Reprinted from Proceedings of Hypertext '98, Frank Shipman, Elli Mylonas, and Kaj Groenback, eds, ACM, New

York. (c) Copyright 1998 by Association for Computing Machinery. All Rights Reserved. Reprinted by permission. Copyright 1999 by the Association for Computing Machinery, Inc. Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on servers, or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from Publications Dept, ACM Inc., fax +1 (212) 869-0481, or permissions@acm.org.

PATTERNS OF HYPERTEXT

Mark Bernstein chief scientist, Eastgate Systems, Inc.

Patterns Of Hypertext
The complexity and unruliness of the complex webs of links we create has fre uently led to calls for !structured! or otherwise disciplined hypertext "##$"%&$"'($. )hile calls for clearer structure ha*e tried to a*oid, consolidate, or minimi+e links, it is now clear that hypertext cannot easily turn its back on complex link structures. )here it was once feared that the cogniti*e burdens of large, irregular link networks would o*erwhelm readers, we find in practice that myriad casual readers flock to the docu*erse. The growth of literary and scholarly hypertext, the e*olution of the )eb, and the economics of link exchange all assure the long,term importance of links. Since large linked constructs cannot be wished away, it is time to de*elop a *ocabulary of concepts and structures that will let us understand the way today-s hypertexts and

)eb sites work. .rogress in the craft of writing depends, in part, on analysis and discussion of the best existing work. /n appropriate *ocabulary will allow us both to discern and to discuss patterns in hypertexts that may otherwise seem an impenetrable tangle or arbitrary morass. The reader-s experience of many complex hypertexts is not one of chaotic disorder, e*en though we cannot yet describe that structure concisely0 the problem is not that the hypertexts lack structure but rather that we lack words to describe it.

Looking for Patterns

This paper describes a *ariety of patterns of linkage obser*ed in actual hypertexts. 1ypertext structure does not reside exclusi*ely in the topology of links nor in the language of indi*idual nodes, and so we must work toward a pattern language through both topological and rhetorical obser*ation. Instances of these patterns typically range in scope from a handful of nodes and links to a few hundred. These patterns "%2$"#$ are components obser*ed within hypertexts, rather than system facilities 3see "4'$5 or plans of a complete work. Typical hypertexts contain instances of many different patterns, and often a single node or link may participate in se*eral intersecting structures. I do not argue that the obser*ed structural patterns are uni uely desirable, that superior patterns cannot be de*ised, or indeed that the writers of these hypertexts meant to use these patterns at all. I do propose that by considering these patterns, or patterns like them, writers and editors may be led to more thoughtful, systematic, and sophisticated designs. These patterns are offered, then, as a step toward de*eloping a richer *ocabulary of hypertext structure. Examples are drawn from published stand,alone hypertexts as well as from the )eb. )eb sites are readily accessible but *olatile6 a site which today illustrates one structure may be unrecogni+able tomorrow. .ublished hypertexts are less accessible, but are also more permanent. Moreo*er, some important patterns depend on dynamic links ,, links which depend on the reader-s past interactions. The )eb itself is state,free, and while *arious implementations of state, dependent beha*iors for the )eb ha*e been proposed, state,dependent beha*ior remains an exceptional case in )eb hypertexts. Some pattern examples are drawn from literary fiction. I do not belie*e these patterns to be useful exclusi*ely for fiction0 rather, a *ariety of economic and cultural factors sometimes encourage experimentation in narrati*e rather than technical writing or 7ournalism. Moreo*er, hypertext fiction tends to be written for general audiences and may remain a*ailable indefinitely, while speciali+ed reference manuals and 1elp systems may be short,li*ed and less readily a*ailable to the general reader. 8or does our interest in structural *ocabulary necessarily imply a structuralist or post,structuralist stance0 we need to describe phenomena, whate*er our theoretical beliefs "9:$";$. Two patterns ,, Tree and Se uence ,, ha*e been described many times in the hypertext literature ";4$"49$. Both are useful, indeed indispensable, and can be found in almost any hypertext.

Cycle
In the Cycle, the reader returns to a pre*iously,*isited node and e*entually departs along a new path. <ycles create recurrence ";%$ and so express the presence of structure.

