Hypertext: an Introduction and Survevj

SURVEY & TUTORIAL SERIES

Hypertext: An Introduction

and SurvevJ

Jeff Conklin Microelectronics and Computer Technology Corp.

ost modern computer sys- design issues that go into fashioning a tems share a foundation hypertext environment. which is built of directories containing files. The files consist of text which is composed of characters. The text Hypertext systems that is stored within this hierarchy is linear. feature machine- For much of our current way of doing graphically (as labelled tokens) and in the business, this linear organization is suffi- supported links-both database (as pointers). (See Figure I .) cicnt. Howcvcr, for morcand moreappli- within and between But this simple idea is creating much cations, a linear organization is not excitement. Several universities have adequate. For example, the documenta- documents-that open created laboratories for research on hyper- tion of a computer program* is usually text, many articles have been written about either squcczcd into the margins of the exciting new the concept just within the last year, and program, in which case it is generally too possibilities for u!Gng the Smithsonian Institute has created a terse to be useful, or it is interleaved with demonstration laboratory to develop and the text of the program, a practice which the computer as a display hypertext technologies. What is all breaks up the flow of both program and communication and the fuss about? Why are some people will- documentation. ing to make extravagant claims for hyper- As workstations grow cheaper, more thinking tool. text, calling it “idea processing” and “the powerful, and more available, new possi- basis for global scientific literature”? bilities emerge for extending the tradi- In this article I will attempt to gc~ at the tional notion of “flat” text files by essence of hypertext. I will discuss its allowing more complex organizations of advantages and disadvantages. I will show the material. Mechanisms are being that this-new technology- opens some very devised which allow direct machine- exciting possibilii;es, particularly for new supported references from one textual pioneers of hypertext, once defined it as ‘?i’ uses of the computer as a communication chunk to another; new interfaces provide Tcorn-.*‘b’ ma I$ onofnaturallanguageiejit’iiriih ‘-’ fand thinking tool. However, the reader the user with the ability to interact directly ‘_Ihe c&puter*s capacity for interacti&* _ who has not used hypertext should espcct with these chunks and IO establish new s branching, or dynamic display. . . of a that at best he will gain a pcrccption of relationships bctwcen them. Theseexten- iL non~neartext_ . ~whichcannot~p,+.,t~ hypertext as a collection of intcrcsting fca- sions of the’traditional text fall under ihe ”conveniently on a conventional page.“’ cures. Just as a description of electronic general category of hyperkxf (also known This article is a survey of existing hyper- spreadsheets will not get across the real elc- as noniineurfext). Ted Nelson, one of the text systems, their applications, and their gancc of that tool, this article can only hint design. It is both an introduction to the at the potentials of hypertext. In fact, one 'D~cumm~a~ion is Ihe uncrccutablc English text world of hypertext and, at a deeper cut, a must work in current hypertext environ- which explains the logic of the program which it accompanies. survey of some of the most important ments for a while for the collection of fea-

September 1987 0018-9162/81/o9ooM)11501.00~~961 IEEE that is common’io many hypertext systems I is the heavy use of windows that have a : one-to-one correspondence with nodes in. ., the database. 1 consider this feature to be of secondary importance. A I One way to delimit hypertext is to point out what it is not. Briefly, several systems B have some of the attributes Sf hypertext CL3 but do not qualify. Window systems fall 4, into this category; while window systems do have some of the interface functional- ’

I I could claim that a file system is a database, I I I i I I : Display screen and that one movef amongnodes(files) by 8I ,8 : II ,1 : t1 8I ,I --AI -L-_I_------r I It I 1 I : I :i , I I @~m~t;,3imilarly, most outline proces- I sor’s (such as ThinkTank) do not qualify. I They provide little or no support for refer- I ences between outline entries, although their integrated hierarchical database and I interface do approximate hypertext better I than the other systems that I have men- I tioned. Text formatting systems (such as I I Troff and Scribe) do not qua!ify. They allow a tree of text fragments in separate I I files to be gathered into one large docu- !-I I- ----_------I ment; however, this structure is hierarchi- Hyperlext database cal and provides no interface for on-line navigation within the (essentially linear) document. Similarly, database manage- Figure 1. The correspondence bctwccn windows and links in the display, and nodes ment systems (DBMSs) have links of var- and links in the dafabasc. In this example, each node in the hypertext database is ious kinds (for example, relational and displayed in a separate window on the screen when requested. The link named “b” object-oriented links), but lack the single in window A has been activated by a pointing device, causing a new window named coherent interface to the database which “B” IO be created on the screen and filled wifh the text from node B in the data- is the hallmark o[ hypertext. _ base. (Gcncrall~. links can have names that arc different from the name of the fqS’~ideodisCie~~orb~~m~~g~.~ .., _I node they point to.) ,‘fhere is,&&in~~~~~~est in theext+& 6&3&i I. $+Xztext to the~n$+~en&l &cept or. . ype&&in which_ th;&‘n;$ii wh’ich’~ ‘i“y’iI P are i n&&k$ _. tog&h&, can be- &t, F r &- -+raphiF, digitized speech, audio record- a ‘T; !’ ISI@, picty~e;s, animatir+ film clips, and ,.. .JE, ‘Sistimably,_ _.“..--ttast&‘pdors --.-a and tactilcsentz. F i tures to coalesce into a useful tool. At thu pomt, httle has been done i the design and engineering t One problem with identifying the i tial aspects of hypertext is that the term these additional modalities, “hypertext” !-I% been used quite loosely in ing capability which allows a nonlinear although many of the high-level design ; the pas[20 years for many diffcrcnt collcc- organimtion of 1~x1. An additional feature issues arc likely IO be shared with hypcr- [ions of featurei. Such tools as window 1~x1. Thercforc, this survey will primarily systems, elccrronic mail, and telecon- *While this arliclc seeks to establish the criterion cl address the more conservative text-based i ferenckg share iearures with hypertext. mxhinc-supvrlcd links as the prims? crilcrion of systems. hypertext. this is by no means an xccp~rd definition. This anicle focuses on machine-supported Thcrcforc I will also rcvicw 3nd discuss 3cvnc syrwnr links (both within and between docu- wblch have a weaker notion 01 links. A glimpse of using hypcrtcxt. It is USC- I I8 COMPUTER ‘._ ~.‘. -. _- _, i i ( - ,_’ ._ _ A. . . “_ ,; . . ._ - ~-. -. : r .-*.

ful IO have a sense of the central aspects of using a hypertext system, particularly if you have never seen one. Below is a list of _ , * Display screen wilh Wwsrr the features of a somewhat idealized hypertext system. Some existing systems I pXnt have more features than these, and some /stems lertext ns fall ‘stems ional- :I” of single and each has a name or title which e lack ,is always d&played in the window. How- le sys- ever, only a small number of nodes are .t; one yver “open” (as windows) on the screen at Ibase, the same time. ‘cs) by l Standard window system operations their are supported: Windows can be reposi- rtext, tioned, resized, closed, and put aside as. icated small window icons. The position and size more of a window or icon (and perhaps also its Yerely color and shape) arecues to remembering :ditor the contents of the window. Closing a win- -oces- dow causes the window to disappear after alify. any changes that have been made aresaved refer- to the database node. Clicking with the ough mouse on the icon of a closed window e and causes the window to open instantly,. ,_ ,. _. )e i \ .* Windows can contain any number of: n Jink.ico,p+ thich repiesent-pointers to C. other nodes in the database. The link icon They contains a short textual field which sug- Irate gcsts the contents of the node it points to. locu- Clicking on a !ink icon with the mouse rchi- causes the system to find the referenced L -----__--____- 1 .-line node and to immediately open a new win- Hypertext database Icar) dow for it on the screen. ._ ._ : _ iage- ,, l The user can easily create new nodes i var- and new links to new nodes (for anhota- Figure 2. The screen at the top illustrates how a hypertext browser provides a direct and lion, comment; elaboration, etc.) or to two-dimensional graphic view of the underlying database. In this illustration, the ;nglc existing nodes (for establishing new con- node “A” has been selected for full display of its contents. Notice that in the hich nections). _,_.. browser view you can tell not only which nodes are linked IO A but also how the I l Thedatabasecaribebrdwsedin tliiii--“subnetwork fits into the larger hyperdocument. (Of course, hyperdocuments of an> age, ways: (1) by following links and opening ‘size cannot be shown all al once in a browser-only portions can be displayed.) an of windows successively to examine their Con-_:? ‘_. -’ i rt of tents, (2) bysearchingthe network (or pat-t -“. ’ hich of it) for some string,*+ keyword, or _+’ ext, attribute value, and (3) ‘by navigating i- 3rd- around -theT%yperdocument using a- 5- - and browserthat displays the network graphi- Pen- zcally. _The user can select whether the _<:rr”rcci”C”‘.Y---F one nodes and links display their labels or npf. $f hypertext syster+ As the hyperdocu- Using a browser can be likened to using -ing The browser is an importantcomponent ment grows more complex, it becomes dis- visual and tactile cues when looking for a its, tressingly easy for a user to become lost or cerfain page in a book. Sometimes wc .ign ‘NOIC 11131 I am arc dcscribrng two uses oficons: those disoricntcd. A browser displays some or ail rcmembcr the pcneral way rhc page looked 1cr- lhal function as placcholdcrs for windows that have of the hyperdocument as a graph, provid- been temporarily pur aside, and those within umdows and about how far it was through the ti thal rcprcseni hnks IO other nodes. ing an important measure of contextual book. although we don’t recall the page sec._ and spatial cues to supplement the user’s number or even which keyword terms model of which nodes he is viewing and would help us find it by using the index or ‘*A wring 1s a scrics of alphabclic and numenc iharac- lcrs of any length, for example “liricning” or how they are related to each other and their table oicontenfs. The browser display can se- “CiOO274.” neighbors in the graph. (See Figure 2.) be similarly scanned and scrolled when the

ER September 1987 IS user has forgotten all but the appearance (tic user interface a&s). while others or location of a node. have focused on the database issues of the buck end (the database server). Table I identifies various features of the different Hypertext hypcrtextual representation. hypertext systems which have been imple- implementations But if one insists, as most modern mented. proponents of hypertext do, that naviga- The history of hypertext is rich and var- tion through hypertextual space must be Macro literary systems. The earliest ied because hypertext is not so much a new computer-supported in order to qualify as visions of hypertext focus on the inttgra- idea as an evolving conception of the pos- true hypertext, then the field is narrowed tion of colossal volumes of information to i sibleapplications of the computer. Many considerably, and the history likewise make them readilyaccessible via a simple people have contributed to the idea, and shortened. and consistent interface. The whole net- each of them seems to have had something In some ways, the people who first work publishing system constitutes a different in mind. In this section, I will described hypertext-Bush, Engelbart, dynamic corpus to be enriched by readers review these theorists and their ideas in an Nelson-all had the same vision for hyper- without defacing the original documents; effort to present a historical perspective as text as a path to ultimate human-computer thus, the difference between authors and well as to sketch some of the hypertext interaction, a vision which is still alive readers is diminished. The advent of the applications that have been devised to today among hypertext researchers. Thus computer has brought this vision closer IO date. I do not describe the individual sys- the historical review below stresses the reality, but it has also revealed the terns and ideas reviewed here in any detail. early development of ideas about hyper- monumental problems inherent in this For more detailed information, the reader text as much as the more contemporary application area. is invited to consult the literature directly. , F”21_ m7 ,--: .‘.,.. .,_ _ _ . ‘~~“u; cpTp’ -7s !?@ementation e ffy * Bush 3 Memex. Vannevar Bush, Presi- ne kmd of manual hypertext s the tra- .i Because of the difficulty of precisely dent Roosevelt’s Science Advisor, is #&tional use of 3 x 5 index cards for note classifying hypertext systems according to credited with first describing hypertext in i taking. Note cards are often referenced’to *their features, my description will list sys- his 1945 article “As We May Think,“’ in -each other, as well as arranged hierarchi- tems according to application. There are which he calls for a major postwar effort cally (for example, in a shoebox or in four broad application areas for which to mechanize tht scientific literature sys- hypertext svstems-. have been developed: rubber-banded bundles). A particular -..“.T . ..-.“. . .._ tem. In the article, he introduces a machine advantage of note cards is that their small mucro lilerarysyslems: the study of for browsing and making notes in an rechn’oiogies G’s&& l&e on-line size modularizes the notes into small extensive on-line text and graphics system. chunks. The user can easily reorganize libraries in which interdocument links , a This memexcontained a very large library set of cards when new information sug- are machine-supported (that is, all as well as personal notes, photographs, gests a restructuring of the notes. Of publishing, reading, collaboration, .- and sketches. It had several screens and a course, a problem \vith note cards is that and criticism takes place within the facility for establishing a labelled link network); rheuser can have difficulty finding a spe- between any IWO points in the entire cific card if he has many of them. problem ezploraffon Tookloo to library. Although the article is remarkably Another kind of manual hypertext is the support early unstructured thin$ing foresightful, Bush did not anticipate the reference book, exemplified by the dic- on a problem when many discon- power of the digital computer; thus his tionary and the encyclopedia. In the sense nectcd ideas come to mind (for exam- memex uses microfilm and photocells to that each of these can be viewed as a graph ple, during early authoring and do its magic. But Bush did anticipate the of textual nodes joined by referential links, outlining, problem solving, and pro- information explosion and was motivated they are very old forms of hypertext. As gramming and design); in developing his ideas by the need to sup- one reads an article or definition, explicit :- browsingJy~fems:sy;frems similar to %. ‘2 . ._.: port more natural forms of indexing and references to related items indicate where macrolite’ra’rjsystem?, but smaller in retrieval: to get more information about those items. scale (for teaching, reference, and The majority of people’s transactions with public information, where ease of use The human mind . . . operates by association. Man cannot hope fully IO duplicate this a dictionary make use of the linear (alpha- is crucial); ~_. _.?-i).~..-r--. -I_ - mental process arMicially, but he certainly betic) ordering of its elements (detkirions) ..* genirul hyp&G&hnologv. genera! Pugh1 IO be able IO learn from it. One cannot for accessing a desired element. An ency- “- ‘purpose.systenis designed to‘allow hope IO equal the speed and flexibility with clopedia, on the other hand, can best be experimentation with a range of which the mind follows an associative trail. but it should be possible to-beat ~hGaind used rg explorGhe local nodes in the “net- hypertext applications (for reading, decisively in regard to the pcrmancnc< and work,” once one has found the desired writing, collaboration, etc.) clarity of the items resurreclcd from entry through the alphabetic index. These categories are somewhat infor- storage.’ There are also many documents in mal. Often thesingle application IO which Bush described the essential feature of which references io other parts of the a system has been applied to date derer- :he mcmcx as the ability to tic IWO ircms documcnr. or IO other documents, consti- mints which category it is dcscribcd in. togcrhcr. The mechanism is complex, bur (MC a major portion oithc work. Both the Bear in mind that some of the systems clever. The user has IWO documents that he Talmud, with its heat,! use of annotations mentioned below are full-scale environ- wishes to join into a trail he is building, and nested commentary, and Aristotle’s ments, while others are still only concep- each document in its own vicwcr; hc raps writings, with their reliance on references tual sketches. Some systems have focused in rhc name of the link, and that name to other sources, are ancient prototypesof more on the development of thejronr end appears in a code space at the bottom of

