Models for hypertext - Semantic Scholar

Models for Hypertext

Mark F. Frisse and Steve B. Cousins Medical lnformatics Laboratory, Department St. Louis, MO 63770

of Internal Medicine,

Features characteristic of hypertext are no longer the providence of specialized hypertext systems. Interfaces exhibiting the “power of linking” can be found in software ranging from document preparation programs to operating systems, providing users with many new ap proaches to individual and group information management. To take advantage of the new opportunities afforded by this migration, it is important to understand hypertext at representation levels beneath the more superficial aspects of the human-computer interface. Three abstract models for hypertext-representative members of a spectrum of popular formalisms-are presented to provide insight into the meaning and potential of hypertext. Each model represents a different level in the design-decision process necessary for effective hypertext development, and each model plays an important role in development and use of information management software exhibiting hypertext features.

Hypertext provides new methods for organizing and presenting information. The technology has been employed to facilitate personal information management, improve use of large information resources, and coordinate complex planning tasks. Support for hypertexts and related systems is motivated by a belief that hypertext can provide more efficient and personalized access to text by complementing the global search techniques of traditional information retrieval systems with focal navigation based on meaningful inter- and intradocument connections or “links” identified by other means. Informally, hypertext is simply a collection of modular text elements and a set of inter- and intraelement relationships that can be ordered in many ways and used to satisfy a broad range of personalized information needs. More formally, hypertext is: (1) a database of text; (2) a semantic net which connects text components; and (3) tools for creating and manipulating this combination of text and semantic net (Conklin, 1987; Rada & Lunin, 1989; Schnase, Leggett, Kacmar, & Boyle, 1988). Hypermedia is hypertext with nontextual elements. Software is said to possess “hypertext func-

0 1992 by John Wiley & Sons, Inc.

Washington

University

School of Medicine,

tionality,” or “hypertext features” if it supports behavior characteristic of hypertext systems. Products integrating hypertext functionality with domain-specific application software have been called “hypertext environments” (Scacchi, 1989). Hypertext’s promise makes hypertext research and development topics of growing importance to the users and developers of information systems. In particular, members of the library and information science communities, eager to see the development of systems that realize the medium’s full potential, express a serious need for a more formal understanding of hypertext. Unfortunately, where hypertext is concerned, the transition from intuition to implementation is difficult; although it is relatively easy to explain the desired behavior of hypertext at an intuitive level or at the behavioral level, computer programs derived solely from ad hoc behavioral descriptions may lose their value as the needs of users change and as the scope of information areas to which the technology is applied expands. If one is to support “virtual documents” that can be used by many programs with hypertext functionality, concerned parties must agree upon interchange formats defined by means of a formal model of hypertext structure. If hypertexts are to be used to coordinate intellectual tasks, a welldefined semantic model for hypertext will be necessary. If hypertexts are to be used for learning or sequential information-seeking tasks, one must understand the browsing semantics that govern hypertext use. This overview uses three different models for hypertext as reference points for examining issues relevant to effective hypertext design and use. Each model clarifies certain aspects of the topic: the Dexter model is presented as a model of hypertext structure; gIBIS is presented as a hypertext system based on a model of rhetoric; and Trellis is presented as a hypertext model that incorporates browsing semantics. Several issues problematic to hypertext design and use are discussed with reference to the models, including: (1) defining and manipulating hypertext structure; (2) resolving differences between informal use and directed problem solving; (3) searching for information using hypertext;

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATIONSCIENCE. 43(2):183-191, 1992

CCC 0002-8231/92/020183-09$04.00

and (4) identifying relationships between hypertext and indexed document retrieval systems. An analysis of these issues from the perspective of the various models for hypertext provides a greater understanding of how this technology can be applied to educational, reference, and collaborative work settings.

The Dexter Hypertext

Model

The Dexter model is a representative “neutral” model for hypertext; it relegates to developers as many design policies as possible. The model is composed of three separate layers: (1) a storage layer defines how text components and links can be related to form hypertext networks; (2) a within-component layer provides a mechanism to define the composition of each type of component (e.g., text, table, equation); and (3) a runtime layer provides a mechanism to define how hypertext components and links will be displayed and manipulated (Figs. 1 and 2). The Dexter model also defines mechanisms for mediating between adjacent layers: the storage layer and the within-component layer communicate through the anchor mechanism, and the runtime layer and the storage layer communicate through presentation specifications.