=olb-s Socrates In The Labyrinth "9($ discusses the role of the <ycle in argumentation, showing how hypertext cycles emerge naturally from traditional argumentati*e forms. <yclical repetition also modulates the experience of the hypertext "99$, emphasi+ing key points while relegating others to the background. )riters may break a cycle automatically by using conditional links, or may use breadcrumbs "'$ to guide the user to depart along a new tra7ectory. >elying on breadcrumbs to break cycles is common on the )eb. In oyce!s Cycle, the reader re7oins a pre*iously,*isited part of the hypertext and continues along a pre*iously,tra*ersed tra7ectory through one or more spaces before the cycle is broken. >e*isiting a pre*iously,*isited scene, moreo*er, may itself pro*ide a fresh experience because the new context can change the meaning of a passage e*en though the words remain the same. The opening lines of afternoon, a story "#:$, when first seen, establish a chilly climate, poetic and o*erwrought6 By fi*e the sun sets and the afternoon melt free+es again across the blacktop into crystal octopi and palms of ice,, ri*ers and continents beset by fear, and we walk out to the car, the snow moaning beneath our boots... ?ater, we may again encounter the same scene. 8o longer does it ser*e as an establishing frame0 later, we may recogni+e that the winter scene the narrator describes might be the wreck of his ex,wife-s car, that the continents of fear, the moaning snow, may be the wrack left after the car 3and the bodies5 ha*e been remo*ed. 1ypertext, @oyce writes elsewhere, demands rereading "#2$. Measured and planned repetition can reinforce the writer-s message6 end,of,chapter summaries and ballad refrains, for example, are a common feature of the pedagogical literature of print and oral culture. <ycles thus lend themsel*es not only to a *ariety of postmodern effects "4;$, but also to familiar writerly motifs6 Af recursus, there is hallucination, de7a *u, compulsion, riff, ripple, canon, isobar, daydream, and theme and *ariation...Af timeshift there is the death of Mrs. >amsay and the near disintegration of the house...?eopold Bloom on a walk, and a man who wants to say he may ha*e seen his son die. Af the renewal there is e*ery story not listed pre*iously. "#2$ In "o#glas!s Cycle "%#$, the appearance of an unbroken cycle signals closure, the end of a section or the exhaustion of the hypertext. / $e% Ring is a grand cycle, a cycle that links entire hypertexts in a tour of a sub7ect. 1ypertexts in a )eb ring agree, in essence, to share readers. Though largely unheralded in the research literature, )eb rings, <.>.E.). and related compacts ha*e pro*ed central to the hypertext economy. 1ypertexts concerning speciali+ed interests ,, obscure actors, or )orld )ar I memoirs ,, may promise little direct professional or commercial importance, and alone they cannot easily find an audience. <ooperation among related sites, howe*er, creates self,organi+ing +ones of autonomous but interrelated acti*ities on a common theme or toward a common goal. The cyclical structure of )eb rings tends to promote e uality of access6 each participant gains one inbound link, at the cost of offering one outbound link. /lternati*e structures 3such as central directories and search engines5 can also offer access, but the cyclical structure of the ring keeps each

participant e ual and resists the tendency to concentrate attention at the directories themsel*es. / conto#r ";%$"9&$ is formed where cycles impinge on each other, allowing free mo*ement within and between the paths defined by each cycle. Mo*ement among the cycles of a contour is easy, and infre uent links allow more restricted mo*ement from one contour to another.

Co#nterpoint
In Co#nterpoint, two *oices alternate, interlea*ing themes or welding together theme and response. <ounterpoint often gi*es a clear sense of structure, a resonance of call and response reminiscent at once of liturgy and of casual dialogue. <ounterpoint fre uently arises naturally from character,centric narrati*es0 for example, Forward Anywhere "(9$ uses a series of e,mail letters between its two central characters to explore their differences and establish their connections. <ounterpoint may be fine,grained. In Bubbe's Back Porch, /bbe Bon-s Bubbe mo*es constantly between tales of the distant past and tales of her own present, telling her great,granddaughter at once what it is like to be old and what it was like, long ago, to be a young @ew in old >ussia "%;$. Bon mo*es between times and *oices within a single lexia, echoing the patterns of traditional Ciddish storytelling 3see, for example, the work of Sholem /leichem "%$5. It is also interesting to obser*e how the same counterpoint techni ues can be adapted to decenter the sub7ect"%:$, for here 3as in Spiegelman-s Maus "';$5 traditional narrati*e techni ues yield postmodern effects. /t a large scale, Bon-s hypertext is essentially linear, and the internal counterpoint 3and the Missing ?ink patterns suggested by recurrent anti ue photographs5 forms the chief hypertextual element. /drienne Eisen-s Six Sex Scenes "#9$, on the other hand, offers three or four outbound links from almost e*ery node. Eisen-s hypertext habitually alternates time frames6 a writing space describing a childhood scene tends to be linked to scenes of adult life, and adult scenes tend to be linked to stories of childhood. Because Eisen, in Six Sex Scenes, works hard to a*oid <ycles, the <ounterpoint of childhood and adult experience is its most prominent structural element. In !Interlocked! "(%$, Beena ?arsen addresses a topic closely allied to Eisen-s6 how memories of childhood or adolescence find expression in the sexuality of the adult protagonist. )here Eisen uses <ounterpoint as a substitute for the structural power of the <ycle pattern, ?arsen builds her hypertext from two interlocked <ycles. These cycles, inspired by a classic uilt pattern, represent self,reinforcing traumas of past and present. <oncurrently, links between cycles create a ! uilted! counterpoint that represents the interplay of memory and action0 the counterpoint, like uilt stitching, distorts the cycles while holding them in place.

Digure ;. In Samplers, a list of links connecting two writing spaces becomes an interstitial *oice in counterpoint to the main text. Interstitial counterpoint adds hypertext commentary notionally situated between writing spaces. Interstices ha*e long been used for uotation, both epigraphic and ironic ";#$. ?inks in ?arsen-s Samplers appear in a dialog box ,, a con*entional list of links that Storyspace authors can use to build an ad hoc multi,tailed link. The dialog is designed to be purely functional, showing a list of links by pathname and destination, but ?arsen has chosen path names so that this list itself can be read as an interstitial poem. Edward Dalco independently disco*ered the same, unexpected <ounterpoint opportunity in his hypertext poetry, !Sea Islands!, where the interstitial writing includes both path names and destination titles "%4$. <ounterpoint writ large, the dialogue amongst hypertexts proposed in !<on*ersations )ith Driends! ":$ is constructed as <ounterpoint among se*eral independent hypertexts, each representing a recogni+able point of *iew and each capable of responding to links and tra7ectories within its own frame and those of other acti*e hypertexts.