20 COizlPUTER

-_ - “.. ._--_.______, : , Toblc 1. Hypertext syslcms rnd lhcir feature.

r Hierarchy Graph- Link Attri- Paths Vcr- Proced- Keyword Text Con- Pictures Graphica based Types butes sions ural or Editor current or Browser Hypertext Attach- String Multi- Graphics Systems ment Search users Boxer Yes Yes Fixed' No’ No No Yes Yes Emacs No Yes Yes earliest . ntegra- CREF Yes Yes Yes No No By link No Yes Zmacs No Yes No 1tion to simple Emacs INFO Yes No No No No No No Yes Emacs Np No No Jle net- IBIS Yes Yes Yes No No Byiink No No A basic Yes No’. No :utes a text readers cdilor .ments; 3rs and Intermedia YeS Yes Yes Yes No’ No No’ Yes Custom Yes Yes Yei : of the ;oser to KMS Multiple Yes Fixed No No’ Yes Yes Yes Text/ Yes Yes No :d the graph. in this WYSIWYG Neptune Yes Yes Yes Yes No Yes Yes Yes Smalltalk- Yes Yes Yei 80 editor , Presi- SOr, is NLS/Augment Yes Yes Yes Yes Yes Yes Yes Yes Custom Yes Yes No rtext in NoteCards Multiple Yes Yes YCi k,“‘in Nodes No No Yes Yes Interlisp Yes Yes . effort 3utline Processors Yes No No No No No No Yes Various No No NO Ire sys- machine PlaneText Unix Yes No No No h‘o No Unix/ SunView Yes Yes Yei in an file sys. grep text ed. ys’ - jymbolics Yes Yes No No Yes No No Yes None No No NO 1. 3ocument rapns, Examiner 3 and a d link XNVIEW Yes No No No No h’o No No linccd.1 No No NO entire Unix rkably lte the rexmet Multiple Yes Yes Yes Yes h’o No Keyword Any No No NO !us his iyperties No Yes No No No No No No’ A basic No Yes No :ells 10 text edilor ire rhe ivaled WE Yes Yes No Fixed No* No: No’ No Smalltalk- No’ Yes Yei 0 sup- 80 editor ig and Yanadu No Yes Yes Yes Yes Yes No No Any No Yes No rsocia- ZOG Yes No No No No No Yes Full text Spec. Pur. Yes No NO UC this rtsinly ’ Can be user programmed. cannot ’ Planned for next version. y with e trail, -In this table, each column rcprcs?nts one possible feature or ability that a hypertext systun can provide. The negative or affirmalive enws in rhe mind - -3able indicate whether the corresponding hypertext system meets the standardcrireria for a specified fealurc. These criteria are lisred bclo\v. x and from Hierarchy: Is there specific support for hierarchical smcturcs? Graph-based: Does the sys[cm support nonhierarchical (cross-reference) links? Link types: Can links have types? Al[ributes: Can user-designated attribute/value pairs be associated \\ith nodes or links? Palhs: Can many links be srrung togcthcr inlo a single persistent objca? Versions: Can nodes or links have more than a single version? Procedural attachment: Can arbitrary csccurablc procedures bc atrachcd IO events (such as mousing) a~ nodes or links? .%ring search: Can the hypcrdocumcnr be searched for strings (including keywords)? TCXI cdilor: What cdilor is used IO create and modify the comems of nodes? Concurrent mul[iusers: Can scvcral users edit the hypcrdocument at Ihc same lime? Plclurcs or graphics: Is some form of piclorial or graphical informalion supporrcd in addition IO ICXL? Graphics browser: Is there a browser which graphically presents [he nodes and links in rhe hypcrdocumenr? i-ER September 1987 t ideas about augmentation had become more specific, and had been implemented as NLS (ON Line System) by the Augmen- ted Human Intellect Research Center at SRI. NLS was designed as an experimental tool on which the research group developed a system that would be adequate to all of their work needs, by placing in computer store all of our speci!i- cations, plans, designs, programs, documentation. reports, memos, bibliography and rcrerence notes, etc., and doing all of our scratch work, planning, designing, debug- ging, erc., and a good deal of our intercom- munication, via the consoles.‘ These consoles were very sophisticated by the standards of the day and included television images and a variety of input devices, including one of Engelbart’s best known inventions, the mouse: Files in NLS were structured into a hierarchy of segments** called slafements, each of which bore an identifier of its level within the file. For example, a document might havestatements”l,““la,” “.lal.” “la&” “ 1 b,” etc., though these identi- fiers did not need to be displayed. Any number of reference links couldbe estab- lished between statements within files and between files. Note that this is a structure which is primarily hierarchical, but which allows nonhierarchical links as well. The ,- Figure 3. Engelbart 31 fhe NLS/Augmenf workstation. Note the chord key se1 importance of supporting both kinds of under Engelbart’s left hand. The chord key set is optional for Augment. If is a structures is a point to which I will return remarkable accelerator for character-driven commands and mouse-select screen - later. The system provided several ways to operands. traverse the statements in files. NLS, like other early hypertext systems, emphasized three aspects: a database of nonlinear text, viewfilrers which selected information from this database, and viebcts each viewer; out of view, the code space is In this stage. the symbols with which the which structured the display of this infor- human tcprcscnts the conccps he is manip- also filled with a phoroccll-rcadablc dot mation for the terminal. The availability code that names the other document and ulating can be arranged before his eyes, moved. stored, recalled, operated upon of workstations with high resolution dis- the current position in that document. according IO extremely complex rules-all in plays has shifted the emphasis to more Thereafter, when one of these items is in very rapid response IO a minimum amoum of graphical depictions of nodes, links, and inlormation supplied by the human, by view, the other can be instantly recalled networks, such as using one window for merely by tapping a button below the cor- means of special cooperative technological devices. In the limit of whal we might now each node. responding code space. Bush admitted that imagine, this could be a computer. with NLS provided viewing filters for the file many technological breakthroughs would which individuals could communicate rapidly structure: One could clip the level (depth) and easily, coupled to a three-dimensional be needed to make his memex practical, of hierarchy displayed, truncate the num- but he felt that it was a technological color display with which extremely sophisti- caled images could be constructed . . .) ber of items displayed at any level, and achievement worthy of major expenditure. write customizedfilters (in a “high-level His proposed sysrem, -AM/T content analysis language”) that displayed Engflban’s M!.S/Augmenr. Just less (Human using Language, Artifacts, and only statements having the specified con- than two decades later Douglas Engelbart, Methodology, in which he is Trained), tent. NLS also introduced the concept of at Stanford Research Institute, was included the human user as an essential ele- influenced by Bush’s ideas. In 1963. Engel- ment: The user and the computer were bart wrote “A Conceprual Framework for dynamically changing components in a the Augmcntarion of Man’s Intcllcct.“’ symbiosis which had the effect of Engelbart envisioned that computers “amplifying” the native intelligence of the would usher in a new stage of human evo- user. This is still a common vision among - luiron, characterized by “aulomatcd dcvclopcrs of hypcrrcst systems. csrcrnal symbol manipulation”: Five years later, in 196s. Engclbart’s “Scamcn~r wxc llmaicd IO 3CCOcharac~rrs m Icngh.

22 COMPUTER

_. -.-_- - -- xcome multiperson distributed conferenc- mcntcd jng/editing. Llf” !q_ NLS has evolved over the years. It is at now called Augment (or NLS/Augment) eb..dzn- and is marketed as a commercial network p devel- system by McDonnell Douglas. In luate to . developing NJ_& the emphasis has been on creating a consistent environment for spccifi- *‘knowledge workers” (that is, office xumen- automation for software engineers). The ohy and system now includes many forms of II of our , debug- computer-supported communication, itercom- both asynchronous (email with links to all documents, journaling of ideas and sticated exchanges, bulletin boards, etc.) and syn- ncluded chronous (several terminals sharing the If input same display, teleconferencing, etc.). It rt’s best includes facilities for document production and control, organizational and pro- o a hier- ject information management, and etnenls, software engineering. (See Figures 3 and ,?Cl STRUCTURE CUTOFF.’ Shou only / the ~ fits level 4.) 7C2 LLUEL‘CLIPPING. For xument 7C3 STRTMENT TRUNCfiTlON. for ._ 7C4 INTER-STATMEM SEPRRNTION., ’ “la1 ,” Nelson’s Xanadu Projecf. During : 76 (Note: 'The foregoing vieu e identi- Engelbart’s development of Augment, ~ “7~6 STRTMEM iii~~fgs AND NMES: ed. Any another hypertext visionary, Ted Nelson, .. -?C7 FROZEN STRTMENTS. b uorker' 3e estab- was developing his own ideas about aug- ,'.?CS USEI+PEClFlED COKTENT ’ ‘:, files and mentation, but with an emphasis on creat- . , E tructure ing a unified literary environment on a u,:’ :h global scale. Nelson coined the term ne “hypertext.” His thinking and writing are kirrrrs of the most extravagant of any of the early Figure 4. Augment display showing five windows. Window 1 (W-l) has a passage ill return workers. He named his hypertext system as if embedded in a message, showing a link to Branch 7c of Document 2250 in the I ways to Xanadu, after the “magic place of literary OAD Journal. A ViewSpcc (“rbtzgm”) provides Ihc following spccific:,iions: tar- memory” in Samuel Ta; lor Coleridge’s get level plus one, 1runcaIe to one line per statement, no blank lines beIcccn stale- systems, poem “Kubla Khan.” InXanadu, storage men& show only Ihat branch (e.g., noI Branch 7d), and Iurn on Location abase of space is saved by the heavy use of links. Numbers. Window 2 (W-2) shows the view obtained with a jump link command. selected Only the original document and the To perform a jump link command, Ihe opcralor clicks on Ihc link in W-1, lbcn .nd views changes made to it are saved. The system moves the cursor into W-2 for Ihc final click. Thr very fop-lcf~ syslcm rncssagc lis infor- easily reconstructs previous versions of announces that the desired Journal Item has been accessed, and Ihc clusIcr al Ihe lilability documents. Nelson describes his objec- top left of the screen verifies that the view is clipped to three levels and the state- ition dis- tives as follows: ments truncated to one line each. Window 3 (W-3) shows an indirecl link fhat to more Under guiding ideas which are not technical specifies Ihe linkage palh. In effect, this link says “go to the statemen! in the file nks, and but Iilcrary, weareimplemenringa system for named ‘Refd,’ follow the link found there to iIs target file, and in Ihat file find tdow for storageand retrieval of linked and window- Location Number 6.” Kale IhaI Ihe same ViewSpec is specified here as for Ihc link ing text. The document, our fundamental in W-l. Window 4 (W-4) identifies Refd and provides its general reference source unit, can have windows to any other docu- )r the file ments. The evolving corpus is continually as the reference section at the end of !he document; a user can jump from Ihe link I (depth) expandable wifhoul fundamental change. citation in W-3 IO see Ihis statement by using the jump name command. To pcr- the num- h’cw links and windows can continually add form this command, he clicks on “Ref-6” in W-3 then clicks on W-4. Window 5 -:vel, and new access pathways M old mPerial. Fast (W-S) shows a view in the OAD-Journal Item 2250. The user can obtain this view proprierary algorithm? render the extreme igh-level by performing a jump link command on the indirect link of W-3. To perform this dala fragmentalion tolerable in the planned displayed back-end service facility.’ command, the user clicks on lhe indirect link of W-3 and then clicks in W-S. fied con- mcept of The long range goal of the Xanadu project has been facilitating the revolutionary process of placing the cntirc world’s liter- tainablc, and that a system for the clcc- The back end of the Xanadu system has ary corpus on line. In fact, Xanadu’s Ironic accounting and distribution of been implemented in Unix and is available design makes a strong separation between royalties is in place. Nelson predicts that in several forms, including as an on-lint Ihc user imcrfncc nnd the dntnbnsc scrvcr. the advent of on-line libraries will crcatc scrvicc (much like Engclbnrt’s Augnicn~). with most of the emphasis placed on the a whole new market for the organization A crude front end for the Xanadu system latter. In particular, great care has been and indexing of this immense information is also available which runs on Sun work- taken that copyright protection is main- store. stations.