Symbolics Document Examiner records (Walker, 1987; Walker & Bryan, 1988), Intermedia documents (Yankelovich, Haan, Meyerowitz, & Drucker, 1988), and the curds of HyperCard and NoteCards (Irish & Trigg, 1989). The notion of modularity inherent in the concept of a component is central to the general practice of classifying and modifying components independently of one another. The Dexter model’s storage layer makes few requirements on the composition or function of components; a component can consist of any combination of text, graphics, or other media. The only component type critical to the definition of hypertext is the link. Links relate two or more components together. Link attachments are usually defined by a portion of a field within the components called an anchor. If there is an element of directionality associated with a link, a portion of one component may serve as a proximal (“from”) anchor and a portion of a second component serves as the distal (“to”) end. Anchors are uniquely described by the combination of a unique global component identifier with a unique intracomponent anchor identifier; this enables component modification without violating the integrity of other components or links. The Within-Component

The Storage Layer The Dexter model emphasizes the storage layer and anchoring mechanisms. The fundamental unit of the storage layer is the component-a syntactically and semantically discrete entity that represents a single, independent concept and that can be classified and viewed in many ways (Raymond & Tompa, 1988). The nodes of generic hypertext networks are defined as components in the Dexter model. Examples of components include KMS frames (Akscyn, McCracken, & Yoder, 1988),

Layer

The Dexter model leaves developers with complete freedom to define new component types, facilitating the transition from systems containing only text (hypertext) to systems containing text, static illustrations, audio, and animation (hypermedia). For each new component type, one defines, in the WithinComponent Layer, a method for representing both the new entity and anchors to the entity. This representation task is not always straightforward. Although our experience with text has been constrained by the evolution of print and the conventions of the printed page, our experience with sound and images has not been constrained by a similar degree-we have far fewer conventions equivalent to the paragraph, period, and semicolon of the print medium and, accordingly, there is less agreement on both what constitutes an isolated component of sound or image and how one represents anchors to such components. The Runtime Layer

FIG. 1. The three layers of the Dexter model. The Dexter model emphasizes a storage layer in which the fundamental model for hypertext is represented. Internal representation of hypertext components is performed by the within-component layer and mediated by anchoring mechanisms. Presentation is accomplished by a runtime layer in conjunction with presentation specifications (after Halasz & Schwartz, 1989).

184

JOURNAL OF THE AMERICAN

SOCIETY

FOR INFORMATION

Design decisions concerning the appearance of a hypertext component are made at the runtime layer and by means of presentation specifications. It is at this level that one determines the correspondence between text representation and text appearance. Programs like HyperCard support a one-to-one relationship between storage layer representation and presentation layer display; small components are displayed in small text fields and large components are displayed in scrolling

SCIENCE-March

1992

--L? some text in a texl

IO=

Window 51

!

Runtime Layer

Storage Layer

Within-Component Layer

FIG. 2. A depiction of the three layers of the Dexter model as embedded in an actual hypermedia system. The storage layer contains four components (one link component, two text components, and one graphics component). The actual component contents reside within the Within-Component Layer. A graphics single component is being presented within a window in the Runtime Layer.

text fields. Other systems create “structured documents” by combining multiple storage layer components. This requires one to develop complex presentation specifications or to model and manipulate composites-components that can incorporate other components. The integration and simultaneous presentation of multiple components is a concern of a broader interest in “compound documents, ” which is discussed in the electronic publishing literature. Link markers-displayed items within a component that signify the presence of link anchors-are also the concern of the runtime layer and presentation specifications. Design options for link markers include highlighting link-marker text using boldface fonts, placing link markers in specific areas of the displayed component, changing the position of the cursor when it is placed over a link marker, and using icons prominent enough to denote the link marker, yet subtle enough to avoid distracting the reader. The reader’s expectations about the content of the component at the distal link will be profoundly influenced both by the choice of link marker and the context in which it is placed (Landow, 1987). glBIS-Hypertext

