Where No Mind Has Gone Before:Ontological Design for Virtual Spaces

Nancy Kr@itlStuartMoulthrop

Institute for Publications DesignThe University of Baltimore

1420 North Charles StreetBaltimore, Maryland U.S.A. 21201-5779

nakaplan@ubmail .ttbahdusamoulthrop@ttbmail. ubalt.edtt

ABSTRACT

Hypermedia designetx have tried to move beyond the directedgraph concept, which defines hypermedia structures asaggregations of nodes and links. A substantird body of workattempts to describe hypertext in terms of extended orglobal spaces. According to this approach, nodes and linksacquire meaning in relation to the space in which they aredeployed. Some theory of space thus becomes essential forany advance in hypermedia design; but the type of spaceimplied by electronic information systems, fromhyperdocttments to “consensual hallucinations,” requirescareful analysis. Familiar metaphors drawn from physics,architecture, and every&y experience have only limiteddescriptive or explanatory value for this type of space. Astheorists of virtual reality point out, new informationsystems demand an internal rather than an externalperspective. This shift demands a more sophisticatedapproach to hypermedia space, one that accounts both forstable design properties (architectonic space) and forunforeseen outcomes, or what Winograd and Flores call“breakdowns.” Following Wexelblat in cyberspace theoryand Dillon, McKnight, and Richardson in hypermediatheory, we call the domain of these outcomes semanticspace, In two thought experiments, or brief exercises ininterface design, we attempt to reconcile these divergentnotions of space within the conceptual system ofhypermedia.

KEY WORD S: Spatial hypertext, interface design,information mapping, navigation

Permission to copy without fee all or part of this materiat is

granted provided that the copies are not made or distributed for

direct commercial advantage, the ACM copyright notice and

the title of the publication and its date appear, and notice is

given that copyright is by permission of the Association for

Computing Machinery. To copy otherwise, or to republish,

requires a fee and/or specific permission.

63 1994 ACM 0-89791 -640-9/94/0009/$3 .50

1 INTRODUCTION: NODE, LINK, SPACE

Hypermedia designers and theorists have tried repeatedly tomove beyond the directed graph concept, which defineshypermedia strttchtres as aggregations of nodes (containersof information) and links (dynamic connectors). Halasz[11] folds these primitives into node/link “composites.”DeRose [7] distributes link functions over a taxonomy of“intensional” and “extensional” links, while Nanard andNanard [23] envision a type scheme for link anchors aswell. Parttnak [27] replaces the concept of linked nodeswith set-theoretic groupings, while Marshall et al. [20]propose a frame-based model. Stotts, Fttruta, and Ruiz [32]present a document-browsing automaton that operates onlink structures alone. Theorists interested in the graphicdimension of hypermedia structure, such as Lai and Manber[15], Gloor [10], and Noik [24], propose formalisms for

surveying and analyzing spatial relationships within a text,

These conceptions are explored most profitably in Marshall

and Shipman’s work [21] on the expression of implicitstructures through spatial arrangement. In their approach,nodes and links acquire meaning in relation to the space inwhich they are deployed. Accordingly, some theory ofspace seems essential for any advance in hypermedia design;but the type of space implied by electronic informationsystems, from hyperdocuments to the “consensualhallucinations” of virtual reality, requires careful analysis.Bolter [4] suggests that hypertext creates a new “writingspace”; writing about space in virtuat reality systems,Wexelblat [34] invokes the term “semantic space.” Thenature of this space resists easy definition. Familiarmetaphors from physics, architecture, and everydayexperience have only limited value here. As Dillon et al.[8] observe, the psycholinguistic or semantic space of a text(electronic or otherwise) can never be represented withperfect accuracy by any physical system. A radicallysubjective element necessarily comes into play. Novak[26] points out that contemporary information systemsdemand an internal rather than an external perspective.Virtual space is an interactive continuum defined by our

ECHT ’94 Proceedings 206 September 1994

committed participation as well as our simple attention. Inthe space of electronic media, we are always more thandktallced observers.

Proceeding from this starting point, we offer a criticalreview of two issues in hypermedia design: transition(“navigation”) and information mapping. Our approach tothese issues is guided by Winograd and Flores’s notion [33]of “ontological design” — a design philosophy thataddresses human-machine interactions in terms of complexenvironments instead of simple, end-directed functionality.In response to Winograd and Flores’s demand that newdesign be based on conceptual complication or“breakdown: we offer two thought experiments that mighthelp redefine our understanding of navigation and mapping.