&irror$orl'
To retain coherence, writers of both texts and hypertexts fre uently adhere to a single *oice and point of *iew. &irror(orl's pro*ide a parallel or intertextual narrati*e that adopts a different *oice or contrasting perspecti*e. The Mirrorworld echoes a central theme or exposition, either amplifying it or elaborating it in ways impractical within the main thread. )here <ounterpoint interwea*es different *oices of e ual 3or nearly e ual5 weight within a single exposition, the Mirrorworld establishes a second *oice that separately parallels 3or parodies5 the main statement. 3 The term !Mirrorworld! is meant to allude to Throu h The Lookin !lass and to funhouse mirrors, not to Eelernter-s monograph "#%$.5 In "ncle Buddy's Phantom Funhouse "(4$, by @ohn McBaid, readers explore the computer files of the late /rthur 8ewkirk through his 1yper<ard home page, which is organi+ed in the image of his house. The back door, obscurely labeled !Egypt!, allows passage to 8ewkirk-s locked files0 these files, once the reader gains access, appear in a distorted image of the house, retitled !/untie Em-s 1aunt 1ouse!. In this haunted

funhouse, the content and concerns of 8ewkirk-s 1yper<ard house are mirrored in darker extensions and parodies. McBaid adds depth to the reader-s knowledge of 8ewkirk through this distorted addition to his life and work. The central thread of Edward Dalco-s A #ream with #emons is a no*el,within,the, no*el6 a story of a woman, her daughter, and her lo*er. wrapped together in lo*e and *iolence "%($. This narrati*e is interrupted periodically by na*igational opportunities that lead the reader into a basement of notes and memories, purportedly belonging the notional author. Dalco thus superimposes two layers of fiction6 the dramatic conflict of incest and abuse in the con*entional narrati*e is echoed by the more complex and ambiguous backstory of the Mirrorworld. The Mirrorworld also here plays an intriguing formal role6 by re*ealing the thoughts and moti*es of the story-s notional creator in a second fiction, the basement in*ites the reader to speculate on the nature of authorship more deeply than the familiar readerFwriter dichotomy"9;$ "%:$. =athryn <ramer-s !In Small G ?arge .ieces! ";:$ defies coherence in its central thread, which is told backward and which *eers unpredictably between a mundane s uabble between children and a horrific fantasy of the grand unified parent. Its Mirrorworld interlea*es brief and impressionistic sketches of interior life ,, perhaps of the protagonist at a later time, perhaps of the author as a younger woman. 1ere, the Mirrorworld is spare and fragmented, resisting e*en the retrograde coherence of the central thread0 without determined effort, the reader finds it difficult to remain in the Mirrorworld. The fragmentation reflects the shattered mirror of the work-s title0 the mirror world cannot be put together again, but sharp, sil*ery splinters are always underfoot.

Tangle
The Tangle confronts the reader with a *ariety of links without pro*iding sufficient clues to guide the reader-s choice. Tangles can be used purely for their *alue as intellectual amusement, but also appear in more serious roles. In particular, tangles can help intentionally disorient readers in order to make them more recepti*e to a new argument or an unexpected conclusion "(&$"2$. An entering a hypertext, a tangle can lead *isitors to different entry points, helping to con*ey the breadth of a hypertext to readers who may not anticipate the hypertext-s scope or co*erage. The home page of designer Ba*id Siegel "42$, for example, opens with four identical icons that lead to four different !home pages! ,, each offering a different design and a different emphasis. 8ew or infre uent *isitors must choose arbitrarily, and thus will likely see different parts of the site on each *isit "4:$. >eaders may, through habit or preconception, form an excessi*ely narrow *iew of a hypertext. Because tangles are difficult to fit into a simple, preconcei*ed structure, they encourage browsing and disco*ery. Tangles may extend through many writing spaces ";2$ or, like Siegel-s entryway, may be limited to a single Montage. Tangles are fre uently encountered near the beginning of a hypertext, where they disrupt orientation and create a sense of depth, but <arolyn Euyer-s $uibblin "#($ places a ma+e at the center of the hypertext, forming a bridge between scenes or episodes. Tangles may be used as pacing de*ices, or to recapitulate moments or pathways encountered earlier in a reading. Tangles are often found within or ad7acent to Mirrorworlds.

Moulthrop terms hypertexts robotic when the logic of the hypertext, not reader choice, tends to dictate the course of a reading "(:$. >obotic tangles like Mary,=im /rnold-s !?ust! "9$ combine complex dynamic structure, rich in broken cycles and other structural cues, with a dearth of interacti*e choice. This structure ser*es to entice the reader while frustrating the uest for release and resolution.

Sie)e
Sie)es sort readers through one or more layers of choice in order to direct them to sections or episodes. Sie*es are often trees, but may be multitrees, B/Es, or nearly, hierarchical graphs0 different topologies may all ser*e the same rhetorical function. )here the choice is informed and instrumental, sie*es become decision trees. The Cahoo directory, for example, pro*ides a large sie*e that readers tra*erse to find topical entry points to the )eb. Sie*es need not be represented as explicit hierarchies0 the 1ot Sauce M<D browser displays sie*es in three,dimensional space and permits readers to !fly! in Sem8et style"%9$ through the sie*e to their destinations, whereas the 1ypertext 1otel ";'$ hides its introductory Sie*e behind a check,in desk and hotel lobby.