:PUTER September l9S7 i . . Trigg’s Textnef. Randall Trigg wrot; ticipant is able lo understand the whole the first and to date the only PhD thesis on problem better. 776s process enables a hypertext. In his thesis, he describes his common understanding of the major Textnet system as supporting nonlinear issues and their implications to emerge. ~exl-text in which documents are order as if he were reading a linear IBIS is designed to support this organized’as “primitive pieces of text con- document. design/planning conversation. nected with typed links to form a network Trigg joined Xerox PARC after com- IBIS systems are thus a marriage of (1) similar in many ways to a semantic net.” pleting his thesis and was one of the prin- teleconferencing systems which enable The thesis focuses on specific link types cipal architects of the Xerox NoteCards many people to participate in one conver- that support literary criticism. system. sation, and (2) hypertext, which allows In the tradition of the field, frigg’s sys- participants to move easily between differ- tem is just a first step in the direction of his Problem exploration systems. These are ent issues and the different threads of vision: highly interactive systems which provide argument on the same issue. The current In our view, the logical and inevitable result rapid response to a small collection of version of Rittel’s IBIS runs on an Apple [of the comp\.ctr revolution] will be the tram- specialized commands for the manipula- PC and is being ported to Sun worksta- fer of all such (text handling] activities to the tion of information. They can be thought tions.* IBIS has three types of nodes computer, transforming communication of as the early prototypes of electronic (issues, posifions, and arguments), and within the scientific community. All paper writing, critiquing, and refereeing will be pcr- spreadsheets for text and symbolic uses nine types of relations to link these formed on line. Rather than having to track processing. One important feature of most nodes. In a typical application, someone down litllc-known proceedings, journals or of these tools is a facility for suppressing posts an issue; then that person or others unpublished technical reports from distant detail at various levels specified by the post positions about that issue; and then universities, users will find them stored in one large distributed computerized national user. For example, the outline processors the positions are argued using argument paper network. New papers will be written all have single keystroke commands for nodes. Of course, any of the three types of using the network, often collaborated on by turning on and off the display of subsec- nodes can be the seed of a new issue. (See multiple aurhors. and submitred IO on-line tions of a section. This is an unusual but Figure 5.) The current set of relationships electronic journals.‘ natural facility. Hypertext and similar between nodes is: responds-to, quesrions. Textnet implements two basic types of tools excel at the collection of large supports, objects-to, specializes, general- nodes: those which have textual content amounts of relatively unstructured infor- izes, refers-to, and replaces. Theresearch (chunks) and those which hierarchically mation. But such cohections are of little on IBIS concentrates 0~1 ways to summa- organize other nodes (foes, for “table of use unless adequate mechanisms exist for rize and present the issue network, both contents”). Thus Textnet supports both filtering, organizing, and browsing. These for participants and decision makers. hierarchical trees (via the tot nodes) and are the primary desiderata of these author- nonhierarchical graphs (via the typed ing/thinking/programming systems. Lowe’s SYNVIEW. David Lowe’s links). SYNVIEW system is similar in concept to Trigg further proposes a specific tax- Issue-Based Informorion Systems. Rittel’s IBIS but goes in a different direc- onomy of link types for use by collabora- Horst Rittel and his students have intro- tion. It proposes that the participants, in tors and critics in Textnet. He argues that duced the notion of issue-Based Informa- addition to posting their own issues and there is generally a specific set of types of tion Systems (IBIS)’ to handle systems arguments, assess previous postings as to comments, and that there is a link type for analysis in the face of “wicked problems.” their validity and relevance. The asscss- each comment. For example, there are Rittel describes wicked problems (as ment is done by a kind of quantitative vot- re_futalion and supper/ links, and, more opposed IO “tame” ones) as problems ing. For example, if you think that Joe’s specifically, thcrc arc links to say that a which cannot be solved by the traditional rcsponsc to Sam makes a good point but point is irrelevant (“Pt-irrelevant”), that systems analysis approach (that is, (1) is not really a direct response to Sam’s data cited is inadequate (“D- define the problem, (2) collect data, (3) posting, you might grade it “5, 1” (where inadequate”), or that the styleis rambling analyze the data, (4) construct a solution). 5 is a high validity rating and 1 is a low rel- (“S-rambling”). Trigg describes over 80 Wicked problems lack a definitive formu- evance rating). These values are averaged such link types and argues that the disad- lation; their problem space cannot be into the existing values for that posting. vantage of having a limited set of link types mapped out without understanding the The various displays of the argument is outweighed by the possibility of special- solution elements; in short, the only way structure show the values for each posting, ized processing on the hyperdocument to really understand a wicked problem is allowing readers to focus, if they choose afforded by a definite and fixed set of to solve it. Wicked problems have no stop- to, on those argument trails having the primitives. ping rule. The design or planning activity highest voted validity. In addition, Textnet sworts the defi- stops for considerations that are external Thro@h debate>on the accuracy of informa- nition of pafhs-ordered lists of nodes- to the problem (for example, lack of time, tion and on aspccls of rhc structures thcm- used to browse linear concatenations of money, or patience). Solutions to wicked selves, a large number of users can text and to dump such scans to hard copy. problems are not “right” or “wrong”; cooperatively rank al1 available irems of information in lcrms of significance and rcl- The path facility relieves the hypertext they just have degrees of sufficiency. Rit- cvancc IO cxh topic. Individual users can rcadcr from having to make an n-~aydcci- tcl argues that solving wicked problems lhcn choose the depth IO which ~hcy wish IO sion at each link; rarhcr, the rcadcr is requires all those involved IO exchange and csaminc [hcsc s~ruc~urcs for IIIC purposes 31 provided a default pathway through the argue their many viewpoints, ideas, ‘A graphical Sun version. allcd glL3lS. is also being network (or part of the network), and can values, and concerns. By coming IO under- dewloped ;I, the hlCC/Softwnrc Tcchnolog) simply read the material in the suggested stand other viewpoints bcrter. each par- I’rogrLJm.

24 COMPUTER whole hand. The function of this debate is not IO b 2.3 arrive LI specific conclusions, but rather IO 2 collect and order the bcsl available evidence J- on each topic.‘ n, txs UNC’s WE. A group at the Unjversity of North Carolina at Chapel Hill has been of(l) developing a wrifing environmen! called :nable WE.9 Their research is based on a cogni- jnver- tive model of the communication process 3llows which explains reading as the process of Jiffer- taking the linear stream of text, compre- ids of hending it by structuring the concepts hier- .Irrent archically, and absorbing it into long-term memory as a nefwork. Writing is seen as Apple Generalizes Resl bnds-to rksta- the reverse process: A loosely structured I \ nodes network of internal ideas and external 1, and sources is first organized into an appropri- these ate hierarchy (an outline) which is then Position 2 ncone “encoded” into a linear stream of words, )t hers sentences, etc. I then WE is designed to support the upstream Ybjects-to tmcnt part of writing. It contains two major view pes of windows, one graphical and one hierarchi- . (See cal, and many specialized comman_ds for Iships moving and structuring material (nodes (ions, and links with attached text) between these Teral- two views. Normally a writer will begin by tarch creating nodes in the graph view, where he nm- can place them anywhere within the win- k dow. At this stage, little or no structure is ‘5. ._ imposed on the conceptual material. The writer can place nodes in “piles” if they zwe’s seem to be related, or he can place individ- :pt to ual nodes between two piles if they are ii:ec- somewhat related to both. As some con- Figure 5. A segment of a possible IBIS-style discussion showing the topology of the ts. in ceptuzl structure begins to emerge from IBIS network. Each node contains information on the type of the node, the time c and this process, the writer can copy nodes into and date of creation, the author, a short phrase describing the content, a longer as to the hierarchy window, which has spccial- body of text with the text of !hc comment, a list of keywords, nnd a list of the &Xs- ized commands for tree operations. The incoming and outgoing links. zvot- hierarchy window has four different dis- Joe’s play modes: (1) the tree can be laid out on I but its side, with the root node on the left; (2) come with further experiments and opening up just those entries that are rele- am’s the tree can be hung vertically with the root analysis. vant or useful to the idea that he is work- ,here at the top; (3) child nodes can be displayed ing on. In addition, each outline entry can \ rel- inside their parent node; and, (4) the hier- Outlineprocessors. An outlineproces- have a textual body of any length ased archy can be shown in the traditional out- sor is a word processing program which is associated with it, and the user can make ring. line view. specialized for processing outlines, in that this body appear or disappear with a sin- ncnt WE uses a relational database for the its main commands deal with movement gle keystroke. This feature is a real boon :,ng, storage of the nodes and links in the net- among, creation of, and modification of to the writing process, becauseit allows the >ose work. The user points with a mouse to outline entries. In this respect, thesepro- %ser IO have a view of boththe immediate the select a node. A third window is an editor grams commercialize many ideas from text that he is composing and the global for the material within the currently Engelbart’s NLS/Augment. Outline context for it. It also facilitates rapid selected node. A fourth window on the processors also have at least simple text movement between sections, particularly screen is for queries to the database. A editors and do some text formatting, so in large documents, because in outline fifth window is used to control system that the user can use the same tool to go mode a rcmotc section is ncvcr more than modes and the current working set of from outline to finished document. One of a few keystrokes a\vay. nodes. the most powerful features of outline Most outline processors are personal WE is designed to be an experimental processing is the ability to suppress lower computer programs, and they have done platform to study what tools and facilities levels of detail in the outline. As with much IO bring some of rhe concepts undcr- will be useful in a writer’s environment. Engelbart’s NLS/Augment, the user can lying hypertext into popularity. The first The real validation of these ideas, as with view just the top level of the outline, or the of these was called ThinkTank. II was so many of the systems described here, will top n levels, or he can “walk the tree,” released in 1984. It has since been joiner’

‘ER Septcmbcr 19S7 . -. . ,. _..

by a host of others, with names like Max- short commands typed at the keyboard. It Think, Executive Writer/Executive Filer. is primarily hierarchical, but a user can Thor, Framework, Kamas, Fact jump to a different place in the hierarchy Cruncher, Freestyle, Idea!, and PC- by typing in the name of the destination Outlinc.‘“There are two very recent addi- node. It is used as an on-line help system tions to the field: Houdini is an extension in Emacs. INFO has the same potential for of MaxThink that supports rich nonhicr- called Knowledge Management System user disorientation which is shared by all archical internode references; and For- (KMS). of the systems which display only a single Comment is a word processor that allows Each segment of the ZOG/KMS data- frame at a time and have no browser. up to 15 people to apply hypertext-like base is called a frume. A frame has, by annotations to a document (and can opcr- convention, a one-line title at the top of the Shneiderman 3 Hype&s. The Univer- ate over a Local Area Network (LAN) in screen, a few lines of text below the title sity of Maryland Hyperties project’ has real time). staring the issue or topicof the frame, asct been developed in two directions-as a Aside from Houdini, most outline of numbered (or lettered) menu items of practical and easy-to-learn tool for brows- processors do not support inter-entry text called selecfions, and a line of standing in instructional databases and as an references, except by “cloning” the whole ard ZOG commands called globolpadsat experimental platform for studies on the entry and displaying it in the new location. the bottom of the screen. (Some of these design of hypertext interfaces. As a prac- Only a few others provide windows for commands are: edif, help. back. nexl, tical tool, it has already seen some use in nodes. None of them provide explicit mark, refurn, and commenf.) The selec- the field at a Washington, D.C. museum “mousable” link icons. For these reasons, tions interconnect the frames. When a user exhibit about Austria and the Holocaust. one could argue whether they qualify as selects an item by typing its number or let- (See Figure 6.) Designers of the exhibit hypertext as I have defined it here. How- ter at the terminal keyboard, the selected emphasized making the system easy and ever, ThinkTank was the first program to frame appears on thescreen, replacing the fun for users who have never used a com- be billed-somewhat pretentiously--as an previous frame. The structure is generally puter before. As an experimental plat- “idea processor,” and all of these pro- hierarchical, though cross-referencing form, it has been used in tive experimental grams treat sections of text as first-class links can be included. In addition, an item studies involving over 228 subjects.” objects and support manipulations that in a frame can be used to activate a In Hyperties the basic units are short coincide with the way one manages ideas. process. articles (SO-1000 words typically), which They share these features with hypertext, In 1982 ZOG was installed and used as are interconnected by any number of links. and in this sense, they anticipate the a computer-based information manage- The links are highlighted words or phrases inevitable proliferation of hypertext fea- ment system on the nuclear-powered air- in the article text. The user activates the tures within the mainstream of computer craft carrier USS CARL VINSON. This links by touching them with a finger (on a applications. system is probably the largest and most touch-sensitivescreen) or using the arrow thoroughly tested hypertext system in sew- keys to jump to them. **Activating a link Structured browsing systems. The sys- ice in the field. ZOG has also been used for causes the article about that topic to tcms rcvicwcd in this section wcrc dcsigncd more interactive process applications such appear in its own window on the screen. primarily for applications involving large as policy analysis, authoring, communica- The system keeps track of the user’s path amounts of existing informarion or requir- tions, and code management. Historically, through the network of articles, allowing ing easy access IO information. These sys- however, ZOG made its name more as a easy return from exploratory side paths. tems pose different problems for their bulletin board/textual database/CA1 tool In addition to a title and a body of text, designers. Ease of learning and ease of use than as an interactive system. Hence it is each article has a short (5- IO 25-word) are paramount, and great care goes into included in this section on browsing. A description which the program can display crafting the interface. On the other hand, drawback of the ZOC/KMS style of view- very quickly. This feature allows the user writing (adding new information) is ing a single frame at a time is that users an intermediate position between bringing usually either not allowed to the casual may become disoriented, since no spatial up the full article and trying to guess from user or is not particularly well supported. event corresponds to the process of mov- the link name precisely what the article is ing from frame to frame. In the KMS sys- about. CMU’s ZOG and Knowledge Sysfems’ tem, this tendency has been offset by Hyperties runs on the IBM PC. KMS. ZOG is a menu-based display sys- minimizing system response time, so that Recently graphics capabilities have been tem developed in 1972 at Carnegie-hlellon frame-to-frame transition takes about half added to the system. Current implementa- University.” It consists of a potentially a second. Thepossibility of user disorien- tion efforts focus on support for videodisc largedatabase of small (screen-sized) seg- tatio& greatlyreduced by the fact that the images. Also. a browser is being developed ments which are viewed one at a time. user can move very quickly among frames ZOG was developed with the particular and thus become reoriented with very lit- 7he “tics’* in “Hypcrtics” stands for “The In~erx- goal of serving a large simultaneous user tle effort. Creating text and graphics is also live Encyclopedia System.” community, and thus was designed IO fast in KMS. operate on standard terminals on a large timesharing system. In 1981 two of the Emacs INFOSubsysrem. The help system principals on the ZOG Project, Donald in the widely used text editor, Emacs, is McCracken and Robert Akscyn, started called INFO. and is much like ZOG. It has the company Knowlcdgc Sysrcms nnd a sirnplcr set of standard commands, 31x4 devclopcd a commercial successor to ZOG its control input is done by single letters or