Based on a Rhetorical

Model

Hypertext designers and users of hypertext systems must decide on the range of expression that will be encouraged within a specific system. Concerns often arise as a result of the unavoidable tension between complete freedom of expression and the degree of constraint and coherence required for specialized problem solving. At one extreme, Nelson’s “everything for any user” philosophy (Nelson, 1987) encourages liberal applica-

tion of links by each and every hypertext user. At the other extreme, tightly knit groups engaged in intense discussion on a narrow topic require clearly defined component and link semantics. The Dexter model’s “neutrality” leaves this issue to models that incorporate more explicit policy decisions. Conklin and Begeman’s gIBlS (generalized IssueBased Information System) is a hypertext implemenfution based on IBIS (Conklin & Begeman, 1989)a rheforicaf model that allows one to relate issues, arguments, and positions by means of a predefined set of components and semantic links. When issues, arguments, and positions are represented as hypertext components and semantic relationships are represented by links, one has a generalized hypertext suitable for a wide range of design and planning processes. With gIBIS, arguments can “support” or “object-to” positions; positions can “respond-to” issues; and issues can “be or “challenge” positions. ” “expand on,” suggested by, gIBIS provides a semiformal structure that may facilitate problem understanding, improve the focus of design meetings, and provide a historical trace of deliberations. Rather than implementing component and link specifications directly into their program, the generalizations of the IBIS model achieved by gIBIS were used as a basis for a program that creates hypertexts tailored to specific applications. This metalevel program, Germ (Graphical entity relationship modeling tool), takes, as input, a “schema” consisting of specifications for component types, link types, and allowable relationships between components and links, and provides as output all information necessary to realize a hypertext appropriate for a specific specialized task (Conklin, personal communication, 1991). Therefore, Germ is an


SOCIETY FOR INFORMATION SCIENCE-March

1992

185

example of a metasystem-a program that accepts user definitions to modify the behavior of another program. Metalevel issues also influence the use of systems like gIBIS, for much of the benefit realized using these systems requires users to share common interpretations of component and link semantics, and frequently participants digress from the issues at hand to discuss how best to discuss the issues at hand-a process akin to setting the agenda for a meeting. Given hypertext’s flexibility, it seems reasonable to expect discussions about the hypertext system to take place within the hypertext system. Trellis-A Model Incorporating Semantics

Browsing

The link semantics imposed by systems like gIBIS complement the structural features defined by the Dexter model and provide a framework for the semantics of associations between hypertext components. The Trellis model’s semantics for browsing allow one to express both the possible states in a hypertext reading session and the preconditions necessary to transfer from one state to another (Stotts & Furuta, 1989). This capability makes the Trellis model extremely attractive in educational settings by ensuring that readers have access to critical hypertext components only in appropriate sequence, and by facilitating adaptation to user behavior and preferences exhibited while browsing. The Trellis model achieves its power and flexibility in two ways. First, the model is based on Petri nets; places represent possible actions such as display of a component. Transitions between places occur only when places are in a prespecified state. Second, the Trellis formalism provides for attributes -variables that can represent information about component and link parameters (Furuta & Stotts, 1989; Stotts & Furuta, 1990). With Trellis, it is possible to specify transitions that will take place after the system has remained idle in a specific state for prespecified amounts of time. For example, under the assumption that idle time represents confusion, designers can specify that a “Help” link marker (button) appears after 15 seconds of idle time and a “Help” window appears after an additional 15 seconds of idle time. (One can also specify that the Help window closes if it is ignored for a prespecified amount of time.) Specifying a value of “infinity” for both link marker and window appearance effectively removes a link from the hypertext network. The value for attributes can be changed dynamically in response to user behavior. Combining Models

Hypertext

and Information

Retrieval

The ultimate goal of the hypertext developer is to facilitate information management, not to assure adherence to any specific model for hypertext. This goal may 186