&ontage
In &ontage, se*eral distinct writing spaces appear simultaneously, reinforcing each other while retaining their separate identities. Montage is most fre uently effected through superimposed windows which establish connections across the boundaries of explicit nodes and links. Montage is prominent in the pedagogical hypertexts of Eeorge .. ?andow "9', (;$, each of which commences with a montage offering multiple points of departure "92$. Shelley @ackson-s Patchwork !irl "#'$ also uses windowed montage with intriguing effect. <hristiane .aul-s "nreal %ity "4($ breaks the frame of the screen, using montage between the screen and a con*entional paper book, held in the *iewer-s hand. /n iconic representation of the printed page mediates the montage, thus freeing screen real,estate.

Digure %. / montage from .atchwork Eirl "#'$ as it appears in &ritin At The 'd e "(&$

Montage is a fact of life in the design of museums and art galleries, where disparate *isual works are collected in a limited space. Thoughtful architecture and cle*er arrangement may minimi+e the disrupti*e effects of montage, while 7uxtaposition may suggest new insights. Some art,historical hypertexts attempt to recreate the architectural montage of real or *irtual museum spaces0 often, as in the masterful (us)e d'*rsay+ ,isite ,irtuelle, the sub7ect of such a hypertext becomes the museum itself rather than its collections ";($. Trellis "'%$ is extensi*ely ,, perhaps primarily ,, concerned with describing and managing montage.

Neig*%or*oo'
/ Neig*%or*oo' establishes an association among nodes through proximity, shared ornament, or common na*igational landmarks. Hn*arying thumbtabs, a na*igation bar, or a miniature site map can all inform readers that the lexia in which they appear are !close! in some planned way. @ust as a prominent church spire shows a walker that two spots separated by long, winding streets are still in the same neighborhood "(#$, deliberate display of commonality in a hypertext can express relationships that indi*idual links might not emphasi+e "'$. 3>osenberg-s episodes "44$ are closely related to our 8eighborhoods0 !neighborhoods! emphasi+es the presence of patterns of meaning in the hypertext while !episode! places greater emphasis on the experience these structures create in the reader-s perception. See also >ossi-s -a.i ational %ontexts pattern"4'$5 Dor example, 8ielsen has described the inherent conflict in large )eb sites between establishing the identify of a particular hypertext and the identity of the site itself "4#$. If each page of a )eb site is separately designed and optimi+ed for its own purposes, the site as a whole may lose its coherent identity and its brand name may be obscured. /s a solution, 8ielsen proposes adopting a uniform na*igational frame or subsite as a 8eighborhood pattern that organi+es the collecti*e site, adding layers of incremental na*igational ornament to subsites as needed to create subsidiary identities. Iisual motifs often reinforce the identity of 8eighborhoods in order to establish organi+ational context or to call attention to relationships among concepts. )hen (us)e d'*rsay+ ,isite ,irtuelle ";($ adopts the structure of the museum to shape the hypertext, it effecti*ely echoes subtle issues of history, historiography, and politics that ha*e shaped both the composition and presentation of the national art collection. Millet leads to <ourbet and on to Manet0 <ourbet-s contemporary <outure, standing outside this tradition, hangs across the allJe centrale rather than in the ad7acent room. The use of inherited ornament and na*igational apparatus to identify and situate a piece of a hypertext as a component of a larger structure traces back to 1yper<ard backgrounds "($ and 1BM "#;$. In II=I "(($ and )eb S uirrel ";&$, spatial proximity is used less to establish Montage than to define spatial 8eighborhoods that represent informal relationships among elements.

Split+ oin

The Split+ oin pattern knits two or more se uences together. SplitF@oin is indispensable to interacti*e narrati*es in which the reader-s inter*ention changes the course of e*ents. If each decision changes e*erything that happens subse uently, authors cannot allow the reader to make many decisions while keeping the work within manageable bounds";9$. Splits permit the narrati*e to depend on the reader-s choice for a limited span, later returning the reader 3at least temporarily5 to a central core. 3By recording state information, the author may design subse uently,encountered se uences to split in conse uence of an early choice0 these splits, too, will usually be reconciled by a 7oin.5 The Ras*o,on pattern "94$ embeds a split,7oin within a cycle. The splitF7oin effecti*ely breaks the cycle, as readers explore different splits during each recurrent exploration, yet the cycle remains a prominent frame that pro*ides context for each strand. Sarah Smith-s /in o0 Space "'&$ uses a three,way split at the end of its entrance se uence to explore the way casual choices may in*ol*e the reader in acts she would ne*er sanction. The split appears tri*ial and game,like when first encountered, but becomes morally meaningful only after the reader has explored alternati*e paths. O)er)ie(s and to#rs "'4$"%'$ are examples of SplitF@oin where the rhetorical intent of each path is similar, but one side of the split is more detailed than the other. )riters typically offer o*er*iews and tours as a ser*ice, but SplitF@oin need not be purely utilitarian. In &o#lt*rop!s &o)e, for example, the hypertext offers a Split0 the hypertext responds ironically to the reader-s apparent moti*ation instead of responding directly to the link-s o*ert message "(2$, in a style later populari+ed by the )eb maga+ine Suck. 1ypertext may resist0 it need not merely ser*e the reader-s whims.