26 COMPUTER xd. It which will provide string search, book- :r can marks, multiple windows, and user annotation. 7:“: n Symbolics Docurnenf E&miner. The L3r most advanced of [he on-line help systems, ;:.G: lZ?e.choslouakia by aill ihis tool displays the pages from the entire single twelve-volume manual set on the Sym- cr. bolics Lisp machine screen.” Certain textual fields in the document (printed in niver- bold) are mouse-sensitive. Touching one .’ has of these fields with the mouse causes the -as a relevant section of the manual to be added rows- to the current working set of manual as an pages. The system allows the reader to In the place bookmarks on any topic and to move prac- swiftly between bookmarked topics. The Jse in protocol for link following is tailored to scum browsing in a refcrencc manual or cncy- :aust. clopedia. Mousing a link only causes it 10 thibit be placed on a list of current topics. Then, y and mousing an entry in this list causes that com- link to be followed, bringing up the refer- plat- enced ropic in the main viewing window. lental The system also supports on-line string 12 - search of preidentified keywords, includ- PIACES: AUSTRlA PAGE 1 OF 3 short ing the search for whole words, leading Jhich subsrrings, and embedded substrings. The Austria (see map) holds a special place in the history of the Holocaust. / inks. system is thus well designed for the specific rases task of browsing through a technical man- Situated between Eastern and Western Europe, possessing a vibrant and .c ual and pursuing several aspects of a tech- culturally creative Jewish community on the eve of World War II. 1 nical question or several levels of detail &.- simultaneously. The user cannot makeany Austria had also provided the young Adolf Hitler, himself an Austrian link changes or additions to the manual set c to (although it is possible to save personalized raised near Linz. with impor’ant lessons in the political uses of 'CC.l. collections of bookmarks). path antisemitism Leading Nazis came from Austria: the names of Adolf .ving General hgpcrtcxt technology. So far I HiUer, Adolf Eichmann, wno organized the deporlations of the Jews to shs. have discussed hypertext sysrems that have ext, particular practical applicarions. The fol- the death camps, and Ernst Kaltenbrunner. the head of the ord) lowing systems also have one or more 2lay applications, but their primary purpose is Reich Main Office for Security, 1943-45. readily cOme to mind. As user experimentation with hypertext itself as a ~~_~_.______~~______~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ zing technology. For example, while Linz - city in northern Austria; childhood home of Adolf Hitler and other iOIll NoteCards has been used for authoring, le is programming, personal information man- lea&n9 Nazis -rgement, project management, legal ‘C. research, engineering design, and CAI, its NEXT PAGE RETURN TO GYPSIES INDEX .een developers view it primarily as a research %a- vehicle for the study of hypertext. disc .--,. - oed Xerox PARC’s NoteCards. Perhaps the Figure 6. The Hyperties Browser enables users to traverse a database of articles and best known version of full hypertextis the piclures by selecting from highlighted items embedded in the text of the arlicles. J’oreCurds system developed at Xerox The photos show the IBM PC version of Hyperties. The upper node shows a map Pr\RC.” The original motivation in of Austria. The lower node shows double-spaced text with link terms highlighted. building NotcCards MS IO dcvclop nn Either a touchscrccn or jump-arrow keys arc usrd for selection of hrirf dcfinitiolls, information analyst’s support tool, one full articles, or pictures. The Hypertics Author permits pcoplc with onl! word Ihat would help gather information about processing skills to create and maintain databases. Research versions of Hyperlies a topic and produce analytic reports. The run on the Enhanced Graphics Adapter to give more lines and multiple windows designers of Notccards observed rhat an and on the Sun 3 \vorkstation to show two full pages of 1~x1 31 a time. Current information analyst usually follows a development efforts will enable rcadcrs to point at pictures and videodisc images IO general procedure rhar consists of a series retrieve further information.

September 1987 NoteCards IIEC?/

,... .,;.. i. _.:..j:. :; :;:: i.;: i:., j:: :,:j . . . :.;c>:. .:$‘.::.\.‘..A‘.I. - Lsc’r 1OLIll .,c : ::. :,,_ :,: .:; ‘.

‘. :j . ;. :

‘: !: : 1..,: .: ! ,: :, ;: . . .: : :, \ ,:: . . . . L

Figure 7. A typical NotcCards screen with five FileBox cards, two unformatted Test cards, and one Text card formatted as a [able. Links bet\+een cards are represented by the boxed test inside the cards. The two menus at Ihe top/middle of the screen control two different note files. The remainder of the icons on the screen belong to non-NoteCard applications running in the Xerox Lisp environment.

ofsteps: (1) reading sources (news reports, interface, the user can easily customizethe screens allowing windows and link and scholarly articles, etc.), (2) collecting clip- browser. NoteCards allows easy creation node icons to be displayed in very high pings and filing them (in actual shoe- of new types of nodes. Forty or fifty such resolution. (See Figure 7.) Currently boxes!), and (3) writing analytic reports. specialized node types have been created between 50 and 100 users use NoteCards, The-designee also observed that through- to date, including text, video, animation, many of them outside df Xer&even outThe process, the analyst forms analyses graphics, and actions. l The new version though it is not a supported product). and conceptual models in his head. The also allows several users IO work in the Several of these users have constructed research goal of the PARC team was to same Notefile at the same time. very large databases in the system (for develop technology to aid the analyst in Part of NoteCards’ success is due to the esample, 1600 nodes with 3500 links forming better conceptual models and fact that it was dcvclopcd on Xerox D- bctwccn them). annl\xcs, ntld to find bcrrcr cxprcssions oi scrics Lisp machines, which 3rc powerful Ihcse models and analysts. workstations that have high resolution Brown L'rriversily ‘s fntermedia. One of A programmer’s interface makes the oldest and largest hypcrtesr research JoteCards an open architecture that groups esists at Brown University, at the allo\rs users to build (in Lisp) new appii- Institute for Rcscarch in Information and cations on rap of NoteCards. Using this Scholarship, (IRIS).” The lnrermcdia

28 COVPUTER .I , _ ::.. _

project builds on two decades of work and pk. ic can show whole documents and the [hrcc prior gcncrations Of hypcrtcxt links between them, or each link and its systems.“ approximate localion within its docu- The first system was the Hypertext Edit- ments. (See Figure 8.) ing System designed by Ted Nelson, Andy The Intermedia project has a long his- For example, in the English lireraiurecourre van Dam, and several Brown students for tory, many participants, and a se_rious the firsr time a student is searching for back- ihC IBM 2250 display in 1968. This system ground informationon Alexander Pope, he institutional commitment to long-term was used by the Houston Manned Space- or she may be interested in Pope’s life and the objectives. It conducts creative hypertext craft Center to produce Apollo documen- political events that prompted his satiric criti- esperiments and uses the classroom as a tation. cism. To pursue this line of thought the stu- proving ground. Although this project is The second system was the File Retrieval dent might retrieve the biography of Pope and a timeline summarizing political evenis sIill in its early’stages, we can expect it to and Editing System (FRESS). FRESS was . taking place in England during Pope’s life. contribute significantly to the develop- a greatly enhanced multiterminal Subsequently, thestudent may want tocom- ment of effective cooperative work envi- timesharing version designed by van Dam pare Pope’s use of satire with other later ronments based on hypertext. and his students. It became available in authors’satiric techniques. This time the stu- 1969 and was commercially reim- dent may look at the same information about Pope but juxtapose it with information about Teh-monkNeptune. Tektronix Neptune plemented by Phillips in the early 1970’s. other satiricists instead of a time line. The is one hypertext system that has been par- FRESS was used in production by instructor (and other students. if permitted) ticularly designed as an open, layered hundreds of faculty and students over could read the student’s paper. examine the architecture.” Neptune strongly separates more than a decade. Its users included an refcrencc material, and add personal annotation links such as comments, criticism, and the front end, a Smalltalk-based user inter- English poetry class that did all of its read- suggestions for revision. While revising the face, from the back end, a transaction- ing and writing on a communal hypertext document, the student could see all of the based server called the Hypertext Abstract document. Like NLS, FRESS featured instructor’s comments and examine the .\lachine (HAM). The HAM is a generic both dynamic hierarchy and bidirectional sources containing the counter-arguments. hypertext model which provides opera- reference links, and keyworded links and Like most of the serious workers on tions for creating, modifying, and access- nodes. Unlike NLS, it imposed no limits hypertext, the Intermedia team is espe- ing nodes and links. It maintains a on the sizes of nodes. On graphics termi- cially concerned with providing the user complete version histor!* of each node in nals, multiple windows and vector with ways of managing the increased com- the hyperdocument, and provides rapid graphics were supported. plexity of the hypertext environment. For access to any version of a hyperdocument. The third project, the Electronic Docu- example, they contend that multiple links It provides distributed access over a comment system, was a hypermedia system emanating from the same point in a docu- puter network, synchronization for mul- emphasizing color raster graphics and ment may confuse the reader. Their alter- tiuser access, a complex network navigation aids. native is to have a single link icon in the versioning scheme, and transaction-based As part of Brown’s overall effort to material (text or graphics) which can be crash recovery_. bring graphics-based workstations into quickly queried via the mouse to show the The inte-face layer provides several effective* use within the classroom, the specific outgoing links, their names, and browsers: A graph browser provides a pic- Intermedia system is being developed as a their destination nodes.” They also pro- torial view ;lf a subgraph of nodes and framewtjrk for a collection of tools that pose a construct called a )veb IO implement links; a docanlenl brotc,ser supports the - allow authors 10 crcatc links to documents context-dependent link display. Every link browsing of hierarchical structures of a of various media such as text, timelines, belongs to one or more webs and is only nodes and links; and a node browser 1 diagrams and other computer-generated visible when one of those webs is active. To accesses an individual node in a hyper- le images, video documentaries, and music. view documents with the links that belong document. Other browsers include oftrib- Two courses, one on cell biology and one to a particular web, a user opens a web and ure browsers, version browsers, node on English literature, have been taught then opens one or more of its documents. dij/erences browsers, and demon using the system. Current applications Although other webs may also reference browsers. (See Figure 9.) include InterText, a text processor; the document, only the links which were In Neptune, each end of a link has an Jnd InterDraw, a graphics editor; Interval, a made in the current web are displayed. As oiiset within its node, whether that node

igh timeline editor that allows users interac- a result, the user does not have to sift is textual or graphical. l The link attach- :tly tively to organize information in time and through the connections made in many ment may refer to a particular version of ds, date sequences; Int_erSpec, a viewer for different contexts. a node, or it may refer to.the current ver- -/en sections of 3D objects; and InterPix, a The IntermediGproject?s also studying sion. The HAM provides IWO mechanisms :I). scanned-image viewer. Under develop- ways of providing an effective bro\vser for that are useful for building application :ed ment are a video editor, a 2D animation a network that can include hundreds or layers: Nodes and links may have an *‘or editor, and more complex methods for even thousands of nodes. The Intermedia un!imited number ofattribute/value pairs; AS filtering the corpus and creating and browser has two kinds of displays: a global ar,d special high-speed predicates are traversing trails. ~nup, which shows the entire hypcrdocu- included for querying the values of these Intermedia is being developed both as a ment and allows navigation within it; and prjrs in theentire hyperdocument, allo\\- of tool for professors to organize and present a local map, which presents a view cen- _.’ their lesson material via computer and as tered on a single document and displaying he ‘- an interactive medium for students to its links and nearest neighbors in the web. Id study the matcrials and add their own In addition. a display can show nodes and ia annotations and reports. links at several levels of detail. For exam-

R Scptcmbcr 1987 29 j%specks =(1859)-

Contemporary Thought: Oarkin’s On tie Origin of *ties by Means of Natural Sekchfl, of tie Preservation of Favoured Races in-tie Sruggk for We< made several points thal hat I major Impact on nlnetrenlh-century lhoupht (1) That biOlOQlCal types or species do not have a ked, Slatit existence but edsl in permzmcnt states ol change and rlbq

Figure 8. The Intermedia System. This figure illustrates materials from an Infermedia corpus catled “Bio 106: Cell Biology in Conlexl.” Three folder windows containing hierarchically organized documents of different types are open in fhc upper left side of the display. An IntcrText document (lower let1 side) and an InterVal document (upper right side) are currently open, as well as an IntcrDraw documcnl containing a scanned electronmicrograph (lower middle). This image has been linked to a corresponding three-dimensional image displayed in an InterSpect document (lower right). The “lower tracking map” (center) shows the links emanaling from the current document. Authors or browsers can manipulate the three-dimensional imnge, edit lexl and graphics, follow links or create links at any time in this environmenf. (The electronmicrograph of Micromonas was published in the Joltrnol oJPhycology and is reprinted with the permission of the Editor.)

ing higher level applications 10 define their example, a program is a box that contains A to box B appears within A as a bos own accessing mechanisms on the graph. some boxes that provide input and output whichsh6ws B. But a port is more than The HAM also provides a demon mecha- variables, and other b&es that specify just a view of rhe destination box, because nism chat invokes arbitrary code when a behavior. The system also supports alter- the destination box can be changed specific HAM event occurs. nate views of some boxes: A box which through any of the porls which lead to it, specifies a graphics routine can also show and the changes willbe reflected in all of diSes.w’s Boxer. Boxer” is a highly that graphic display. these ports. interacIive programming languagespecif- Since Boxer is a programming language, Hicrsrchy is more nalurally expressed in icall! tailored IO be easy for noncomputcr it Ire315 cross-reference links in a special Boxer than in many of the o:her hypertext specialists co learn. Boxer uses a box to tvay. Rather than using mousable icons as systems. Boxes are nested within each rcpresenr a unit of information in the sys- links, Boxer uses a specialized box, called other Iwo-dimensionally. and arc fillcrcd tem. In Boxer, one box cm contain other a “port,” \vhich giws a direct view imo rhc 10 reduce rhc level of clutter on the screen. - boxes, or data such as ICXI or graphics. For desrination. For example, a port from box This system oi representation has the

30 COMPUTER

- NCDtUnt OOCUmCnU: Crwh IrowIer

NeDtUne Documl

Context3 becificatio CSW Paper

[Kc6861 KdQ, R.H.. chang. E. and Bhatela, R. Venlon modeling concepts for MufUmedla content computer-tided derlgn dtiua. Proc. ACM SlCMOD ‘86. (May 1986) 379-386. - The name hypertext Is actually a misnomer for many of IMeytLl Meyrow~ N several Wtems. lncludlng Augment, Xanad , . Intermedla: The Archttecture and ConsQuctlon of an and the Oectronlc Document Syste NeV-unc+md Object-Oriented HypenIJCdla System and Clgplicationr Framework. contents of a node to text. In general the content of a Proc. KM OOPSLA’86, (Nov 1986) 186-201. can be tikrtrary dlgltal data whose Interpretadon may )nc or digked speech. [NclKl~~~ Nelson, T.H. Mrrory Mathfnrs. T.H. Nelson. SwarKhmore. PA.. 1981.