JOURNAL

OF THE AMERICAN

SOCIETY FOR INFORMATION

best be realized by integrating models for hypertext with information retrieval models commonly applied to full-text document collections. Hypertext systems that emphasize link-based browsing and full-text document retrieval systems that emphasize index-based access play complementary roles in information management. Indexing hypertext components identifies components not associated by a generally sparse collection of links (Coombs, 1990). Adding hypertext links to a conventional full-text document retrieval system allows for representation of relationships not identified by indexes and term co-occurrence. Combining hypertext links with index-based retrieval mechanisms provides a new approach to the information seeking process (Marchionini, 1992, p. 156). Problem articulation is performed by matching query terms with items in a global index of an entire document collection; examination of results is accomplished by pursuing links denoting inter- and intradocument relationships that are not represented in the index. Hypertext links can also be used to order information derived through an analysis of component content. Content can be used to classify components by source, time, scope, degree to which specialized terminology is used, and by the perceived value to various groups as determined by authors and previous readers. With this information, hypertext links could be combined into collections or “webs” useful for specific tasks, including: (1) identification of items contributed by specific authors or groups, allowing one to remain current in specific areas; (2) establishing chronological relationships between components, allowing an author the opportunity to provide a historical perspective, and allowing the reader to review the most recent components first; (3) establishing an ordering from the most general to the most specific, allowing a reader to examine items that most appropriately meet his or her level of experience; (4) identifying component clusters using term co-occurrence techniques, facilitating discovery of new items of interest; and (5) joining components that domain experts believe are most relevant to a defined line of inquiry, allowing novices “guided tours” of a specific domain (Marshall & Irish, 1989) (Fig. 3). No matter how links are established, link selection remains a “guessing game,” where the ultimate result cannot be determined until one actually chooses a link; the number of irrelevant selections presented must be minimized. Accordingly, it is crucial to consider both the content of the link marker and the con&t in which the link marker is placed (Landow, 1987). It is also important to perform the actions dictated by link marker selection as rapidly as possible (Akscyn, McCracken, & Yoder, 1988), and it is equally critical to provide an easy way for users to “go back” by reversing the results of a link marker selection. Hypertext users, deprived of many visual and tactile cues afforded traditional paper documents, must be provided with compensatory navigational aids that SCIENCE-March

1992

FIG. 3. Guided tours. Guided tours are a facility to organize hypertext networks for presentation. Material added about the tour is presented in a narrative fashion through the use of visual cues like large bold fonts. Reprinted with permission of the ACM.

allow them to retain a sense of “place” within the hypertext. Link markers and backtracking mechanisms are one example of a navigational aid; they act as signposts that tell the user what is immediately ahead or what has just been passed. “Maps” or webs are a second form of navigational aid; they can consist of prespecified orderings, or they can be general-purpose graphical browsers. Ideally, one would prefer an intelligent “guide” whose behavior could change as the needs of the user changed (Weyer & Borning, 1985). Prototypic guides exploit user models, domain models, and user feedback to alter the apparent structure and behavior of the document; structural alterations are achieved by hiding or making visible various links within the document; behavioral adaptations are exhibited by changing link sequence timing, linking sequence representation, and providing automated help. The value of each navigational aidsignposts, maps, and guides-varies with circumstance: Signposts may be of most value when the content of the hypertext is well understood and the search goal is well defined; maps may be of use when the content includes a relatively small number of components that can be arranged in a coherent and logical way; and guides can be of most value when the domain is broad, the hypertext size is large, and the needs of the user are not well defined (Frisse & Cousins, 1990). Despite impressive gains, retrieval performance in systems combining hypertext browsing and query-based searching do not always live up to users expectations

(Halasz, 1988). This sense of dissatisfaction is increased by the manner in which conventional documents are decomposed into smaller units. When this process occurs, very often related terms appearing on a single document and displayed prominently during the retrieval process are distributed across several different hypertext components, eliminating both the semantic and statistical relationships that account for the performance of many retrieval algorithms (Frisse, 1988). Disappointing retrieval performance is due in part to a failure of query mechanisms to represent the semantic relationships between various components; “display all refutations for the claim that hypertext is superior to the printed book” is a query that one cannot pose in most hypertext systems (Consens & Mendelzon, 1989). Until more effective means of expressing component context are developed, retrieval performance of a specific hypertext system applied to a specific task can be determined only through careful empirical measurement. Opportunities

for Hypertext

The range of published hypertext models allow for great latitude in the construction and use of programs with hypertext functionality. At one extreme, systems can consist of conventional full-text documents and interdocument links, denoting document similarity. At the other extreme, systems can consist of small units of text, a prespecified set of component and link types,