&issing Link
/t times, a hypertext may suggest the presence of a link that does not, in fact, exist. Dor example, Stuart Moulthrop, re*iewing Forward Anywhere "(9$, describes his hunt for a link that his reading of the hypertext led him to expect6 /t this point I began to think the two !nightmare! passages must be connected by a hypertext link, so I launched the reading program and made my way to Malloy-s screen about the freight trains of yesteryear.... there were many links to other screens, mainly screens written by Marshall 3this alternation of narrators is pre*alent throughout the work5. 8one of the links I followed, howe*er, brought me to Marshall-s *ignette about ?B@ and the headless doll.... Dor those less in lo*e with bindings, howe*er, this case of the apparently missing link may tell a different story. /s Forward Anywhere brilliantly demonstrates, hypertexts are structured in more dimensions than the line. If a link is not apparent it may be implicit. "4&$ /llusion, iteration, and ellipsis can all suggest a Missing ?ink. Structural irregularity, introduced in a context where regular structure has been established, presents an especially powerful Missing ?ink, for a place to which we cannot na*igate may seem, by its inaccessibility, uni uely attracti*e. 1arpold and @oyce ha*e argued separately that the Missing ?ink is a common if not uni*ersal hypertext motif, that na*igational choice re uires the reader to imagine not only what might appear on the chosen page but also what might ha*e appeared had she followed a different link "#4$"#2$.

Na)igational Feint
The Feint establishes the existence of a na*igational opportunity that is not meant to be followed immediately0 instead, the Deint informs the reader of possibilities that may be pursued in the future. By re*ealing na*igational opportunities e*en where they may not be immediately pursued, a hypertext writer con*eys *aluable information about the scope of the hypertext or about the organi+ation of the ideas that underlie it. Deints often appear in the guise of na*igational apparatus. Dor example, a hypertext may begin with a map or table of contents that pro*ides an o*er*iew of the entire work and pro*ides direct access to selected places within the hypertext. )hile the na*igational function is not unimportant, the rhetorical importance of the o*er*iew itself should not be o*erlooked. .rominent and detailed na*igational Deints are especially useful for establishing the scope and shape of a hypertext. @ust as important, Deints may help establish what the hypertext omits. 8otice that the feint need not always be strictly accurate0 it is sometimes useful to deli*er more than what was initially promised. Dor example, the classic 1yper%ard 234 1elp "($ presented a thumbtab o*er*iew that suggested to new readers that instructions on programming were only a minor part of the hypertext0 readers who might be deterred from using a complex product were reassured that programming appeared to be a minor feature. In fact, o*er half of the hypertext was de*oted to a programming reference manual. The na*igational feint on the co*er concealed this from programming,a*erse users, while those who wanted to consult the programming section were pleasantly surprised by its unheralded scope. Moulthrop-s ,ictory !arden opens 3in some readings5 with garden maps that schemati+e the narrati*e "(2$. The core narrati*e in =athryn <ramer-s !In Small G ?arge .ieces! ";:$ is epitomi+ed in episode outlines, cryptic epigrammatic lists that begin each narrati*e section and that lend the central narrati*e an apparent order and regularity that contrasts sharply with the disorder of the story-s Mirrorworld.

Digure #6 a typical )eb page in which a uniform na*igational frame encloses topical content.

Stephanie Strickland-s True -orth "'#$ and @. Cellowlees Bouglas- !I 1a*e Said 8othing! "%%$ use utilitarian Storyspace maps as uncon*entional Deints6 the layout of lexia simultaneously describes a structure and illustrates a central motif.

Digure 96 True 8orth-s Storyspace map In addition to their utility as introductory and framing de*ices, Deints may form a recurrent motif throughout the hypertext-s structure. Spatial narrati*es like (yst "('$ offer na*igational feints in the form of doorways, structures, and other pathways that intersect the reader-s route0 here, Deints signal possible openings for new narrati*es, roads the reader,protagonist may later choose to tra*el. In narrati*e, na*igational feints can establish spatial and temporal relationships without interrupting the narrati*e strand. By establishing a con*entional link type ,, for example, an icon denoting !link to a simultaneous e*ent occurring elsewhere! ,, a narrator can clarify and interconnect disparate e*ents without interrupting the topic under discussion. /rtful use of feints may also manage dramatic tension through foreshadowing6 if we pro*ide a link from /lice and 1erschel-s inauspicious first meeting in a Tulsa oncology clinic to the birth of their daughter in Stockholm, the knowledge gained from the existence of the link sets up undercurrents of expectation and in uiry off which the rest of the narrati*e may play. By disclosing some parts of the future we may refocus the reader-s attention and shift tension from one dramatic thread to another, or may shift energy from wondering how e*ents unfold to permit better concentration on why they unfold as they do "9#$"#&$"4$. The Deint is also important in the design of hypertextual catalogs. /s department stores disco*ered long ago, it is important both to offer the shopper a comprehensi*e array of desirable goods and to arrange those goods to form a coherent and compelling tra7ectory as the customer mo*es through the store. /t its best, this pro*ides efficiencies for both the shopper and the store6 shoppers disco*er items they want to buy but might otherwise ha*e o*erlooked, and the store gains additional transactions without incurring additional marketing costs. <atalogs similarly benefit from appropriate interconnection and by pro*iding useful Deints en route to the ob7ect of desire "4:$. By indicating the presence of other rele*ant items, the hypertext catalog can increase its efficiency without incon*eniencing or delaying the reader.