2 ARCHITECTONIC AND SEMANTICSPACES

Marshall and Shipman’s work on VNS, NoteCards., andAquanet represents a crucial advance in thinking about thespatiality of hyperdocuments. In place of “explicit” linksthey concentrate on the arrangement of nodes in a graphicrepresentation, permitting their system to address an“implicit” level of structure. This approach escapes theconstraints of the traditional directed-graph model withoutsacrificing the visual dimension of hypermedia. However,Marshall and Shipman’s conception reflects only one aspectof the complex phenomenology of virtual space. InMarshall and Shipman’s work, as in other major systemssuch as Writing Environment, SEPIA, and Storyspace, thegeneral idea of space tends to collapse into the muchnarrower domain of screen real estate. The user’smanipulation of objects within a graphic representationimplies some related transformation in a mentad orlinguistic space, but that space is accessible only throughthe representation. “Space” comes to be defined in terms ofthe active window on a display screen.

Graphic display is of course indispensable to the concept ofvirtual space, since connecting directly to the visual cortexremains for the moment the stuff of science fiction.However, screen real estate is not the only sort of spaceinvolved in hypermedia design. The space of the screen isdefined in terms of pixels and other coordinate systems. Itrepresents architectonic space. This is a regular,mathematically precise space in which absolute principlesapply: objects are uniquely identified, have exclusivelocations, and obey rules of geometry and perspective.Most important, the architectonic space of screen display isbinary. Absence and presence, or the relation of figure toground, are immediately and unambiguously apparenlt. Apixel is either devoted to a graphic element or it is not.Elements may overlap or include each other, but in doingso they complicate the visual field, introducing anappearance of depth. Space on the screen imitates space inthe physical world, where architecture involvesmanipulating stable objects according to regular principles.Along with this space of geometries and built structures,

hypermedia also requires us to operate in a very different

environment. We might call this semantic space, since itis deeply connected to the production of meaning,interpretation, and other activities involving symbols.Marshall and Shipman invoke architectonic space in thecontext of writing, semantic space emerges more clearly inthe act of reading or reception — though since hypertexttend to blur the roles of reader and writer, these distinctionscannot be absolute. To understand semantic space, we needto follow Novak’s prescription and imagine ourselveswithin an information environment: for instance, ahypertext. Hypermedia researchers have expressedconsiderable concern that higher levels of complexity ininteractive documents might impair effectivecommunication (see Raskin [29], Lesk [191, Carlson [51,Wright [35], Charney [6], and for a critical review, Dillonet al. [8]). These reservations are justified. Hypermediatexts are undeniably more complicated and cognitivelychallenging than conventional texts. As some studies haveindicated (Egan et al. [9], Jones and Spiro [14]), thiscomplexity may have positive value for interpretive tasksrequiring multiple frames of reference. But encounteringthese multiple frames, or heterogeneous streams oflanguage, always invites confusion. The conceptual orsemantic space of a hyperdocument is inevitably morecomplicated than that of a comparable linear text.

The inherent complexity of hypermedia structures becomesmost apparent in navigation, or selecting and activatinglinks. Even in a highly structured, task-specific applicationsuch as technical documentation or argumentation,following a link brings ambiguity and uncertainty into thereading process. If the author has given adequate attentionto Landow’s rhetoric of arrivals and departures [16], or whatDillon et al. call “landmarks” [8], then we shouldunderstand clearly enough where we are going and why. In

terms of local linearity, a hypermedia text need be no morechallenging than a print document. But this does notneutralize the complexity of hypermedia reception. Even aswe follow the locally linear track we have selected, we

become aware that we could have chosen otherwise. Everylink we follow must be accompanied by a number ofalternatives. Some of these may have been explicitlysignaled by the system, as in NCSA Mosaic, where linkanchors are underlined and color-coded to show whether wehave already visited their destinations. Other alternatives,which Moulthrop [22] has called “implicit” destinations, arepresent to the reader as speculations or mental projections.In this regard, hyperdocuments invite what Hofstadter [13]calls “jumps outside the system.”

The concept of space used in this account of navigationshares little of the clarity and unambiguousness ofarchitectonic space. As Dillon et al. [8] see it, “we cannotnavigate semantic space, at least not the way we navigatephysical environments, we can only navigate the physicalinstantiation that we develop of the semantic space” (p187). Harpold [12] discusses hypertextual linking as“detour,” not a definitive trajectory from departure point toarrival point, but an elliptical and fundamentally uncertaindisplacement. The hypertextual detour, he says, “is a turn

ECHT ’94 Proceedings 207 September 1994

around a place you never get to, where something dropsaway between the multiple paths you might follow” (p173). Perhaps that which “drops away” as we traverse ahypermedia link is indeed our orientation in architectonicspace, with its stable geometries and singularity ofexpression. In the directed graph we can see clearly wherewe have come from and where we are going; but this is notthe case in semantic space. For every pint of actual arrivalin a hypermedia text, there are an unspecified number ofplaces we never get to, alternative destinations which the

system has either disclosed to us, and which we have chosennot to visit, or which may be simply undeveloped orunexpressed in the current version of the text. Semanticspaces are n-dimensional, as Dillon et al. [8] point out,while architectonic spaces have at best three dimensions,and more usually two. Architectonic space is always eitherempty or filled (see Benedikt [1]). In the absence of anobject, the space is empty — or in any event, that is whatthe assignment of a null or neutral value to a pixel is takento represent. In semantic space, however, the defaultcondition is not defini~ely empty but rather indefinitely

filled. A semantic space is a domain of possibleexpression. It is “semantic” because it is the place wheremeanings or interpretations come into existence; and inhypertext, as the cognitive critics cited earlier insist, there isalways a surplus of meaning.