CRMN11 I Robertson. C., McOacken. 0. and Newell, A. The 2OC approach to man-machlne communication. lnrrrnoclonol)ournolof Mon-Uocblnt StudIts. 14, 461-488. 1981.

in Figure 9. Neptune browsers. Three browsers from Neptune are illustrated. A pictorial view of a network of nodes and links is It shown in the Graph Browser (the upper window). The lower righf window and the lower pane of the Document Browser arc 1, as viewers for lest nodes. Icons reprcscnfing link affachmrnfs are shown embedded wifhin fhc fexf in each of fhe nodes. :or- ) gdit ts

advantage of showing a natural hierarchy tigate more general hypertext design /ions, which can be defined implicitly b) of nodes: The windows of lower-level issues. I9 Much of the interactive feel of a predicate (called an absrracr collenion) ! box nodes are nested directly within their par- CREF reflects the style of use and pro- or explicitly by a list (called a sroric collec- than ents. In;most hypertext systems, no gramming of the Symbol& Lisp machine, lion). At any rime, the selectedcollectiijh lause attempt is made to display the parent-child on which it was built. Chunks of text. appears as a continuous length of text with mged relationship once the nodes are opened as called segments, constitute the nodes in the the segment boundaries marked by named to it, windows. system. Segments are arranged in linear horizontal lines (such as “Segment 1,” aLI of series, and can have key\vords and various “Segment 2.” etc.). This view can be Pirman’s CREF. The Cross-Referenced kinds of links to other segments. The edited as if it were a sin& document. cd in Editing Facility (CREF) is a prototype of notion of a linear set of segments is naru- One way of forming an abstract collec- Ti f a specialized text and graphics editor ral to the protocol analysis problem, since tion is by selecting segments using a

et which was developed originally as a tool the first srcp with such protocols is to seg- boolean predicate over keywords. To .crfb- for use in analyzing the transcripts from mcnt them into the cpisodcs of the cxpcr- cslcntl lhc power of this keyword f;icilily. .--cccn , psychological cspcrimcnts (known aspro- imcntal session. CREF allows the user to dcrinc a type hicr- the rocols), but which was also used to inves- CREF organizes segments into co//ec- archy on the keywords. For example, if

TER September 1987 31 I . . .

“card 105” is defined as a type of (i.e., a features of the experimental hypertext childof) “card,” then collections based on technologies. Indeed, this area of invcsti- the keyword “card” will also contain seg- gation may become an important fifth cat- ments which have only “card 105” as a egory for hypertext systems of the future. keyword. special support for txewablr links. which The history of hypertext presented here CREF supports four kinds of links: will give it the flavor of a programming suggests that the concept and the advan- rejerences links cross-reference among language. HyperCards will be bundled tages of hypertext were clear several segments; summarizes links impose hier- with the system software in new Macin- decades ago, but that widespread interest archy (a summary is a segment which has toshes. in hypertext was delayed until the support- one or more summarizes links to other seg ing technology was cheap and readily nents); supersedes links implement ver- MCC’s PlaneText. PlaneText, devcl- available. This suggestion may be mislead- sioning by copying the superseded segment oped in the MCC Software Technology ing. Many of the “elders” of the field feel and freezing it; and precedes links riace a Program (STP), is a very recent addition that something else has changed as well. linear ordering on segments. to the family of general hypertext sys- They feel that today computer users eas- Finally, CREF allows multiple analysts tems. l PlaneText is based on the Unix file ily accept the role of the computer as a tool to compose different 0zeorierabout a pro- system and the Sun SunView window for processing ideas, words, and symbols tocol, using the same segmented data. manager. Each node is a Unix file. Links (in addition lo numbers and mere data), Each theory imposts its own structure on appear as names in curly brackets (I}) and as a vchiclc of intcrhuman communi- the data, and has its own collections, dia- whose display can be turned on and off. cation. Those theorists who gave presen- grams, keywords, and annotations. This Links are implemented as pointers saved tations of their hypertext systems 20 years mode of selection is similar to the notion in separate files, so that the linked files ago, using expensive state of the art hard- of contexts or webs used in other systems. themselves are not changed by creating ware, report that the computer science hypertext references between them. This community showed little interest. This Hyperfexl on rhe Mucintosh. At least design allows for the smooth integration lack of interest seemed to stem as much two programs have been written for the of hypertext into the rest of the Unix-based from a lack of understanding of the basic Apple Macintosh that provide hypertext computational environment, including concepts of hypertext as from a lack of facilities: FileVision and Guide. such tools as Mail and News. It allows for hardware resources. File\‘ision is primarily oriented to the hypertext annotation of standard If this is so, then the recent upsurge in graphics nodes and to applications which source code files. In addition, the Unix file interest in hypertext may signal that the can exploit visual indexing. The advertis- directory system serves as a “free” mech- computer community is now ready IO con- ing for FileVision describes applications anism for creating hierarchical structures sider its technology as much a tool for that encourage visual indexing. For exam- among nodes. l * communication and augmenting the ple, in the database for a travel agency, the PlaneText supports color graphics human intellect as for analysis and infor- map of a region may contain icons for the nodes which can be freely linked into a mation processing. Hypertext is certainly main-cities in that region. The user clicks hyperdocument. a large step in that direction. on the icon for a city to obtain a display of a map of that city. The map of the city may Summary. The systems in this section have icons for the major landmarks in the were presented in terms of four broad cat- city. The user clicks on one of these icons egories: macro literary systems, problem The Essence of to obtain a display of data about the land- exploration systems, structured browsing Hypertext mark, or perhaps even to obtain a picture systems, and general hypertext technol- of the landmark itself. ogy. Table 1 summarizes this discussion It is tempting to describe the essence of Guide is a more recent program which and provides a breakdown of the various hypertext as its ability to perform high- is based on an earlier Unix version devel- features which current hypertext systems speed, branching transactions on textual oped in England.” It does not provide the can include. chunks. But this is a little like describing graphics capabilities of FileVision One additional area of research cur- the essence of a great meal by listing its (graphics are supported but cannot con- rently is the development of systems which ingredients. Perhaps a better description tain links), but it does support textual’ aid the entire process of design, particu- would focus on hypertext as a computer- hypertext data very well. Guide uses three larly the informal upstream aspects. Such based medium for thinking and commu- kinds of links: replacemenl /inks, which systems require the features of hypertext nication. cause the text in the current window to be problem exploration and structured - The tinking process does not build new completely rep&d by the text pointed to browsing systems as well as the advanced -ideas one at a time, starting with nothing by the link; note links, which display the and turning out each idea as a finished destination text in a pop-up window; and -11 is perhaps loo early IO say. however. how pearl. Thinking seems rather to proceed on reference links, which bring up a ne\y win- P4rneTc.u will rank in Ihe world of hypcrlexl, since ;I several fronts at one, developing and will only be publicly available rrom the pjrlicipanl dow with the destination text. Guide is companies in the hICC;Sof~warc Technology Pro- rejecting ideas at different lcvcls and on now available for PCs as well. gram. For mote informalion call Bill SwLcrbcry x different points in parallel, each idea As this article goes to press, there is news hlCC. (512) 343-097s. depending on and contributing to the that Apple will soon have its own hyper- ‘7hc use of an existing Ircc-slruc!uring mechanism others. text system, called HypcrCards. Hyper- limits any hypwexl sy$lcm IO only being able IO ban- The recording and communication of dlc J single hwarchical vrwurc. Single hicrarrhisal Cards will bc similar in some ways to organirnllonr may be too limucd for some advanced such cntwir!cd lines of thought is challcng- Xerox PARC’S NotcCards. It will provide applicalions. ing because communication is in practice

32 COhlPUTER . -. _’ .’ ;-;.

.I “. . -:- ,: ertext . a serial process and is, in any case, limited However, not all link traversals can be by the bandwidth of human linguistic instantaneous. Perhaps as important as IVCSli- ,i :h cat- processing. Spoken communication of rapid response is providing cues to the user about the possible delay that a given query ‘Jt y-.”c :, parallel themes must mark items with Jh stresses, pauses, and intonations which the than the nodes. The links form the “glue” or traversal might entail. For example, dr listener must remember as the speaker that holds the nodes together, but the some visual feature of the link icon could :vezl develops other lines of argument. Graphi- emphasis is on the contents of the nodes. indicate whether the destination node is in terest cal forms can use lists, figures, and tables From a computer science viewpoint, the memory, on the disk, somewhere else on @port- ‘to present ideas in a less than strictly lin- essence of hypertext is precisely that it is a the network, or archived off line. .adily ear form. These visual props allow the hybrid that cuts across traditional bound- ilead- reader/viewer to monitor the items which aries. Hypertext is a durubuse mqthod, Properties oJlinks. Links can be used d feel he must understand together. One of the providing a novel way of directly access- for several functions. These include the well. challenges of good writing, especially good ir,g data. This method is quite different following: i eas- technical writing, is to present several par- from the traditional use of queries. At the l They can connect a document refcr- I tool allel lines of a story or an argument in a same time, hypertext is a represenfalion encc IO the document itself. lbols way that weaves them together coherently. scheme, a kind of semantic network which l They can connect a comment or anno- ata), Traditional flat text binds us to writing mixes informal textual material with more tation to the text about which it is written. luni- and reading paragraphs in a mostly linear formal and mechanized operations and l They can provide organizational :sen- succession. There arc tricks for signaIling processes. Finally, hypertext is an inrerfuce information (for instance, establish the icars branching in the flow of thought when mod&y that features “control buttons” relationship between two pieces of text or lard- necessary: Parenthetical comments, foot- (link icons) which can be arbitrarily between a table of contents entry and its cnce notes, intersectional references (such as embedded within the content material by section). This “see Chapter 4”), bibliographic refer- the user. These are not separate applica- l They can connect two successive nuch ences, and sidebars all allow the author to tions of hypertext: They arc metaphors for pieces of text, or a piece of text and all of 2asic say “here is a related thought, in case you a functionality that is an essential union of its immediate successors. .k of are interested.” There arealso many rhe- all three. l They can connect entries in a table or torical devices for indicating that ideas figure to longer descriptions, or to other ;e in belong together as a set but are being The power of linking. In the next two tables or figures. the presented in linear sequence. But these are sections of this article, I will explore links Links can have names and types. They Jon- rough rools at best, and often do not pro- and nodes in more detail as the basic build- can have a rich set of properties. Some sys- for. vide the degree of precision or the speed ing blocks of hypertext. tems allow the display of links to be turned tt and convenience of access that we would on and off (that is, removed from the dis- for. __ like. Linkjollowing. The most distinguishing play so that the document appears as ordi- Ully Hypertext allows and even encourages characteristic of hypertext is its machine nary text). the wrirer to make such references, and support for the tracing of references. But The introduction of links into a text sys- allows the readers to make their own deci- what qualif::s a particular reference- tem means that an additional set of sions about which links to follow and in tracing device as a link? How much effort mechanisms must be added for creating is permissible on the part of a user who is what order. In this sense, hypertext eases new links, deleting links, l changing link the restrictions on the thinker and writer. attempting to trace a reference? The names or attributes, listing links, CIC. It dots not force a strict decision about accepted lower limit of refcrcncing sup- whether any given idea is either within the port can be specified as follows: To qualify Rejerentiul links. There are two sof flow of a paper’s stream of thought or out- as hypertext, a system should require no methods for explicitly linking two points gh- side of it. Hypertext also allows annota- more than a couple of keystrokes (or in hypertext-by reference and by organ- aal tions on a text to be saved separately from mouse movements) from the user to follow ization. The reference method is a nonhi- ‘ng the reference document, yet still be tightly a single link. In other words, the interface erarchical method. It uses referential links its bound to the referent. In this sense, the must provide links which act like “magic that connect points or regions in the text. on “linked-ness” of hypertext provides much buttons” to transport the user quickly and Referential links are the kind of link that “r- of its power: It is the machine processible easily to a new place in the hyper- most clearly distinguishes hypertext. They u- links which extend the text beyond the sin- document. generally have two ends, and are usually gle dimension of linear flow. Another essential characteristic of directed, although most systems support CW At the same time, some applications hypertext is the speed with which th+sys- “backward” movement along the link. ng demonstrate that the “node-ness” of tern responds to referencing requests. Only The origination of the link is called the ed hypertext is also very powerful. Particu- the briefest delay should occur (one or two “link source,” and usually acts as the Xl lariy when hypertext is used as a thinking, seconds at most). Much design work goes rejerence. The source can logically be a.d writing, or design tool, a natural cor- into this feature in most systems. One rea- either a single point or a region of tesl. At In respondence can emerge between the son for this concern is that the reader often the other end. the “destination” ofrhc link ra objects in the world and the nodes in the does not know if he wants to pursue a link usually functions as the rejpren/, and can 1e hypertext database. By taking advantage reference until he has had a cursory look of this object-oriented aspect, a hypertext at the referenced node. If making this ‘Link dclction is problematical. For example. what user can build flexible networks which judgement takes too long, the user may should the policy be lo: nodes which art slrandcd ‘f - when all their links have been dclcwd? Should they be 2 model his problem (or solution). In this become frustrated and not bother with the placed in “node limbo” umil the user dcudcs u t3l IO :e application the links arc less important hypertext links. do with them?