Concl#sion- Co,%ining Patterns

/ll the patterns discussed here may 3and usually do5 contain other patterns as components. / <ycle, for example, may contain se uences and cycles as well as indi*idual nodes. Two parallel cycles might be composed to form a <ounterpoint pattern, or a group of cycles might con*erge to a Tangle. The great utility of structural patterns, in fact, deri*es in large measure from the ways that patterns can be combined to form larger structures. )here a familiar pattern appears prominently, its components are percei*ed as a coherent unit, what other writers ha*e called an episode "44$ or a region "99$. By de*eloping a richer *ocabulary of hypertext structure, and basing that *ocabulary on structures obser*ed in actual hypertexts, we can mo*e toward a richer and more effecti*e hypertext criticism, one that can mo*e beyond the presentation,centered rhetoric so pre*alent in current discussions of the )eb. Simple names help us formulate concise ueries and con7ectures. / shared *ocabulary of structures can facilitate both critical and editorial discussion, not only by facilitating the study of structure but also by helping us refer succinctly to the composites and aggregates that make up a hypertext. Dinally, we may note that our current tools for *isuali+ing hypertext are not particularly effecti*e in representing the patterns described here. Many )eb,mapping programs, for example, unco*er spanning trees on the hypertext graph and so tend to hide <ycle patterns. <on*entional node,link *iews like Storyspace "9%$ and Mac)eb "4%$ represent isolated cycles fairly well but pro*ide little support for *isuali+ing contours created where many cycles intersect. The elision implicit in 8ote<ards tabletops "'9$ or the nested boxes of Storyspace ";;$ helps to keep displays simple but hides patterns that span multiple containers. Some patterns 3Mirrorworld, Missing ?ink, Deint, Montage5 are not easily represented by con*entional tools and re uire new *isuali+ations to help writers 3and readers5 percei*e, manipulate, and understand the patterns of their hypertexts.

Ackno(le'g,ents
My understanding of hypertext structure is deeply indebted to discussions with Eastgate editors Eric /. <ohen, =athryn <ramer, and Biane Ereco, and with many Eastgate authors. Eric /. <ohen, Ba*id B. ?e*ine, and Ba*id E. Burand read drafts of this work, and I am grateful for their many suggestions and impro*ements.

References
;. Espen @. /arseth, <ybertext6 .erspecti*es on Ergodic ?iterature. ;22', Baltimore6 @ohns 1opkins Hni*ersity .ress. %. Sholem /leichem, Te.ye the #airyman and The 5ailroad Stories3 ;2:', 8ew Cork6 Schocken Books. #. <. /lexander, S. Ishikawa, M. Sil*erstein and et al., A Pattern Lan ua e3 ;2'', Axford Hni*ersity .ress.

9. Mary,=im /rnold, !?ust!, 'ast ate $uarterly 5e.iew o0 1ypertext, ;22#. ;3%5. (. Bill /tkinson, 1yper%ard. ;2:', /pple <omputer <o.6 <upertino </. 4. @ohn Barth, !<lickK!, The Atlantic (onthly3 ;22', '. Mark Bernstein, !The Bookmark and the <ompass6 Arientation Tools for 1ypertext Hsers!, SI!*IS 6ournal, ;2::. 23;2::56 p. #9,9(. :. Mark Bernstein, !<on*ersations )ith Driends6 1ypertexts )ith <haracters!, in 1ypermedia #esi n, S. Draise, et al., Editor. ;22(, Springer6 ?ondon. p. %&',%;(. 2. Mark Bernstein, !Beeply Intertwingled 1ypertext6 The 8a*igation .roblem >econsidered!, Technical %ommunication, ;22;. ;22;6 p. 9;,9'. ;&. Mark Bernstein, !8eighborhoods in Spatial 1ypertext!, SI!LI-/ Bulletin, ;22:. in press. ;;. Mark Bernstein, @ay Ba*id Bolter, Michael @oyce and Elli Mylonas. !/rchitectures for Iolatile 1ypertext!, 1ypertext '723 ;22;. San /ntonio6 /<M. p. %9#,%4& ;%. Mark Bernstein, Michael @oyce and Ba*id B. ?e*ine. !<ontours of <onstructi*e 1ypertext!, European <onference on 1ypernedia Technology. ;22%. Milano6 /sssociation for <omputing Machinery. p. ;4;. ;#. @ay Ba*id Bolter, &ritin Space3 ;22;, ?awrence Erlbaum /ssociates. ;9. @ay Ba*id Bolter and Michael @oyce, !1ypertext and <reati*e )riting!, 1ypertext '893 ;2:'. <hapel 1ill6 /<M. p. 9;,(& ;(. Bomini ue Brisson, (us)e d'*rsay+ .isite .irtuelle3 ;22', BME Interacti*e6 .aris. ;4. .eter @. Brown, !Bo we need maps to na*igate round hypertext documentsL!, 'lectronic Publishin :: *r ani;ation, #issemination and #esi n, ;2:2. %3%56 p. 2;, ;&&. ;'. >obert <oo*er, ed. 1ypertext 1otel. ;22;, http6FFduke.cs.brown.edu6::::F ;:. =athryn <ramer, !In Small G ?arge .ieces!, 'ast ate $uarterly 5e.iew o0 1ypertext, ;22#. ;3#5. ;2. )illiam <rowther and Bon )oods, Ad.enture. ;2'4. %&. ?. BeCoung, ?inking <onsidered 1armful, in '%1T'74 : 1ypertext+ %oncepts, Systems and Applications, S. >i+k, /ndre. ;22&, <ambridge Hni*. .ress6 p. %#:,%92. %;. /bbe Bon, Bubbe-s Back .orch. ;224, http6FFwww.bubbe.comF %%. @. Cellowlees Bouglas, !I 1a*e Said 8othing!, 'ast ate $uarterly 5e.iew o0 1ypertext, ;22#. ;3%5.