3 “ ONTOLOGICAL DESIGN”

These speculations appear to move hypermedia design a bitout of its usual conceptual range. Figures like Harpold(along with Bolter, Landow, and the authors of this paper)come to hypermedia through Landow’s controversial“convergence” between information technology and literarytheory [17]. Though all of us also develop hypermediasystems and documents, we became interested inhypermedia partly because of its similarity to certainnotions associated with the poststructuralist critique of printand its cultural institutions. Harpolds concept of “detour,”for instance, draws upon ideas developed by thepsychologist Lacan and the philosopher Derrida. Theseconcepts may have more apparent bearing on philosophyand literature than on the design of information systems;but appearances can be deceiving. It is possible toimplement an architectonic space by means of graphicrepresentations, but as Winograd and Flores observe [33],semantic spaces are notoriously hard to capture in software.Yet one kind of space inevitably implies the other. Inanalyzing strategic behavior, we may speak of “decisionspaces,” implying the mathematical formalism (orarchitectonic space) of a specific formula. As Winograd andFlores point out, these spaces never encompass the fullrange of options available to an agent in the real world.Architectonic constructs like decision spaces and hypertextwebs inevitably imply other, less formally tractablestructure in semantic space.

In Winograd and Flores’s theory of design, this coupling ofarchitectonic and semantic spaces manifests itself as“breakdown,” or the moment when the constraints of a

particular formal system become apparent by juxtapositionwith their alternatives, As a defining instance ofbreakdown, they cite ELIZA’S infamous response to anexasperated subject who announces, “I am swallowingpoison.” The program replies, “How long have you beenswallowing poison?” (p 121). Assuming we are notdealing with a Hannibal Letter, we would not expect anyhuman psychiatrist to give such a response. ELIZASimitation breaks down at this point because we areconfronted with a case that falls outside its linguisticparameters. We have dropped out of architectonic space (thespace of rule-bound knowledge representation) into the morechaotic space of language and cultural assumptions.

Winograd and Flores base their “new foundation for design”on the proposition that advanced information systems willalways be susceptible to such awkward transitions — muchin the same way that hypermedia texts must inevitably carryheavier cognitive overhead than print or video. Given thislimitation, designers might be better off abandoning ahopeless struggle; or in other words, since we can’t get ridof the bug, why not call it a feature? We can accomplishthis conversion, Winograd and Flores suggest, byembracing a new concept of “ontological” design. “Newdesign can be created and implemented,” they note, “only inthe space that emerges in the recurrent structure ofbreakdown. A design constitutes an interpretation ofbreakdown and a committed attempt to anticipate futurebreakdowns” (p 78). It is this latter element, committedanticipation, which most often seems missing fromsoftware design — even in the field of interactive media,where it might arguably be prominent. Much attention hasbeen given to minimizing complexity and cognitivedemands in hypermedia (see especially Charney [6]). Thesecritiques envision better forms of architectonic space,whether in the form of clearer graphical representations ormore coherent rhetorics of arrival and departure. Suchefforts are undeniably important — architectonic space isthe only kind of space we can directly manipulate, after all— but from the perspective of ontological design, they areat least potentially misguided. Any hypermedia documentis extended simultaneously in both architectonic andsemantic space. It occupies both a domain of function (thereliable, connective operation of the link) and one ofbreakdown (the susceptibility of the link to “detour” orambiguity). Accordingly, we might turn our efforts towarddesigning structures that integrate or mediate between thesetwo varieties of space.

4 DESIGN EXERCISES

We will need to re-think our conception of space inhypermedia, and by extension, the dominant metaphor of“navigation” that we use to describe transactions within it.Such re-thinking can be fully realized only in viable systemdesigns and texts; but it might begin with a few conceptualexperiments, two of which we offer here. Our general aimin these experiments is to describe architectonic structureswhich, though still engaged in precise graphical mapping,are better adapted to the multiplicity of semantic space.

E(2HT ’94 Proceedings 208 September 1994

This exercise was inspired by a case of technical breakdown.Several years ago, one of the present authors produced afairly large literary hypertext in Story space. The

Storyspace stand-alone reader in its then current versionrequired that both the code for the reader and the entirehypertext, including all nodes and links, be loaded intoRAM at launch. The combined textual material andprogram amounted to about 1.2 megabytes, a memorydemand large enough to put the product beyond the reach ofconsumers with older, less capacious computers. Inresponse to the author’s request, Bolter designed an

experimental Storyspace reader which loaded only part ofthe hypertext into RAM. This reader module maintained a500-kilobyte memory store, into and out of which it wouldrotate sections of the text as necessary. The author tried tomake these sections as large and as locally coherent aspossible in order to minimize memory shuffling.