R September I987 33 1 . . : -- , from its neighboring material within the same node raises a tough implementation issue-how lo display the selected region to the user. It must be highlighted some- A how, using reverse video, fonts, or color, but each of these options poses difficulties in keeping overlapping regions clearly kl dlksjdf dkf a1 lclrldfj rrlkdfj rljrrd asdlkIj;l fjrlkja rl;kjd jl highlighted. The Intermedia designers pro- pose to draw a light box around regions and a darker box around region/region overlaps, thus shoying a single level of WC vklnrd; vkluv cvolcw.m k oi ionolf flkd. overlapping”; however, this technique is not effective if there are more than two sov VI voinrek voicwd. vicnm overlapping regions. aio vim”8 lvdrll vinocrn vi dinv war alkr 111 relubc rcbicr uvrcbx Another difficulty posed by link regions rhkrkb ‘Ikdrlkt,d djnvuln’ dfol is how to show the name of the link. Unlike rew cv ebnwe iubvi cvubw viu a link point, a link region has no obvious iuhrdibn dcncwin niouc dsiub - indri iourk dk d# iuwi. position for a title, unless it is placed arbitrarily at the beginning oi end of the L B region. Bgb rdkj dlkjx kcwoj vodiu od roi Link regions can also be difficult to rlkcid rdoi rionrl dsia.fl sol dir of I roijsdi sdoijrd oid8 dior of dsoi manipulate. Designers must devise a sys- rdoi ldoir rdoids droijrd doi dsoids tem for copying, moving, modifying, and I s oiordj twos sdoie dioicwnrxon oridoiw roidn. deleting the region and the substrings within it. The movement of regions Ywe WC d coid soinvcoiz zxoid di rdoia aroinv aoidrna oianaois dnoi involves logistical dilemmas which are not rcrndoia a aoindr roid oid odno easy to resolve: For example, when one roir roidnroinao o oirdo dcioiod aoiaal. moves a major portion of the text in a destination region lo someplace else in the node, should the link destination move with it or stay with what remains? Also, , designers must make special provisions for - Figure 10. An example of a link with a point source and a region destination. The deleting, moving, or copying the defining source of the link is a token in the lext of document A which contains a textual end points of a region. identifier (“xxxx”). The identifier may be (1) the name of the destination node (in this case it would be “B”), (2) the name of the link, or (3) an arbitrary string which Organizational links. Like reference is neither the name of the link nor the destination node. The destination of this link links, organizational links establish is node B which is a region. The link has an internal name (5327) which is normally explicit links between points in hypertext. visible to the user. Organizational links differ from rcferen- lial links in that lhcy implcmcnt hicr;lrchi- cal information. Organizational links connect a parent node with‘ its children and thus form a strict tree subgraph within the hypertext network graph. They correspond to the IS- also be either a point or a region. (See Fig- ure 10 illustrates the most common form A (or superconcept) links of semantic net ure 10.) of hypertext link, in which the source is a theory, and thus operate quite differently A linkpoint is some icon indicating the point and the destination is a region. This than referential links.’ For example, presence of the link. It usually shows the example typifies many of the link applica- rather than appearing as explicit high- link’s name and perhaps also its type. Or tions listed above, because it shows how a lighted tokens in each node, organiza- it may show the name and/or type of the -chunk of text-a region-is written about tional links are often traversed by a destination node. In systemssuch asNep- %r referenced by some smaller chunk of separate mechanism at the node control tune which support links with both point text, often a sentence. Since most readers level (i.e., special goto-parent, goro-jifst- source and point destination, the icon also are accustomed to single point references child, and golo-nexl-sibling commands). indicates which type of linb is indicated, In to sentences (i.e., footnotes), they have no In other cases, there are organizational some systems, the display of links can be problem accepting a link with a point nodes (such as tot nodes in Tcxtner and supprcsscd. so that the documents appear source. There can be regions in graphics as FilcBoses in NotcCards) which record the linear. well-either bordered regions or collcc- organizational structure. A fink region is a set of contiguous Cons of graphic objecrs in a figure. -- characters which is displayed as a single Link regions can pose difficult design unit. In Figure 10, the link destination is problems. They arc casiesl to implemcnr a link region, namely, an entire node. Fig- as whole nodes. since setting a region OTT

34 COMPUTER tin the Keyword links. In addition to the several ways. Links can connect more than station explicit linking performed by referential two nodes to form clusterlinks. Such clus- reee, and organizational links, there is a kind of ter links can be useful for referring to implicit linking that occurs through the use several annotations with a single link, and C of keywords. This type of linking is yet to for providing specialized organizational CUlliES be fully explored. (73 percent) of which connected FileBoxes structures among nodes. Indeed, the tot Aearly One of the chief advantages of text stor- to each other or to individual notecards, nodes of Textnet and the FileBoxes of rs pro- age on a computer is the ability to search 261 (7.5 percent) of which were nonhier- NoteCards are both forms of clusIer links. egions large and complex documents and sets of archical referential links, and the One useful way to extend the basic link region documents for substrings and key- remainder of which were mail links (used is to place attribute/value pairs on links ,vel of words. l Naturally, this ability is also a by the system to tic mail messages to other and IO query the network for them. The ique is valuable aspect of hypertext. Indeed, most nodes). This example, for what if is worth, Neptune system, for example, has an .n two users of large hyperdocuments insist on suggests that hierarchical structure is very architecture that is optimized for this func- having some mechanism for scanning their important in organizing a hypertext net- tion. Coupled with specialized routines in tgions content, either for selected keywords work, and that referential links are impor- the daIabase interpreter (the HA1\1), these Jnlike (which can apply to nodes, links, or tant but less common. aIIribute lists allow users to customize lvious regions) or for arbitrary embedded strings. One advantage of a strictly tree-oriented links in several ways, including devising llaced From a funcIiona1 standpoint, link fol- system is that the command language for their own type system for links and per- of Ihe lowing and search are similar: Each is a navigation is very simple: From any node, forming high-speed queries on the types. way to access destination nodes that are of the most one can do is go to the parent, a It is also possible to perform procedural ult to possible interest. Link following usually sibling, or a child. This simplicity also attachments on alink so that traversing the a sys- yields a single node, whereas search can diminishes the disorientation problem, link also performs some user-specified side :, and yield many; hence, a keyword is a kind of since a simpler cognitive model of the effect, such as customizing the appearance Irings implicit computed link. The value of this information space will suffice. of the destination node. This ability is .gions insight is that it may allow design of a Of course, the great disadvantage of any provided in Neptune and Boxer. ie not hypertext interface which is consistent hierarchy is that its structure is a function n one across all link-tracing activities. of the few specific criteria thaI were used Hypertext nodes. Although the essence a des- in creating it. For example, if one wishes of hypertext is its machine-supponed link- n the To tree or nor to Iree. Some hypertext to investigate what sea-based life forms ing, the nodes contribute significantly IO :110vr systems (for example, Emacs INFO) sup- have in common with land-based life defining the operations that a hyperrext 11, port only hierarchical structures, others forms, one may find that the traditional system can perform. Most users of hyper- 1s (such as Xanadu and Hyperties) provide classification of life forms into the plant text favor using nodes which express a sin- ~I&~ no specificsupport for hierarchical struc- and animal kingdoms breaks up the infor- gle concept or idea, and are thus much tures, and others (such as Textnet and mation in the wrong way. The creator of smaller than traditional files. When nodes NoIeCardsj support both kinds of a hierarchical organization must anticipate are used in Ihis fashion, hypcrIcxI rcnce struclures. the most important crireria for later access introduces an intermediate level of blish One could question just how sufficient to the information. One solution to this machine support between characrers and ‘text. strictly hierarchical structures are, and for dilemma is to allow the informaIion ele- files, a level which has the vaguely seman-

crcn- which applications they are sufficient and mcnIs IO be sIrucIured into mulIiple hier- tic aspect of being oricnIcd IO the cxprcs- trchi- for which they are not. On the one hand, archies, thus allowing the world IO be sion of ideas. Bur this sizing is compleIely abstraction is a fundamental cognitive “sliced up” into several orthogonal ar the discretion of the hypertext writer, arent process, and hierarchical structures are the decompositions. Any hypertext system and Ihe process of determining how to I-m a most natural structures for organizing which has hierarchy nodes, such as Text- modularize a document into nodes is an rtext levels of abstraction. On the other hand, net (tot nodes) and NoteCards (FileBox art, because its impact on the reader is not :c IS- cases obviously exist where cross- nodes), can perform this operation quite well understood.” ic net hierarchical links are required. Frank easily. These are the only systems which enIly Halasz, one of the developers of explicitly claim to support multiple hierar- The rnodulorizotion ofideos. Hb-pertext nple, NoteCards, has gathered statistics on the chies. Indeed, one early user of NoteCards invites the writer to modularize ideas into Tigh- hypefspuce of a single representative used the system in doing the research and units in a way IhaI allows (1) an individual Gza- NoteCards user; this person had 1577 writing for a major project paper; he idea to be referenced elsewhere, and (2) 7y a nodes (cards) in all, 502 of whi& were File- imposed one organization on the data and alrernative Successors of a unir IO be :Irol Boxes (hierarchical nodes). Connecting his writings while doing the research, and offered to the reader (for instance, more ‘irsr- these nodes were a total of 3460 links, 2521 then quite a differenr (yer coexistent) detail, an example, bibliographic refer- &). organization on the same maIerial IO pro- ences, or the logical successor). But the inal duce his paper. As a generalization, it writer must also reckon \viIh the fact IhaI 7hcrc 1s some con~r~verry owr Ihc rela~ivc mcriu of and kcy\vord reuieval PS opposed IO full WI search. On the seems thar engineering-orienIcd hypcrIext a hypcrIexI node, unlike a texIua1 para- i the one hand, keyword rclricval is only as good as Ihe skill users prefer hierarchical organizations, graph, tends to be a stricr unit which does and lhoroughncss of Ihe person sclwing Ihe key- whereas arts- or humanities-oriented users words. On Ihe olhcr hand. full 1~x1 search does no1 not blend seamlessly with its neighbors. find all the rclcvan\ documcnls. nor does ir always find prefer cross-referencing 0rganizaIions. Some hypertext systems (Norecards, only~hcrclcvan~documcncs. Ilrshor~omingsarcduc CREF, Boxer. FRES, NLS) nllow nodrs IO in par* 10 fhc commonness Of synotlyms in English. In addIllon, full ICX~ search can be compulalionally pro- Euetlsions IO basic links. CerIain fea- be viewed Iogether as if they were one big hibuisc in large nelworks. tures of the link enable it to be extended in node, and this op:ion is essential for some

ER Seprember 1967 35 ,. :. applications (for example, writing and Scntisrrrr&reti Nodes.’ So far I have .reading prose). But the boundaries around ‘spoken of the hypertext node as a struc- nodes are always discrete and require tureless “blank slate” into which one sometimes difficult judgcments about how might put a word or a whole document. 10 cleave the subject matter into suitable ment hierarchically (i.e., as an outline) at For some applications, there is growing chunks. the same time that one adds new sections interesr in semistrucfured nodes-typed The process of identifying a semanti- and embellishes existing ones. People nodes which contain labelled fields and cally based unit, such as an idea or con- don? think in terms of “screenfulls”; they spaces for field values. The purpose of cept, with a syntactic unit, such as a think in terms of ideas, facts, and evi- providing a template for node contents is paragraph or hypertext node, is not unique dence. Hypertext, via the notion of nodes to assist the user in being complete and IO to hypertext. Manuals of style notwith- as individual expressions of ideas, provide assist the computer in processing the standing, traditional text has rather loose a vehicle which respects this way of think- nodes. The less that ihecontent of a node conventions for modularizing text into ing and working. is undifferentiated natural language (for paragraphs. This looseness is acceptable example, English) text, the more likely that because paragraph boundaries have a rela- Typed nodes. Some hypertext systems the computer can do some kinds of limited tively minor effect on the flow of the read- sort nodes into different types. These processing and inference on the textual ing. Paragraph boundaries are sometimes lypednodes can be extremely useful, par- subchunks. This notion is closely related provided just IO break up the text and give ticularly if one is considering giving them to Malone’s notion of semistructured the eye a reference point. Thus, decisions some internal structure, since the types can information systems.” ~ about the distribution of sentences among be used IO differentiate the various struc- To continue wirh the example of the paragraphs is not always critical. tural forms. Design Journal, we have developed a Hypertext, on the other hand, can For example, in our research in the model for the internal structure of deci- enforce a rather stern information hiding. MCCSoftwareTechnologyProgram,we sions. The model is named ISAAC. It In some systems, the only clue a user has have been implementing a hypertext inter- assumes that therearc four major compo- as to the contenrs of a destination node is face for a design environment called the nents 10 a design decision: Design Journal. The Design Journal is the name of the link (or the name of the (1) an issue. including a short name intended 10 provide an active scratchpad node, if that is provided instead). The for the issue and a short paragraph writer is no longer making all the decisions in which the designer can deli5erate about describing it in general terms; design decisions and rationale, both about the flow of the text. The reader can (2) a set of olfernarives, each of which individually and in on-line design meet- and must constantly decide which links IO resolves the issue in a different way, pursue. In this sense, hypertext imposes on ings, and in which he can integrate the each having a name and short design itself with this less formal kind of both the writer and the reader the need for description, and each potentially information. For lhis purpose we have more process awareness, since either one linked to the design documents or ele- has the option of branching in the flow of provided a set of four typed nodes for the ments that implement the alternative; -noles, goals/consfrainrs, the text. Thus hypertext is best suited for designer to use (3) an analysis of the competing alter- applicarions which require these kinds of ani/octs, and decisions. Nofes arc used for natives, including the specific criteria cvcrything from reminders, such as “Ask judgcments an)-way, and hypcrlcxt mcrcly being used IO evaluate them, rhe offers a way 10 act directly on these judge- Bill for advice on Module X,” to specific trade-off analysis among these alter- menrs and see [he results quickly and problems and ideas relating 10 the design. natives, and links to any data that the graphically. Goals/constraints are for the initial analysis diaws upon; and requirements as well as discovered con- (4) a commifmenf IO one of the alter- ldeosus objects. While difficult to docu- straints within the design. Artifacts are for natives (however tentatively) or co a ment, there is something very compelling the elements of the output: The Design. about reifying the expression ofideas into vector of preferences over the alter- And decisions are for capturing the branch natives, and a subjective rating about discrete objects [o be linked, moved, and points in the design process, the alterna- changed as independent entities. Alan Ka) the correctness or confidence of this tives considered by the designer, and some commitment. and Adele Goldberg” observed of of the rationale for any commitment (how- Small[alk that ir is able to give objects a ever remalive) that has been made. The Without getting into the details of the perceptual dimension by allocating to designer captures assumptions in [he form underlying theory, I merely wish 10 stress them a rectangular piece of screen real of decisions with only one alternative. Our here that the internal structure of ISAAC estate. This feature offers enhanced prototype of the Design Journal uses color suggests that the author of an IStiC deci- retrieval and recognition over computer- to distinguish between note types in the sion is engaged in a+ch more structured processed flat documents, because 10 a browser, and we have found this to be a activity than just “writing down the deci- much greater degree abstract objects are very effective interface. sion,” and the reader is likewise guided by directly associated with perceptual Hypertext systems that use typed nodes the regularity of the ISAAC structure. objects --rhe windows and icons on the generally provide a specialized color, size. Of course, it may not be clear why \ve do screen. or iconic form for each node type. The dis- not treat each of [hc elements lisred above Paragraphs, sections. and chaprers in a tinguishing fearures help the user differen- as its own typed hypertest node. The rea- book, viewed through a standard text edi- tiate at a glance rhe broad classes of typed son is that the parts of an ISAAC frame tor or word processor, don’t stand out as nodes that he is working wirh. Systems are much more rightly bound logether first-class entities. This is particularly such as NoteCards, Intermedia, and IBIS than ISAAC frames are bound IO each apparent when one can view one’s docu- make cstensivc USC of typed nodes. other. For example, we could not have an