SOCIETY FOR INFORMATION SCIENCE-March

1992

187

and formal rules declaring preconditions that must be satisfied before text can be displayed. Each extremeand many points in between-can serve as a useful representation for applying hypertext to some pressing information management need. Perhaps the most dominant expression of hypertext models can be found in the hypermedia systems prevalent in interactive learning environments. Perseus, discussed in another article in this Perspectives issue (Mylonas, 1992), is an excellent example of a system whose realization required a significant investment in resources; it is a hypermedia system consisting of heavily cross-referenced primary Greek literature, commentaries, dictionaries, reference grammars, maps, and illustrations. Production of less ambitious systems has

.

accept

l

prompt-and-accept

.

accept-from-string

l

dwmenu-choose

l

dw:menu-choose-from-set

become a “cottage industry” (Frisse, 1990); effective educational programs can be constructed with inexpensive authoring environments like Apple’s HyperCard; examples of these systems flourish at many educational conferences. As technology evolves and the range of component types grows, self-contained hypermedia systems will be replaced by operating systems that exhibit the ability to support anchors and links across a broad range of applications. The SUN Link Service (Pearl, 1989) is one example of such a system. Apple Computer’s System 7.0 balloon help and QuickTime multimedia standards also exemplify hypermedia support at the operating system level. Hypertext features are also playing a prominent role in supporting reference works. The New Oxford English

The primary faclllty for accepting Input of presentations Is the Symbollcs Common Lisp function accept. Presentations can be accepted via keyboard or mouse Input. Characters typed in at the keyboard In response to a” accept prompt are parsed, and the presentation they represent is returned to the calling function. AlternatIvely, If a presentatlon of the type specified by the accept call has previously been dlsplayed, the user can click on It with the mouse and accept returns It directly (that Is, no parsing is required). Examples: (accept ‘(string) Enter a string: ‘abracadabra’ STRING (accept

‘string)

Enter a string ‘abracadabra’ STRING

==> abracadabra .

D-CHRRRCTER Function Displaying Pronpts in the Input Editor BULTIPLE-URLUE-SET0 Function LOOP Initiallration ElndlnSs Section UNTIL Keyuord for LOOP Section RRPCRR Function File-Openins Options Sectjon COPY-LIST Function PRONPT-RND-RERD Fun&lo” ACCEPT Function AcceptinS Single Objects Scctlon

==> [default

abracadabra]:

abracadabra

FIG. 4. Screen display of the symbolics document examiner. The screen is divided into four panes. The largest pane (upper I@) displays documentation about “Accepting Single Objects” in a “booklike” manner. The lower pane contains both a fixed command menu (lower right) and a command interactor area for user-input (lower leff). The upper right pane displays a list of “Current Candidates”-topics that appear relevant to the reader’s request for “Documentation [about] Accepting Single Objects.” Beneath the “Current Candidates” pane is the “Bookmarks” pane. This area displays a chronological list of topics that the user has read in a session. Candidates, Bookmarks, and some portions of the documentation are mouse sensitive, so that appropriate contents can be displayed quickly and easily. The Document Examiner-an independent window-based program that is closely integrated with the rest of the Symbolics Genera Environment-is discussed in detail by Walker and Bryan (1988). (Illustration courtesy of Janet Walker).

188

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE-March

1992

Section

John P. Mi&hdl

General Care of the Rtient

I.

Hospital

Orders

A. InltW Orda should be wriltcn u soon .I possible following rdmimion md cwhution of the p&m Each set d orden rhould bcu the date md time wrhtcm md the legible rignrtwe of the physicim. All ders shmld be clar, cmpletc. orgmiz.4, md legible B. Or&m

should be reevaluated

u ftqumt

intervals

B. O&xv hould be mmulw C. Oders for Mediutiom to D. Cmlmt md Orgtiticm 1. Admittiq dir~mis md

ad lltasd w the prtiart’l MU dictatu.

when changing sn orda. Ihe old orda should be #pccifmlly ancelled before the new o&x in wriaea.

ttcd Q limituimr

C. Orders for medkAkm to be laken as needed (pm) require cudol considastioo. especially chore ccllccming the dnlinistntim of u1 opiod nucmc uulgaic. Dosing intawll muIt be lpecifisd (eg.. qQh). ian of orders. ENbhbing . mutine for wriIing ordan (q.. the mnemonic SE ovcrlwk~. An example muue thr no impatxot lhapmtic m-