Though the new reader proved reliable, constraints ofhardware and software design prevented it from workingeffectively as a delivery system for narrative hypertext.Bolter and ~he author had anticipated that users wouldusually follow links to material within local sections,where “local” referred to proximity in the architectonicgraph of the Storyspace document. Early inforrnad testsstrongly refuted this assumption. About one time in four(roughly speaking) users chose links which carried them to

distant places in the document, “distant” again defined interms of the graph. These non-local excursions requireddisk access and memory operations, which even onrelatively swift machines added noticeably to the transitiontime between narrative places. These time lags made usersaware of a difference between “local” places, which could besummoned up instantly, and “distant” places encoded onsome remote sector of the drive platter.

The author of the hypertext in question found these effectsundesirable, so development of the new reader module wasabandoned. It is important to ask why the author made thisdecision. What made him assume that all transitions in ahyperdocument should be of equal and apparently nullduration? The “breakdown” in this case involved theauthor’s assumption that architectonic space (the space ofthe Storyspace graph) could adequately model semanticspace (the narrative dimensions of the story). Like allbreakdowns, this one is instructive. The author insisted oninstantaneous transitions because in his view, semanticspace has no fixed dimensions. Part of his narrative projectwas to create interconnections between scattered storyelements, connections which would overcome the sequentialseparation imposed by the printed page. To Speik notentirely fancifully, his design agenda involved creating astory that moved faster than the speed of books.

This last formulation suggests a useful though admittedlyfar-fetched analogy between the semantic space ofhypermedia and the interstellar spaces represented in sciencefiction. To imagine journeys beyond the solar system,science fiction writers have to invent methods of faster-than-light travel. These inventions often involvediscontinuities in the space-time continuum: wormholes,trans-dimensional portals, or the solution most familiar toaddicts of American TV, the “warp field effect,” in whichvessels bend or fold space around themselves. Fortunately,the physics of such inventions need not be considered here.(Someone please explain how the Star Trek universe avoidscatastrophic collapse with thousands of ships constantlystretching space-time all over the place. For furtherdiscussion of Star Trek’s implausible spaces, see Benedikt[1].) We are concerned here only with the analogy betweenRoddenberry’s famous “final frontier” and another imaginaryspace, the semantic domain in hypermedia. Both are in acrucial sense elastic — spaces in which dimensionalproperties like distance and contiguity can be easilyannulled.

If we follow this analogy, however, we must reconceive thefamiliar concept of navigation in virtuat space. Literallyspeaking, “navigation” refers to the operation of floatingvessels; but the spacecraft of Star Trek and other sciencefiction fantrsies are not much like ships. Since they moveby “distorting] the space-time continuum” [26], starshipsare not propelled through a medium as are aircraft or boats.Instead they warp or wrap space around themselves. If wethink of transitions in hypermedia as somehow analogousto this effect, then we need a new conceptual vocabulary —and most important, new representational schemes. Theobject of this design exercise is to create an architectonicrepresentation for “warp-effect” transitions in semanticspace shifts in textual location which we understand not interms of motion-toward (propulsion) but of gathering-in(warping).

Figure 1 illustrates a hypothetical variant of the Storyspacehypertext environment. On the tool palette, the standardnavigation tool has been replaced by a warp tool. This toolshows all link anchors in the current place with boxedoutlines or “wire wraps.” It also displays for each anchor an

overlaid field containing an integer value from 1 to 9,shown here in Roman numerals. These numbers may bethought of as warp coefficients. They represent, in ageneral sense at least, the amount of gathering or distortionthat must be applied to the semantic space in order to effectan instantaneous transition from the current place to thedestination point of the link anchor. In the illustration,material about “coupling and decoupling” lies relativelyclose to the current place and may be reached with a warpfactor of 2. The explanation of “apparent mass reduction,”on the other hand, exists in some remote textual quadrantand requires a warp-8 transaction to bring it in view.

ECHT ’94 Proceedings 209 September 1994

!x3w!:!_ “ ~~ Warp Field Theory ~~

❑ 4 The cumulative effect of the force applied drives the Q

* We~$eld

vehicle forward and is known as a

“ ‘%

peristaltic field manipulation (APFM),*;cl-:”’ coils in the engine nacelles are energized in

/sequential order, fore to aft, The firing frequency

❑determines the number of field layers, a greaternumber of layers per unit time being required athigher warp factors. Each new field layer expandsoutward from the nace eriences a rapid forcecoupling and decouplin riable distances fromth e nacelles, simultaneously transferring energy andseparating from the previous layer at velocitiesbetween 0,SCand 0,9c, This is well within thebounds of traditional physics, effectivelycircumventing the limits of General, Special, andTransformational Relativity, During force couplingthe radiated ener s the necessary transitioninto subspace, app arent mass reductioneffect to the spacecraft, lh is facilitates the shppageof the spacecraft through the sequencing layers of ~warp field energy, a

Figure 1: Storyspace, The Final Frontier (with apologies to Bolter, Joyce, and Smith; textfrom Sternbach and Okuda).