36 COXIPUTER

- have analysis part without an alternatives part; importance of hypenelt is simply that illUC- yet if we treat them as separate hypertext references are machine-supported. Like I f!,, nodes, we have failed to build this con- hyperlext, traditional literature is richly straint into the structure. The general issue interlinked and is hierarchically organircd. ‘-4 here is that some information elemeqts One problem with composite nodes is In traditional literature. the medium of :yped must always occur together, while others that as the member nodes grow and change print for the most part resfricts the flow of i and may occur together or not, depending on the aggregation can become misleading or reading IO folicw the tlow of linearly se of how related they are in a given context and incorrect. A user who encounters this arranged passages. However, the process nts is how important it is to present them as a problem is in the same predicament as a of following side links is iundamental even nd to cluster distinct from “surrounding” infor- writer who has rewritten a section of a in the mediumof print. In fact, library and : the mation elements. This problem is recur- paper so thoroughly that the section title information science consist principally of node siVe: /.n element that is atomic at one level is no longer accurate. This “semantic the investigation of side links. Anyone : (for may turn Out on closer inspection to con- drift” can be difficult to catch. who has done research knows that a con- y that lain many components, some of which are siderable portion of that effort lies in nited clustered together. Analog\’ to semantic networks. The idea obtaining referenced works, looking up xtual In hypertext this tension presents itself of building a directed graph of informal cross-references, looking up ccrms in a dic- lated as the twin notions of scmistructurcd textual elements is similar lo the Al con- tionary or glossary, checking tables and .urcd nodes and composite nodes. cept of semonlic nerbvofks. A semantic figures, and making noles on notecards. Composite nodes. Another mechanism network is a knowledge representalion Even in simple reading one is constantl) f the for aggregating related information in scheme consisting of a directed graph in negotiating references to other chapters or ed a hypertext is the composife node. Several which concepts are represented as nodes, sections (via the table of contents or refer- deci- related hypertext nodes are “glued” and the relationships between concepts are ences embedded in the test), index entries, c. It together and the collection is treated as a represented as the links between them. footnotes, bibliographic references, side- mpo- single node, with its own name, types, ver- What distinguishes a semamic network as bars, figures, and tables. Often a text sions, etc. Composite nodes are most use- an AI representation scheme is fhat con- invites the reader to skip a section if he is ne ful for situations in which the separate cepts in the representation are indexed by not interested in greater technical detail. ,h items in a bulleted list or the entries in a their semantic content rather than by some But there are problems with the rradi- table are distinct nodes but also cohere into arbitrary (for example, alphabetical) tional methods. ordering. One benefit of semantic net- hicb a higher level structure (such as the list or l Most references can? be traced back- :y table). This practice can, however, under- works is that they are natural to use, since wards: A reader can nor easily find where mine the fundamental association of one related concepts tend to cluster together in a specific book or article is referenced in interface object (window) per database the network. Similarly, an incompletely or a document, nor can the aurhor of a paper ele- object (node), and rhus must be managed inconsistently defined concept is easy to find out who has referenced the paper. spot since a meaningful context is provided ;...e; well to avoid complicating the hypertext l As the rcadcr winds his way down vnr- ;‘“r_ idiom unduly. by those neighboring concepts IO which it ious refercncc trails, he must keep track of tria A composite node facility allows a is alreadY linked. which documents he has visited and which group of nodes to be treated as a single The analogy to hypertext is straightfor- he is done with. :r- node. The composite node can be moved ward: Hypertext nodes can be thought of l The rcadcr must squeeze annotations the and resized. and closes up to a suitable icon as representing single concepts or ideas, into the margins or place rhem in a sepa- reflecting its contents. The subnodes are internode links as representing the seman- rate document. er- separable and rearrangeable through a tic interdependencies among these ideas, l Finally, following a referential trail a subedit mode. The most flexible means of and the process of building a hypertext net- among paper documents requires subsran- r- displaying a composite node is to use a work as a kind of informal knowledge tial physical effort and delays, even if the out constraint language (such as that devel- engineering. The difference is that Al reader is working at a well-stocked library. :is oped by Symbolics for Constraint Frames) knowledge engineers are usually striving to If the documents are on linc, the job is eas- which describes the subnodes as panes in build representations which can be ier and faster, but no less tedious. the composite node window and specifies mechanically interpreted, whereas the goal the the interpane relationships as dynamic of the hypertext writer is often to capture New possibilities for authoring and design. -ess constraints on size and configuration. an interwoven collection of ideas without Hypertext mayoffer new Ilays for authors ;4C Composite nodes can be an effective regarda their-machine interpretability. and designers to work. Authoring is Xi- means of managing the problem of having The work on semantic networks also sug- usually viewed as a word- and sentence- red gests some natural extensions to hypertext, a large number of named objects in one’s level activity. Clearly the \\ord processor l tci- environment. Pirman described the prob- such as typed nodes, semistructured nodes : by lem this way: (frames), and inheritance hierarchies of node and link types. In this sari of system. there is a never-ending do tension bctwecn trying 10 name everything (in >V? which case, the number of named lhings can 2 grow quickly and rhescr can become quickly ne unmanageable) or to name as little as Dossi- The advantages and blc (in wvilich cm, ~hmgs thn! look J iot of tcr uses of hype&t trouble to COIISI~UCI can bc I~rd IO rctricvc ch if one acciden[ally drops the poinrers [O an rhem).lY Interrextual references are not new. The

.R Scptcmbcr 1987 37 .*I , r is a good tool for authoring at this level. if the text is moved, even to anolher However; authoring obviously has much document, the linkinformation still to do with structuring of ideas, order of provides direct access to the presentation, and conceptual exploration. reference; Few authors simply sit down and pour out mcmex. First he runs through an cncyclopc- ruskstucking: the user is supported in a finished text, and not all editing is just dir, finds an interesting but sketchy article. having several paths of inquiry active “wordsmithing”and polishing. In a broad leaves it projected. Next, in a history, he finds and displayed on the screen at the sense, authoring is the design o/u docu- another pertinent hem, and ties the two together. Thus he goes, building a trail of same time, such that any given path menf. The unit of this level of authoring is many items. Occasionally he inserts a com- can be unwound to theoriginal task; the idea or concept, and this level of work ment of his own. either linking it into the collaboration: several authors can can be effectively supported by hypertext, main :rail or joining it by a side trail to a par- collaborate, with the document and since the idea can be expressed in a node. ticular item. When it bccomu evident that the comments about the document being As the writer thinks of new ideas, he can elastic properties of available materials had a great deal to do with the bow, he branches tightly interwoven (the cxploratio-, of develop them in their own ,lodes, and then off on a side trail which takes him through this feature has just begun). link them to existing ideas, or leave them textbooks on elasticity and tables of physical isolated if it is loo early to make such constants. He inserts a page of longhand associations. Thespecialized refinements analysis of his own. Thus he builds a (perma- nent] trail of his interest through the maze of The disadvantages of of a hypertext environment assist the materials available IO him.* movement from an unstructured network hypertext As we have seen, Bush’s notion of the to the final polished document. “trail” was a feature of Trigs’s Testnet, There are two classes of problems with allowing the hypertext author to establish hypertext: problems with the current New possibilities for reading and a mostly linear path through the docu- implementations and problems that seem retrieval. Hypertext may also offer new ment(s). The main (default) trail is well 10 be endemic to hypertext. The problems possibilities for accessing large or complex marked, and the casual reader can read the in the first class include delays in the dis- information sources. A linear (nonhyper- text in that order without troubling with play of referenced material, restrictions on text) document can only be easily read in the side trails. names and other proper:ies of links, lack -the order in which the text flows in the of browsers or deficiencies in browsers, book. The essential advantage of non- Summary. We can summarize the opera- etc. The following section outlines two linear text is the ability to organize text in tional advantages of hypertext as: problems that are more challenging than different ways depending on differing these implementation shortcomings, and viewpoints. Shasha provides the following ease of tracing references: machine that may in fact ultimately limit the useful- description of this advantage: support for link tracing means that all references are equally easy to follow ness of hypertext: disorientation and cog- are Suppose you a touris. interested in visit- nitive overhead. ing museums in a foreign city. You may be forward to their referent, or back- interested in visual arrs. You may want tosee ward to their reference; museums in your local area. You may only ease of crearing new references: users Gttling “lost in space.” Along with the bc inrcrcstcd in incxpcnsivc muscums. You can grow their own networks, or sim- power to organize information much more ccrrainly want 10 make sure the museums you ply annotate someone else’s docu- complexly comes the problem of having to consider arc open when you want IO visit them. Now your guidebook may be arranged ment with a comment (without know (1) where you are in the network and by subject, by name of museum, by location. changing the referenced document); (2) how to get to some other place that you and so on. The rroublc is: if you are inrercsted infortuorion slrucluring: both hierar- know (or think) exists in the network. I call in any arrangcmcnt other than the one it uses, chical and nonhierarchical organiza- this the “disorientation problem.” Of you may have IO do a lot of searching. You tions can be imposed on unstructured course, one also has a disorientation prob- are not likely IO find all the visual arts museums in one secrion of a guidebook that information; even multiple hierar- lem in traditional linear text documents, has been organized by district. You maycarry chies can organize thesame material; but in a linear text, the reader has only two several guidebooks. each organized by a global views: browsers provide table- options: He can search for the desired text criterion you may be inrcrcsred in. The num- of-contents style views, supporting earlier in the .tcxt or later in the text. ber of such guidebooks is a measure of the need for a nonlinear texi system.” easier restructuring of large or com- Hypertext offers more degrees of freedom, plex documents; global and local more dimensions in which one can move, Another advantage is that it is quite nat- (node or page) views can be mixed and hence a greater potential for the user ural in a hypertext environment IO suspend effectively; to become lost or disoriented. In a ner\vork reading temporarily along one ,line of cusfouCzed documenrs: text segments of 1000 nodes, information can ea_sily investigation while one looks into some can-& threaded together in many become hard to find or even forgoTten detail, example, or related topic. Bush ways, allowing the same document to altogether. (See Figure i 1.) described an appealing scenario in his 1945 serve multiple functions; There are two major technological solu- article: rnodularily of infornlarion: since the tions for coping with disorientation- The owner of the mcmcs. lcr us say, is same test segment can be referenced graphical browsers and qucry/scarch inlcrcslcd in the origin and propcrtics of the from scverai places. ideas can be mechanisms. Bro\rscrs rely on the bow and arrow. Specifically hc is srudying expressed with less overlap and dupli- extremely highly developed visuospatial why the shorr Turkish bow was apparently cation; processing of the human visual system. By superior IO rhe Enghsh long bow in [he skir- mishes of lhc Crusades. He has dozens of consislency Of inforn~ofion: rcfer- placing nodes and links in a two- or thrce- possibly pcrlincni books 2nd articles in his cnccs arccnlbcddcd in their test. and dirncnsionnl spasc, providing them with