%#. @. Cellowlees. Bouglas, Print Pathways and Interacti.e Labyrinths+ 1ow 1ypertext -arrati.es A00ect the Act o0 5eadin 3 ;22%, 8ew Cork Hni*ersity6 8ew Cork. %9. =. M. Dairchild, S. E. .oltrock and E ). Durnas, !Sem8et6 three,dimensional graphic representation of large knowledge bases!, in %o niti.e Science and its Applications 0or 1uman:%omputer Interaction, >. Euinon, Editor. ;2::, ?awrence Erlbaum6 1illsdale 8@. p. %&;,%##. %(. Edward Dalco, A #ream &ith #emons3 ;22', Eastgate Systems, Inc.6 )atertown, Massachusetts. %4. Edward Dalco, !Sea Island!, 'ast ate $uarterly 5e.iew o0 1ypertext, ;22(. %3;5. %'. >ichard Duruta, et al., !1ypertext .aths and the )orld,)ide )eb6 Experiences with )alden-s .aths!, in Proc3 o0 1ypertext'793 ;22', Southampton, H=. p. ;4',;'4. %:. Sil*io Eaggi, From Text To 1ypertext+ #ecenterin the Sub<ect in Fiction, Film, the ,isual Arts, and 'lectronic (edia3 ;22', .hiladelphia6 The Hni*ersity of .ennsyl*ania .ress. %2. Erich Eamma, >ichard 1elm, >alph @ohnson and @ohn Ilissides, #esi n Patterns+ 'lements o0 5eusable *b<ect:*riented So0tware3 ;22(, >eading, M/6 /ddison,)esley. #&. @ohn Eardner, The Art o0 Fiction+ -otes *n %ra0t 0or =oun &riters3 ;2:#, 8ew Cork6 Iintage Books. #;. Dranca Ear+otto, .aolo .aolini and Mark Bernstein, !Tools for Besigning 1yperdocuments!, in 1ypertext>1ypermedia 1andbook, E. Berk and @. Be*lin, Editor. ;22;, McEraw,1ill6 Mew Cork. p. ;'2,%&:. #%. B. Eelernter, (irror &orlds : or The day so0tware puts the uni.erse in a shoebox3 ;22;, Axford Hni*. .ress. ##. >obert @. Elushko, Besign Issues for Multi,Bocument 1ypertexts, in 1ypertext'873 ;2:2, .ittsburgh. p. (;,4&. #9. /drienne Eisen, Six Sex Scenes+ a no.ella in hypertext3 ;224, www.altx.comFhyperxFsssFindex.htm6 #(. <arolyn Euyer, Muibbling. ;22%, Eastgate Systems, Inc.6 )atertown M/. #4. Terry 1arpold, !Threnody6 .sychoanalytic Bigressions on the Sub7ect of 1ypertexts!, in 1ypermedia and Literary %riticism, .. Belany and E... ?andow, Editor. ;22;, MIT .ress6 <ambridge. p. ;';,;:;. #'. Shelley @ackson, Patchwork !irl+ by (ary>Shelley>and 1ersel03 ;224, Eastgate Systems, Inc.6 )atertown M/. #:. Michael @oyce, a0ternoon, a story3 ;22&, Eastgate Systems, Inc.6 )atertown, M/.