If fantasy physics are not to your taste, it is possible toregard this proposal simply as a weighting scheme forhypertext links, similar to that incorporated inSchneiderman’s HyperTIES system [30]. But though notespecially original, this design solution nonetheless raisestwo important questions: how are the weighings

generated, and what do they represent? Since this schemefunctions in architectonic space, we might generate the“warp coefficient” by referring to the structure graph, as inMarshall and Shipman’s work [21] or Bernstein’s “LinkApprentice” [2]. That is, the warp number might simplyrepresent the amount of screen space, or the number ofnodes and links, that intervene between one place andanother. One might object, then, that this schemeemphasizes architectonic certainty over the n-dimensionalityof semantic space.

To some extent this must be true the scheme of weightedlinks must break down at some point, such as in the case ofsmall differences between weighings. In terms ofmeaning, what distinguishes a warp-2 link and a warp-3link? Moreover, what would become of warp geometries ina dynamically evolving, multi-user hypertext application?Why should a deformable information space be mapped inthe same way twice? Fortunately from one point of view,though unfortunately for those expecting precision, the

metaphor of folded semantic space renders suchmathematical difficulties moot. The coefficients proposedhere are not precise, digital measurements, but ratherevocative descriptions of a continuous, flexible, analogspace. The metaphor of warped space, referring as it does toan imaginmy enterprise (or Enterprise), reminds us that thegraphically generated link weighings are partly metaphoric,one might even say science-fictional. They thus allow usto mediate between architectonic and semantic spaces. Ifthese coefficients have any utility in actual applications, itmust be as a rough indicator rather than as a strict metric;still, their theoretical significance seems less ambiguous.

Exe rcise 2: A Map of Deto~

Design Exercise 1 reveals that proximity has twomeanings, only one of which is usually mapped. Thus, insystems like SEPIA, Story space, Aquanet, NoteCards, andVNS, where both global and local maps emerge as usersinscribe nodes and links, the maps represent chieflygeometric or architectonic notions of proximity. AlthoughStoryspace offers writers and readers tke representations orvisualizations of the emerging hypertextual structure —story space view, chart view, and outline view, all threerepresentations in fact map only the architectonic dimensionof the hypertextual space, All three occlude the semantics

E(2HT ’94 Proceedings 210 September 1994

of cross-hierarchical linkages, privileging whuteversemantics the geometries can construct and maintain at theexpense of the semantic density and multiplicity oflinguistic connections.

Fisheye views represent schemes for mapping boih theglobal architecture of the hyperdocument and one or more~gional “contents,” what Noik [24] terms the local detail inan area of interest. Noik conceptualizes the representationalproblem in terms of links from a user’s current location thatlead to a geometrically distant “region” of the document.The system is designed to visualize three aspects of thetotality simultaneously: the architecture of the entirehyperdocument, displayed with a high level of abstraction,the current area of interest, displayed with a high level ofenlarged detail, and one or more other (related) areas ofinterest, displayed with more detail than the globalrepresentation but with less detail than the current focalpoint.

The goal of Noik’s fisheye representation is to distort thevisual elements of the global map so as to foreground or

enlarge the target regions, the areas of interest the reader hasidentified. In developing this scheme, Noik identifies twoimportant anchors: a current position or focal point (IT)and a desire or query intended to explore possible “next”locations. Possible “next” regions can have variableDegrees of Interest (DOI) which influence the size of their

graphic representation. This scheme recognizes that“nextness” or semantic relatedness need not be confined

either to geometrically proximate spaces or to thoseexplicitly linked: the concepturd relations can be expressedas collocations or gatherings of nodes according to someother userdefined pmpcrty.

The problem with Noik’s scheme for maintaining a global(architectonic) perspective while users actually attend tolocal detail is that the global is already well expressed in ahierarchical table of contents. That is, any document sofully structured as the two to which Noik refers — a FreeTrade Agreement and an extensive dictionary — can berepresented for readers’ purposes as an outline. Forexample, Bolter’s hypertext Writing Space offers usersseveral representations of the document’s structure. Figure2 shows a geometric map (nested boxes) and Figure 3 aconcepturd hierarchy (essentially a dynamic table ofcontents). Neither represents the semantic space of the tex~the text a user encounters when she reads the documentthrough the EasyReader interface, where the links representmultiple opportunities, multiple proximities, only some ofwhich are “near” the current location in architectonic terms.As long as some sort of table of contents is alwaysavailable to a user of such a highly structured document, agraphical representation of that structure serves chiefly as anorienting landmark for “users who set out to understand theoverall structure of a large hypertext” (Noik [24], p 192,emphasis added), where “structure” seems to mean the nestedhierarchy or architectonic space.