38 COLIPIJTER _ . 1 * t .: other . . II still Ihe

flC.,_

e H-&e path task; ; can t and oeing on of

with -rent ,eem lems : dis- 1s on lack sers, IWO han and !f

the lore gto 3nd JOU call Figure 11. Tangled web of links. This experimental implementation of a global map in the Infermedia system shows lhe diffi- Of culty of providing users with spatial cues once a linked corpus contains more than a few doien documents. This global map Db- only represents about one tenth of the documents in a corpus designed for a survey of English literature course. IIS. ‘WO ext :xt. m, ve, properties useful in visual differentiation large hyperdocument, one is by definition standard database search and query tcch- XT (color, size, shape, texture), and maintain- disoriented. In addition, an adequate vir- niques to locating the node or nodes which trk ing cert@n similarities to our physical envi- tuality is very difficult IO maintain for a the user is seeking. This is usua!J done by ily tinment”(for example, no two objects large or complex hypertext network. Such using boolean operations IO apply some en occupy the same space, things only move parameters as (1) large numbers of nodes, combination of keyword search, full string if moved, etc.), browser designers are able (2) large numbers of links, (3) frequent search, and logical predicates on other :u- 10 crcaie quite viable virtual spatial envi- changes in the nerwork, (4) slow or awk- attributes (such as author, time of crea- ronments. Users orient themselves by vis- ward response to user control inputs, (5) lion, Iypc, c1c.) of nodes or links. Simi- zh ual cuts, just as when fhcy arc walking or insufficicnr visual diffcrcnriation among larly, one can fihcr (or ellide) informaGon he driving through a familiar city. However, nodes and/or links, and (6) nonvisually so thal the user is presented with a manage- the:e is no narural topology for an infor- oriented users combine IO make it pracri- able level of complexity and detail, and can j.__ malion space, except perhaps that higher tally impossible IO abolish rhe disoricnta- shifr the view or the dclnil suppression e- lcvcl conccprs go a[ rhc top or on the left [ion problem with a browser alone. white navigafing rhrough ~hc ncf\vork. :h side, so until one is familiar with a given One solurion 10 this dilemma is IO apply Howcvcr, much research remains IO bc

9 September 1987 39 .

done on effective and standardized how do you decide if following the side for. But-1 also intended something more: methods for eliision. path is worth the distraction? Does the that the reader come away from this rrti- label appearing in the link tell you enough de excited, eager IO try using hypertext for The cognitive task scheduling problem. to decide? This dilemma could be called himself, and aware that he is at the btgin- The other fundamental problem with “informational myopia.“The problem is ning of something big, something like the using hypertext is that it is difficult to that, even if the system response time is invention of the wheel, but something that become accustomed to the additional men- instantaneous (which it rarely is), you still has enough rough edges that no one is tal overhead required to create, name, and experience a defmite distraction, a “cog- really sure that it will fulfill its promise. keep track of links. I call this “cognirive nitive loading,” when you pause to con- To that end, I mention one more book overhead.” Suppose you are writing about sider whether to pursue theside path. This that might be considered to belong to the X, and a related thought about Ycomes to problem can be eased by (1) having the literature on hypertext. Neuromoncer2’ is mind and seems important enough to cap- cross-referenced node appear very rapidly a novel about a time in the distant future ture. Ideally, hypertext allows you to sim- (which is tI-.e approach of KMS), (2) when the ultimate computer interface has ply “press a button” (using some mouse providing an instantaneous one- to three- been perfected: One simply plugs one’s or keyboard action) and a new, empty line explanation of the side reference in a brain into the machine and experiences the hypertext window pops onto the screen. pop-up window (which is the approach of computer data directly as perceptual enti- You record Yin this new window, then Intermedia), and (3) having a graphical ties. Other computers look like boxes you press another button, the Y window browser which shows the local subnetwork floating in three-dimensional space, and disappears, and you are in the Xwindow into which the link lcads. passwords appear as various” kinds of right where you were when Yoccurred to These problems are not new with hyper- doors and locks. The user is completely you. text, nor are they mere byproducts of immersed in a virtual world, the “operat- Unfortunately, the situation is a bit computer-supported work. People who ing system,” and can move around and more complex than this scenario implies. think for a living-writers, scientists, take different forms simply by willing it. If Y has just occurred to you, it may still artists, designers, etc.-must contend with This is the ultimate hypertext system. be hazy and tentative; the smallest inter- the fact that the brain can create ideas The basic idea of hypertext, after all, is ruption could cause you to lose it. Coming fasrcr than the hand can write them or the that ideas correspond to perceptual up with a good word or short phrase to mouth can speak them. There is always a objects, and one manipulates ideas and summarize Y may not be easy. You have balance between relining thecurrent idea, their relationships by directly manipulat- to consider not just what is descriptive but returning to a previous idea to refine it, ing windows and icons. Current technol- also what will be suggestive for the reader and attending to any of the vague “proto- ogy limits the representation of these when he encounters the link to Ywithin X. ideas” which are hovering at the edge of objects to static boxes on a CRT screen, In addition, you must determine whether consciousness. Hypertext simply offers a but one can easily predict that advances in you should name the link to Y to suggest sufficiently sophisticated “pencil” IO animation, color, 3D displays, sound, the contents of Yor to show Y’s relation- begin to engage the richness, variety, and etc.-in short, Nelson’s hypermedia-will ship to X. Some systems (for example, interrelatedness of creative thought. This keep making the display more active and NoteCards) provide that links can have aspect of hypertext has advantages when realistic, the data reprcsentcd richer and both a type (such as “idea”) and a label this richness is needed and drawbacks more detailed, and the input more natural (such as “subsume A in 5”). Coming up when it is not. and direct. Thus, hypcrtcxt, far from with good names for both can impose even To summarize, then, the problems with being an end in itself, is just a crude first more load on an author srruggling with an hypertext are step toward the time when the computer is uncertain point. (One way to reduce this disorientation: the tendency to lose a direct and powerful extension of the problem is for the authoring system IO sup- one’s sense of location and direction human mind, just as Vannevar Bush envi- port immediate recording of the substance in a nonlinear document; and sioned when he introduced his Mcmex of the idea, deferring the creation and cognitive overheud: the additional four decades ago.0 labeling of the link and/or the node unril effort and concentration necessary to after the thought has been captured.) maintain several tasks or trails at one Beyond that, you must also consider if time. Acknowledgements you have provided sufficient links to Y These problems may be at least partially before returning to work on X. Perhaps resolvable through improvements in per- I wish IO thank Les Bclady, Bill Curtis. Susan there are better ways to link Y to the net- Gerhan. Raymondc Cuindon, Eric Gullichscn. formance and interface design of hyper- work of thoughts than at the point in X Frank Halasz. Peter Marks, and Andy van Dam text systems, and through research on _ wher+_Y came to mind. for their rhoughtful reading of previous drafts. information filtering techniques. Th? problem of cognitive overhead also occurs in the process of reading hypertext, References which tends to present the reader with a n this article, 1 have reviewed exist- lar_ee number of choices about ‘which links ing hypertext systems, the opportu- 1. T.H. Nelson. “Gclting II Our of Our Sys- (cm.” Injorrfraiio~l Rclrrcvul: 11 Crllrctrl IO follow and which IO lcnvc alone. Thcsc nities and problems of hypcrtcxt, and I\‘cv;cw. C. Schcchlcr, cd.. ‘I‘hompron Clioiccs cngcndcr a certain ovcrhcad of some of the top-level design issues of Books. Wash., D.C., 1967. melalcvel decision making, an overhead building hypertext systems. It has been my 2. V. Bush, “As WC Xlny Think.” ,4llanrir thar is absent when the author has alread) intention to give the reader a clear sense of ,Voonrh/~. July 1945. pp.lOl-109. made many of thcsc choices for you. At what hypertext is, what its strengths and 3. D.C. Engclbart. “A Conccplual Frnmc- the moment that you cncountcr a link, wcakncsscs arc. and what it can bc used work for the Augmcnr~lion or ,Man’s

40 COMPUTER I ; .. .. . : _, :_-. more: Intcllcct ,” in V&as in In/wmolion Hon- 20. P.J. Brown, “interactive Docutncma~ion.” in Softwore: is arri- dliq, Vol. I, Spartan Books, London, Practice ond Experi- 1963. ence. March 1986. pp. 291-299. y”/! for 4. D.C. Engclbart and W.K. English, “A 21. D. Shasha. “When Does Non-Linear Text .- Research Center for Augmenting Human Help? fiperr Dotobase Systems, Proc. 01 1e Intellect,” AFIfS Con/. froc.. Vol. 33. the First Inf’l Con/.. April 1986. pp. ,;hat Part 1. The Thompson Book Company, 109-121. *one is WashingI&, D.C.. 1968. 22. A. Kay and A. Goldberg, “Personal *misc. 5. T.H. Nelson. “Replacing the Printed Dynamic Media,” Compurer. March 1977, .’ Word: A Complctc LitcrarySystem.“fFJP pp. 31-41. : book E. Jeffrey Conklin is a m:mber of the rcscarc: Proc., October 1980, pp. 1013-1023. to the 23. T.W’. Malone CI al. “Jntelligent staff and GE’s liaison to the Software Technol 6. R.H. Trigg, A Network-basedApproach IO Information-Sharing Systems,” Communi- :ei’ is ogy Program ht the Microzlectronicr and Corn 7ex1 Handling jar Ihe Online Scientijic colionr o/ fhe ACM, May 1987. pp. puler Technology Corporation (MCC). Hi future Communily, PhD. Thesis, University of 39042. research centers on constructing informatior cc has Maryland, 1983. 24. W.Cibson. Neuromoncer. Ace Science Fit- syrtcms for the capture and USC of dcsigr one’s 7. H. Rittcl and M. Webbcr, “Dilemmas in a lion. 1984. rationale. :es the General Theory of Planning,” PO/icy Conklin Received his BA from Ant&h Col- Sciences, Vol. 4, 1973. lege and his MS and PhD from the Universit) .I enti- A more detailed version of this article, 8. D.G. Lowe, “Cooperative Structuring of of Massachusetts at Amherst. boxes including an extended bibliography, is avail- Information: The Rcprcscntatio? of able from the author. To obtain a copy, cir- :, and Reasoning and Dcbatc,” in In/‘/. 1. ofMon- cle number 181 on the Rcadcr Service Card Readers may write to Conklin at MCCSofr- ds of Machine Sludics,” Vol. 23, 1985, pp. at the back of the magazine. wareTechnology Program, P.O. Box 20019J. Aetely 97-111. Austin, TX 78720; (512) 3.t3-0978. ?erat- 9. J.B Smith et al, “WE: A Writing Environ- 3 and ment for Professionals,” Technical Report 86-025, Department of Computer Science, ing it. University of North Carolina at Chapel stem. Hill, August 1986. all, is 10. W. Hershey, “Idea Processors,” BYTE. .ptual June 1985, p. 337. s and 11. D. McCracken and R.M. Akscyn, “Expe- )ulat- rience with the ZOG Human-computer InterfaceSystem,” Jnl’lJ. o_fMon-Machine User-Oriented Content-Based hnol- Studies, Vol. 21, 1984, pp. 293-310. these 12. B. Shneidcrman and J. Morariu, “The Text and Image Handling Interactive Encyclopedia System (TIES),” :c_ . Department of Computer Science, Univer- MASSACHUSETTS INSTITUTE OF TECHKOLOGY und, sity of Maryland, College Park, MD 20742, June 1986. Cambridge, MA n March ZI- 24, 1988 -will 13. J.H. Walker. “The Document Examiner.” General Theme: Information rctricval sysrcms for full-test a,ltl mixed : and SIGGRAPH’ Video Re,ic.v. Edited Compi- . and lation from CHI’85: iiumon Fuclors in media databases handling: Computing System, 19F5. tural l techniques designed to reduce m darn entry and control 14. ‘ram F.G. Halasr. T.P. Moran, and T.H. Trigg, imprecision in full-text database 9 “friendly” cni-user inrcrf3ccs first “NoteCards in a Nutshell,” Proc. ojrhc searching 9 new media :cr is ACM Con/. on llurnon Focrors in Cotnpul- &?.S.vsrems, Toronto, Canada, April 1987. SpeciGc Themes: * the IS. N.L. Garrett, K.E. Smith, and N. Mcy- A) Linguistic processing and interrogation of full-text databases. :nvi- rowitz. “Intermedia: Issues, Strategies, and B) Automatic thesaurus construction. mex Tactics in the Design of a Hypermedia Document System,” in Proc. Con/. on C) Expert system techniques for rctricving information in full-tcxr and Compufer-Supporfed Cooperafive Work, multimedia databases. MCC Software Technology Program, Aus- D) Friendly user interfaces to classical information retrieval svstcms. tin, Texas, 1986. E) Specialized machines and system architecture designed fir treating full- 16. N. Yankelovich, N. Meyrowitz, and A. van text data, including managing and accessing widely distributed databases. Dam, “Reading and Writing the Elcc- Automatic database construction using scanning techniques, Jsan tronic Book,” Computer, October 1985. F) optical ircn, 17. N. Delislc and M. Schwartz, “Neptune: A character readers, output document preparation, ctc . . . Iam Hypertext System for CAD Applications.” G) ScwapplicaZjons and pcrspcctivcs suggested by emerging new IfIS. Proc. o_f ACM SIGMOD Inr’l Con/. on technologies. Management of Dafa, Washington, D.C., May2830, 1986, pp. 132-143. (Alsoavail- DEADLCI’ES: able as SIGhlOD Record Vol. IS, No. 2, SEPTE.\IBER 30. 1987: Inrmtion ofsvbmi~?iqapnpcr. June 1986). OCTOBER 30. 1487: Four copier sioufdbe r&&l/! rhc Conjrrrnrc Sn-rrrnn’nr. 16. A. diScssa. “A Principled Design for an DECE.\IBER 16. 1987: Arcrpranrr nortjicarion. Intcgrntcd Compufational Environment,” Humon-Computer Inferoaion, Vol. 1, Lawrence Erlbaum. 1985, pp. l-47. T‘..- 19. K.M.Pitman. “CREF: An Editing Facility for Xtanaging Structured Text,” A.I. nC- Memo No. 829, M.J.T. A.1. Laboratory, n’s Cambridge, .Mass.. February 1985. Submit papers LO, or for more information: ER September 1987 RIAO 88 Confcrcncc Scnicc Of&c. MIT. Bldg. 7. J1m. 111, Cambridge. \I.-\ 02 139 USA