`to Visual
`Approach
`13, Gibson,
`J. J.
`The Ecological
`Perception.
`Boston: Houghton Mifftin,
`1979.
`
`A graphical
`on Office
`
`The use
`14. Henderson, D. A., Jr. & Card, S. K. Rooms:
`space
`of multiple
`virtual
`workspaces
`to
`reduce
`contention
`in a window-based
`graphical user interface.
`ACM Transactions
`on Graphics 5(3, July 1986),21 1-
`243.
`computer.
`and the personal
`15. Kay, A. Microelectronics
`Scientific American
`237(3, September 1977), 230-244.
`16. Mackinlay,
`J. D. Automating
`the design of graphical
`ACM
`presentations
`of
`relational
`information.
`Transactions
`on Graphics, 5(2, Aprit 1986), 110-141.
`17. Mackinlay,
`J. D., Card, S. K., & Robertson, G. G. A
`semantic analysis of
`the design space of
`input devices,
`Human-Computer
`Interaction
`5(2, 1990), 145-190.
`18. Mackinlay,
`J. D., Card, S. K., & Robertson, G. G.
`Rapid
`controlled
`movement
`through
`a virtual
`3D
`workspace.
`SIGGRAPH ’90 Conference Proceedings
`(Dallas, Texas, August 1990).
`In Computer Graphics,
`24(4, August, 1990), 171-176.
`19. Mackinlay,
`J. D., Robertson, G. G., & Card, S. K.
`Perspective
`wall:
`Detail
`and
`context
`smoothly
`integrated.
`ACM CHI
`’91 Conference
`on Human
`Factors
`in Computing Systems (New Orleans, April),
`ACM, New York, 1991.
`20. Madison, A. W. Characteristics
`Ann
`Arbor,
`Michigan:
`International,
`1992.
`Harvard
`of Cognition.
`The