~ File Edit Write Uiew Features Windows Colors ❑ I

+%*

mIl. Conceptua

d

&

m+

monc us..

m

Figure 2: Storyspace view (geometric map) of a portion of Bolter’s Writing Space,

ECHT ’94 Proceedings 211 September 1994

~ File Edit Write Uiew Features Windows Colors ❑13 The book pj:

HThe book

❑ Preface”

❑ Introduction”

❑ 1.Visual Space”

IZ!l2. Computer”

❑ 3. Writing as Technology-

E 4. The Elements of”

E 5. Seeing and Writing-

I!ll I 1.Conceptual”

!!!!6. The Electronic Book”

S7. The New Dialogue-

❑ 8. Interactive Fiction-

❑9. Electronic Writing”

Rlll. Mind-

❑ 10. Artificial-

❑ 11. Electronic Signs”

❑ 12. Writing the Mind”

❑ 13. Writing Culture”

❑ 14. Conclusion-

Figure 3: Outline view (conceptual hierarchy) for the same portion of Writing Space.

Instead of pursuing the double representation of Noik’s One way to represent the relationship between “now” (IT)

svstem. we could sutmose that an auurouriate representation and “next” (POI) is to display information about the. . . .of the semantic space need be onl~ &l. The design issuewould then be, what constitutes locality: where can weusefully draw the boundaries of the neighborhood? Clearlythe current location remains, in Noik’s terminology, thefocal point. Yet, the focal point of a user’s experience isnecessarily unstable it shifts each time a user “moves” toa new location or warps the document space around her.What the user may need to know at each transition may notbe where he or she “is” in relation to a geometry of thewhole document but where he or she stands relative to someset of potential next focal points. That set constitutes theAevant Points of Interest (KM).

destinations accessible from the current location. But sucha limited representation of locality reduces every link’sfunction to successive replacement — what one mass mediacritic derisively calls the grammar of “Now... this” [28].Our previous thought experiment constructs just such alimited grammar. Each link anchor points to a potentialPOI while the “warp coefficient” signifies the relationsbetween the architectonic structure of the document and asemantic interest the user might pursue. However,

displaying a set of possible next nodes fails to tell the uservery much about the trajectory or contours of her reading[3]. A more helpful view might situate the user in relation

ECHT ’94 Proceedings 212 September 1994

to all POI as well as the set of Subsequent Points of through the current set POI. The resulting depiction,Interest (SPOI). SPOI are all those nodes accessible Figure 4, would look much like a familiar tree diagram.

POI

El1

D2

D3

c14

D5

D6

07

D8

D9

Figure 4: The node “War Zones’’from Victory Garden as FP, with representation of its POI and SPOI.

The head of the tree, the FP, leads to a set POI whlich in we collapse redundancies and re-introduce link lines between

turn yields a set SPOI. However, this representation planes. The resulting graphic, Figure 5, makes available to

obscures any potential redundancy. It is difficult to notice the user a representation of potential elements for

that the node marked Gl, a descendant from the node marked exploration in a more liberaIly defined neighborhood.7, also appears as a descendant from the node madked 2.The value of this representation might be enhanced, then, if

ECHT ’94 Proceedings 213 September 1994

Fl?

/

.\9,,#\ ‘, ,,

\ . . /

L’

Figure 5: Mapping the neighborhood of the node “War Zones” in Victory Garden.

Though it remains an architectonic depiction, Figure 5constructs proximity in terms of semantic relation to thecurrent focal point (FP). Moreover, it displays trajectories,some of which may overlap at a later point in the user’sinteractions. For example, a user choosing the member ofset POI marked 5 could subsequently choose the member ofset SPOI marked Il. Alternatively, the user could reach 11through the members 1 or 3 of the set POI.

More important, perhaps, the graphic information displaysthose avenues foreclosed by any given choice. It showsthat if the user chooses the member of the set POI labeled3, she will then be able to reach the unique members of theset SPOI descendant from 3. But she will not then see theunique members of the set SPOI that descend from thenodes marked 8, 6, 1, and so forth, at least not withoutshifting her current focal point. By declining to display thetotality of the document and by suppressing representationsof all nodes in the system in favor of only proximate andnearly proximate semantic regions, this strategy attends to

the succession of Harpoldian detours that define the user’sinteractions with the text. It might thus be regarded asanother means of mediating between screen real estate andconceptual space.