#2. Michael @oyce, !8once Hpon Some Times6 >ereading 1ypertext Diction!, (odern Fiction Studies, ;22'. 9#3#56 p. ('2,(2'. 9&. Michael @oyce, *0 Two (inds+ 1ypertext Peda o y and Poetics3 ;229, /nn /rbor6 Hni*ersity of Michigan .ress. 9;. Michael. @oyce, !Siren Shapes6 Exploratory and <onstructi*e 1ypertext!, in Academic %omputin 3 ;2::, p. ;; ff. 9%. Michael @oyce, Storyspace as a hypertext system for writers and readers of *arying ability, in 1ypertext'723 ;22;, San /ntonio. p. #:;, #:'. 9#. Michael @oyce, Twili ht, a Symphony3 ;224, Eastgate Systems, Inc6 )atertown Massachusetts. 99. Ba*id =olb. !Scholarly 1ypertext6 Self,>epresented <omplexity!, 1ypertext 793 ;22'. Southampton, H.=.6 /<M. p. %2,#' 9(. Ba*id =olb, Socrates in the Labyrinth+ 1ypertext, Ar ument, Philosophy3 ;229, Eastgate Systems, Inc.6 )atertown, Massachusetts. 94. /kira =urosawa and Bonald >itchie, 5ashomon. >utgers Dilms in .rint, Iol. 4. ;2:', >utgers Hni*ersity .ress. %&;. 9'. Eeorge .. ?andow, The #ickens &eb3 ;22%, Eastgate Systems, Inc.6 )atertown, M/. 9:. Eeorge .. ?andow, 1ypertext ?34+ The %on.er ence o0 %ontemporary %ritical Theory and Technolo y, %nd edn. ;22', Baltimore6 @ohns 1opkins .ress. 92. Eeorge .. ?andow. !>elationally Encoded ?inks and the >hetoric of 1ypertext!, 1ypertext 893 ;2:'. <hapel 1ill6 /ssociation for <omputing Machinery. p. ##;,#99 (&. Eeorge .. ?andow, &ritin At The 'd e3 ;22(, Eastgate Systems, Inc.6 )atertown, Massachusetts. (;. Eeorge .. ?andow and @on ?anestedt, The In (emoriam &eb3 ;22#, Eastgate Systems, Inc.6 )atertown M/. (%. Beena ?arsen, Samplers+ -ine ,icious Little 1ypertexts3 ;22:, Eastgate Systems, Inc.6 )atertown Massachusetts. (#. =e*in ?ynch, The Ima e o0 the %ity3 ;2:%, MIT .ress. (9. @udy Malloy and <athy Marshall, Forward Anywhere3 ;224, Eastgate Systems, Inc.6 )atertown, Massachusetts. ((. <atherine <. Marshall, Drank .M. Shipman III and @ames 1. <oombs, II=I6 Spatial 1ypertext Supporting Emergent Structure, in '%1T'7@3 ;229, Edinburgh. p. ;#,%#.

(4. @ohn McBaid, "ncle Buddy's Phantom Funhouse3 ;22%, Eastgate Systems, Inc.6 )atertown, M/. ('. >and and >obyn Miller, (yst. ;22#, <yan6 (:. Stuart Moulthrop. !1ypertext and -the 1yperreal-!, 1ypertext '873 ;2:2. .ittsburgh6 /<M. p. %2;,%2: (2. Stuart Moulthrop, ,ictory !arden3 ;22;, Eastgate Systems, Inc.6 )atertown, M/. 4&. Stuart Moulthrop, !)here ToL / >e*iew of Forward Anywhere by <athy Marshall and @udy Malloy!, %on.er ence+ The 6ournal o0 5esearch into -ew (edia Technolo ies, ;22'. 4;. @anet Murray, 1amlet *n The 1olodeck+ The Future o0 -arrati.e in %yberspace3 ;22', 8ew Cork6 The Dree .ress. 4%. @ocelyn 8anard and Marc 8anard. !Hsing Structured Types to incorporate =nowledge in 1ypertext!, 1ypertext'723 ;22;. San /ntonio6 p. #%2,#9# 4#. @akob 8ielsen, !The >ise of the Sub,Site!, Alertbox, ;224. http6FFwww.useit.comFalertboxF24&2.html. 49. 1.*an Byk .arunak, 1ypermedia Topologies and Hser 8a*igation, in .roc. of 1ypertext '873 ;2:2, .ittsburgh. p. 9#,(&. 4(. <hristriane .aul, "nreal %ity+ A 1ypertextual !uide To T3 S3 'liot's The &aste Land3 ;224, Eastgate Systems, Inc.6 )atertown, M/. 44. @im >osenberg, The Structure of 1ypertext /cti*ity, 1ypertext '7A3 ;224, /<M .ress6 )ashington B<. p. %%,#&. 4'. E. >ossi, B. Schwabe and /. Earrido. !Besign >euse in 1ypermedia /pplications Be*elopment!, Proc3 o0 1ypertext'793 ;22'. Southampton, H=6 p. (',44 4:. Ba*id Siegel, %reatin /iller &eb Sites3 ;224, 8ew Cork6 1ayden Books. 42. Ba*id Siegel, )eb Site. ;22', http6FFwww.dsiegel.comF '&. Sarah Smith, /in o0 Space3 ;22;, Eastgate Systems, Inc.6 )atertown M/. ';. /rt Spiegelman, (aus3 ;2:4, 8ew Cork6 >andom 1ouse. '%. .. Ba*id Stotts and >ichard Duruta, !.etri,net based hypertext6 Bocument structure with browsing semantics!, A%( Transactions on *00ice In0ormation Systems, ;2:2. '3;56 p. #,%2. '#. Stephanie Strickland, True -orth> ;22', Eastgate Systems, Inc.6 )atertown, M/.

'9. >andall 1. Trigg, !Euided Tours and Tabletops6 Tools for <ommunicating in a 1ypertext En*ironment!, A%( Transactions on *00ice In0ormation Systems, ;2::. 43956 p. #2:,9;9. '(. =. Htting and 8icole Cankelo*ich, !<ontext and orientation in hypermedia networks!, A%( Transactions on *00ice In0ormation Systems, ;2:2. '3;56 p. (:,:9. '4. .olle T. Nellweger, Scripted Bocuments6 / 1ypermedia .ath Mechanism, in 1ypertext'873 ;2:2, .ittsburgh. p. ;,;9.