Fmm1dImmdwtimtoWALT ---, himlul d MedlcJl nlmpeutia

xxx Sum-dsy.Fchruq 23.19912:W M From1 ofMmtulofMcdicrlThenpeth --a Allergia md Sasitivitia From 2345 Allqia md Scmitivitia

---a Searchfor termGbmerulmeprhltk Weighted

0 All Books

0

OR

(BNSearch)

(paste)

@

FmmX345Alleqlaad.ksitiviUm ---> 11. Rmll Dh-

FIG. 5. WALT-a representative system combining hypertext models with an “Electronic Book” interface. The WALT interface is dominated by a Browsing Pane used to examine documents. To the right of the Browsing Pane are the Book Shelf (with its associated title and chapter fields) and the Book Spine; these panes are used for selecting items to be examined in the Browsing Pane. The remainder of the right-hand portion of the interface is relegated to the Table of Contents and the Path Clipboard. Along the bottom of the Browsing Pane are the Control Pane and the Reader Feedback Pane. Like the Document Examiner, contents of panes are mouse sensitive and support hypertext link mechanisms.

Dictionary presents an outstanding example of how conventional text can be structured and used in a reference setting; it presents a combination of link-based browsing and free-text search capabilities that is becoming prevalent in CD-ROM reference systems (Raymond & Tompa, 1988). The Symbolics Document Examiner is perhaps the first large hypertext reference and demonstrates the value of this technology in an applications programming environment (Walker & Bryan, 1988) (Fig. 4). Its book-like interface is representative of a class of information management systems that includes SuperBook (Egan et al., 1991), the KMS implementation of the ACM Hypertext Compendium, WALT (Frisse & Cousins, 1991) (Fig. 5), and a number of commercial systems. These systems complement the book metaphor with some combination of links, full-text indexing, rapid access to cross-references, and history traces. These systems employ a relatively standard structural model for components and links, but differ

significantly in their models for rhetoric and browsing semantics and the manner in which they integrate more conventional full-text document retrieval techniques. These differences simplify tailoring each hypertext to its application domain, but make it far more difficult to promote hypertext document interchange. Hypertext support for collaboration, once a somewhat arcane subject, is moving into the mainstream as both operating systems supporting link services and networks supporting access become more widely available. Systems like gIBIS are moving into the commercial marketplace to play a role complementary to those played by products designed to facilitate coordination (Flores et al., 1988), or to communicate and order semistructured electronic mail messages (Lai, Malone, & Yu, 1988). In 1968, Douglas Engelbart profoundly influenced the information science community by demonstrating at the Fall Joint Computer Conference a hypertext sys-

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATIONSCIENCE-March

1992

189

tern supporting collaborators separated by geography, but united by a common network and human-computer interface (Nielsen, 1990; Rada & Lunin, 1989). Realization of this tour deform required specialized networks, person-years of programming effort, technologic vision, and considerable amounts of good luck. Now it is possible to replicate much of this demonstration using household phone lines, personal computers, and computer programs widely available in software supermarkets, suggesting that the themes of hypertext-once technologic wonders observed only in specialized settings using custom programs-have been incorporated into a growing list of features expected from personal computer software. Software is expected to support integration of components stored on different computer systems and contributed by more than one author. Changes in the established correspondence between document authoring, storage, and use produced by these new systems will require new approaches to issues as far ranging as copyright law and the common restrictive perception of a library as a physical place where books are held. But many challenges remain if the true potential for hypertext is to be realized. Coherence and effective communication will require us to understand the medium at a deeper level, matching appropriate models for hypertext with our information management needs.

Acknowledgments

This work was supported in part by equipment grants from Apple Computer, Inc., and by grants from the Center for Intelligent Computing at Washington University and Southwestern Bell Technology Resources, Inc. We particularly thank Steve Weyer and Tim Oren for their support and contributions. Charles Mead and Michael Kahn reviewed early drafts of this paper. Dr. Frisse is a Teaching and Research Scholar of the American College of Physicians.