Yet this mapping scheme is subject to breakdown just likeany other. The map shown in Figure 5 may represent thecomplexity of textual relations within a document. It mayeven suggest how these relations are, in Nelson’s words,“deeply intertwingled.” However, it cannot reliably depictHarpolcl’s “place you never get to,” since in at least somecases, that place may exist only as a hypothetical attemativein the mind of the user. Like the weighted linking schemeof the first exercise, this relativistic map can suggest thedimensions of semantic space, but it cannot exhaustivelyrepresent them. Semantic space thus constitutes aninevitable limiting factor for any architectonicrepresentation

ECHT ’94 Proceedings 214 September 1994

5 CONCLUSION: DESIGN LIMITS

Both our design exercises invoke a more problematicdefinition of space than is usually applied in hypermedia.This is in line with the general emphasis on complexspatiality in virtualizing information systems. Novalk [26]describes design principles for virtual reality as “liquidarchitecture.” In our notion of semantic space, we havetried to develop a similar idea for hypermedia. Like Novak,we believe that users of virtual texts must be able to situatethemselves within a dynamic information space; hence ourinterest in warping rather than navigation, and in relativisticlocal maps instead of global representations. But unlikesome proponents of all-encompassing or “seamless”artificial environments (for instance LauA [18]), we believethat any attempt to represent this internal situation instable, objective terms must inevitably reach a point ofobvious constraint. If our designs are to reflect anintelligent anticipation of such breakdowns, we mustunderstand that the two domains of virtual space, thearchitectonic space of mapping and the semantic space ofconceptual development, do not perfectly correspond.

Having anived at a very similar conclusion, Dillon et al.suggest that we must refine the conceptual apparatus we useto describe information spaces: “Our view is that theprecise nature of the representation is less important toworkers in the field of interactive technology than theinsights any theory or model of navigation provides. Tothis extent we propose that a model based on schema theoryand including landmarks, routes and surveys asinstantiation of basic knowledge is of some utility inconsidering the design of electronic information spaces” ([8]p 175). This conclusion represents a pragmatic, positivisticapproach to what Norman [25] calls the “Gulf of Execution”— the inherent mismatch between architectonic andsemantic spaces. As Norman argues more generally., suchan approach gives rise to an immensely valuable enterprise,the field of cognitive engineering, with its emphasis onuser-centered system design. Though we do not questionthe importance of this approach for the development ofinteractive technologies, we suggest that cognitiveengineers may benefit fmm the critical observations of theirco-workers in ontological design. Dillon, McKnighl, andRichardson are no doubt correct in their call for lbetterspatial metrics and metaphors for hypermedia. In their verylimited ways, our design exercises respond to this call. Butthe obvious limitations of our exercises demonstrate thatour pragmatism functions only within clear constraints.Semantic and architectonic spaces cannot be perfectlyreconciled. We should aim for systems that harmonize thetwo as well as possible, but which acknowledge thecontingent nature of any such harmony.

REFERENCES

[1] Benedikt, M. “Cyberspace: Some Proposals”, inM. Benedikt (cd.), Cyberspace: First Steps, pp.119-224, MIT Press, 1991.

[2]

[3]

[4]

[a

[a

[7]

[8]

[9]

[10]

[11]

[12]

[13]

Bernstein, M., J. Bolter, M., and E. Mylonas..“Architectures for Volatile Hypertext”, Hypertext’91 Proceedings, ACM Hypertext Conference (SanAntonio TX) pp. 243-260, December, 1991.

Bernstein, M., M. Joyce, and D. Levine.

“Contours of Constructive Hypertext”,.Proceedings of the ACM Hypertext Conference,Second European Conference on Hypertext(Milano) pp. 161-170, November 1992.

Bolter, J. Writing Space: The Computer,Hypertext, and the History of Writing, LawrenceErlbaum Associates, 1991.

Carlson, P. “The rhetoric of hypertext”,Hypermedia 2,2, pp. 109-132, 1990.

Charney, D. “The Effect of Hypertext on Processesof Reading”, in C. Selfe and S. Hilligoss (eds.),Literacy and Computers: The Complications ofTeaching and Learning with Technology, pp. 238-263, Modem Language Association, 1994.

DeRose, S. “Expanding the Notion of Links”,.Hypertext ’89 Proceedings, ACM HypertextConference (Pittsburgh PA), pp. 249-258,November 1991.

Dillon, A., C. McKnight, and J. Richardson..

“Space — The Final Chapter or Why PhysicalRepresentations are not Semantic Intentions”, in

C. McKnight, A. Dillon, and J. Richardson (eds.),Hypertext: A Psychological Perspective, pp. 169-91, Ellis Horwood, 1993.

Egan, D., M. Lesk, R. Ketchum, C. Lochbaum,J.. Remde, M. Littman, and T. Landauer.“Hypertext for the Electronic Library? CORESample Results”, Hypertext ’91 Proceedings,ACM Hypertext Conference (San Antonio TX) pp.299-312, December 1991.

Gloor, P. “CYBERMAP — Yet Another Way ofNavigating in Hyperspace”, Hyper[exf ’91Proceedings, ACM Hypertext Conference (SanAntonio TX), pp. 107-21, December 1991.

Halasz, F. “Reflections on NoteCards: SevenIssues for the Next Generation of HypermediaSystems”, Communications of the ACM, 31, 7,pp. 836-855, 1988.

Harpold, T. “Threnody: PsychoanalyticDigressions on the Subject of HypertextS”, in P.Delany and G. Landow (eds.), Hypermedia andLiterary Studies, pp. 171-184, MIT Press, 1991.

Ho fstadter, D, Godel, Escher, Bach: An EternalGolden Braid. Basic Books, New York, 1979.

ECHT ’94 Proceedings 215 September 1994

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

[22]

[23]

[24]

Jones, R. and R. Spiro. “Imagined ConversationsThe Relevance of Hypertext, Pragmatism, andCognitive Flexibility Theory to the Interpretationof ‘Classic texts’ in Intellectual History”,Proceedings of the ACM Hypertext Conference,Second European Conference on Hypertext(Milano) pp. 141-148, November 1992.

Lai, P. and U. Manber. “Flying throughHypertext”, Hypertext ’91 Proceedings, ACMHypertext Conference (San Antonio TX), pp. 123-132, December 1991.

Landow, G. “RelationalIy Encoded Links and theRhetoric of Hypertext”, Hypertext ’87 Papers,ACM Hypertext Conference (Chapel Hill NC), pp.331-344, November 1987.

Landow, G. Hypertext: The Convergence ofContemporary Critical Theory and Technology,Johns Hopkins University Press, 1992.

Laurel, B. “Interface as Mimesis”, in D. Normanand S. Draper (eds.), User Centered SystemDesign: New Perspectives on Human-ComputerInteraction, pp. 67-86, Lawrence Erlbaum, 1986.

Lesk, M. “What to do When There’s Too Muchinformation”, Hypertext ’89 Proceedings, ACMHypertext Confemece (Pittsburgh PA), pp. 305-318, November 1989.

Marshall, C., F. Halasz, R. Rogers, and W..Janssen, Jr. “Aquanec A Hypertext Tool to HoldYour Knowledge in Place”, Hypertext ’91Proceedings, ACM Hypertext Conference (SanAntonio TX), pp. 261-275, December 1991.

Marshall, C. and F. Shipman III. “Searching forthe Missing Link Discovering Implicit Structurein Spatial Hypertext”, Hypertext ’93 Proceedings,ACM Hypertext Conference (Seattle WA), pp.217-230, November 1993.

Moulthrop, S. “Toward a Rhetoric of InformatingTexts”, Proceedings of the ACM HypertextConference, Second European Conference onHypertext (Milano), pp. 171-180, November1992.

Nanard, J. and M. Nanard. “Should Anchors BeTyped Too? An Experiment with MacWeb”,Hypertex( ’93 Proceedings, ACM HypertextConference (Seattle WA), pp. 51-62, November1993.

Noik, E. “Exploring Large Hyperdocuments:

[25]

[26]

[27]

[28]

[29]

[30]

[31]

[32]

[33]

[341

[35]

Norman, D. “Cognitive Engineering”, in D.Norman and S. Draper (eds.), User CenteredSystem Design: New Perspectives on Human-Computer Interaction, pp. 31-61, LawrenceErlbaum, 1986.

Novak, M. “Liquid Architectures in Cyberspace”,in Michael Benedikt (cd.), Cyberspace.’ First Steps,pp. 225-254, MIT Press, 1991.

Parunak, H. “Don’t Link Me In: Set-BasedHypermedia for Taxonomic Reasoning”, Hypertext’91 Proceedings, ACM Hypertext Conference (SanAntonio TX), pp. 233-242, December 1991.

Postman, N. Amusing Ourselves to Death:Political Discourse in the Age of Show Business,Penguin, 1985.

Raskin, J. “The Hype in Hypertext A Critique”,Hypertext ’87 Papers. ACM Hypertext Conference(Chapel Hill NC), pp. 325-330, November 1987.

Schneiderman, B. “User Interface Design for theHyperTIES Electronic Encyclopedia”, Hypertext’87 Papers. ACM Hypertext Conference (ChapelHill NC), pp. 189-94, November 1987.

Stembach, R. and M. Okuda. Star Trek: The NextGeneration Technical Manual. Pocket Books,1991.

Stotts, P., R. Furuta, and J, Ruiz.“Hyperdocuments as Automata: Trace-BasedBrowsing Property Verification”, Proceedings ofthe ACM Conference on Hypertext, SecondEuropean Conference on Hypertext (Milano), pp.272-281, November 1992.

Winograd, T. and F. Flores. UnderstandingComputers and Cognition: A New Foundation forDesign. Addison Wesley, 1985.

Wexelblat, A. “Giving Meaning to SemanticSpaces”, in M. Benedikt (cd.), Cyberspace: FirstSteps, pp. 255-272, MIT Press, 1991.

Wright, P. “Cognitive Overheads and Prostheses:Some Issues in Evaluating Hypertext”, Hypertext’91 Proceedings, ACM Hypertext Conference (SanAntonio TX), pp. 1-12, December 1991.

Fisheye Views of Nested Networks”, Hypertext ’93Proceedings, ACM Hypertext Conference (SeattleWA), pp. 192-205, November 1993.

ECHT ’94 Proceedings 216 September 1994