References Akscyn, R. M., McCracken, D. L., & Yoder, E. A. (1988). KMS: A distributed hypermedia system for managing knowledge in organizations, Communications of the ACM, 31, 820-835. Conklin, J. (1987). Hypertext: an introduction and survey, Computer, 20, 17-41. Conklin, J. & Begeman, M. L. (1989). gIBIS: A tool for all reasons, Journal of the American

Society for Information

Science, 40, 200-

213. Consens, M. P. & Mendelzon, A. 0. (1989). Expressing structural queries in GraphLog. In N. Meyrowitz (Ed.), Proceedings of Hypertext ‘89 (pp. 269-292). Pittsburgh: ACM Press. Coombs, J. J. (1990, September). Hypertext, full text, and automatic linking. In J.-L. Vidick (Ed.), Proceedings of International Conference on Research and Development in Information Retrieval (SIGIR ‘90), (pp. 83-98). Brussels: Presses Universitaires de

Bruxelles. Egan, D. E., Lesk, M. E., Ketchum, D., Lochbaum, C.C., Remde, J. R., Littman, M., & Landauer, T. K. (1991). Hypertext for the

190


SOCIETY FOR INFORMATION

electronic library’? CORE sample results. In R. Furuta and D. Stotts (Eds.), Proceedings of Hypertexf ‘91 (pp. 299-312). San Antonio, TX: ACM Press. Flores, F., Graves, M., Hartfield, B., & Winograd, T. (1988). Computer systems and the design of organizational interaction. ACM Transactions on Office Information Systems, 6, 153-172. Frisse, M. E. (1990). The case for hypermedia. Academic Medicine, 60, 17-20. Frisse, M. E. (1988). Searching for information in a hypertext medical handbook. Communications of the Association for Computing Machinery, 31, 880-886.

Frisse, M. E. & Cousins, S. B. (1990). Guides for hypertext: An overview. Artificial intelligence in Medicine, 2, 303-314. Frisse, M. E. & Cousins, S. B. (1992). Information retrievyl using a “digital book shelf.” In P. Clayton (Ed.), Proceedings of the Symposium for Computer Applications in Medical Care (pp. 803-807). Washington, DC: McGraw-Hill. Furuta, R. & Stotts, P. D. (1989). Programmable browsing semantics in trellis. In N. Meyrowitz (Ed.), Proceedings of Hypertext ‘89 (pp. 27-42). Pittsburgh: ACM Press. Halasz, F. & Schwartz, M. (1989, January). The Dexter hypertext reference model. Proceedings of NIST Hypertext Standards Workshop, (NIST Publication 500-178 [95-1331. Gaithersburg, MD: National Institute of Standards and Technology. Halasz, F. G. (1988). Reflections on Notecards: Seven issues for the next generation of hypermedia systems. Communications of the ACM, 31, 836-855.

Irish, P. & Trigg, R. (1989). Supporting collaboration in hypermedia: Issues and experiences. Journal of the American Society for Information Science, 40, 192-199. Lai, K.T., Malone, T.W., & Yu, K. (1988). Object Lens: A spreadsheet for cooperative work, Transactions of Office Information Systems, 332-353.

Landow, G. (1987). Relationally encoded links and the rhetoric of hypertext. In F. Halasz (Ed.), Proceedings of Hypertext ‘82 (pp. 331-344). Chapel Hill, NC: ACM Press. Marchionini, G. (1992). Interfaces for end-user information seeking. Journal of the American Society for Information Science, 43, p. 156. Marshall, C. C. & Irish, P. M. (1989). Guided tours and on-line presentations: How authors make existing hypertext intelligible for readers. In N. Meyrowitz (Ed.), Proceedings of Hypertext ‘89 (pp. 15-26). Pittsburgh: ACM Press. Mylonas, E. (1992). An interface to classical Greek civilization. Journal of the American Society for Information Science 43, p. 192. Nelson, T. H. (1987). All for one and one for all (invited talk). In F. Halasz (Ed.), Proceedings of Hypertext ‘82 (pp. v-vii). Chapel Hill, NC: ACM Press. Nielsen, J. (1990). Hypertext & hypermedia. Boston: Academic Press. Pearl, A. (1989). SUN’s Link Service: A protocol for opening linking. In N. Meyrowitz (Ed.), Proceedings of Hypertext ‘89 (pp. 137146). Pittsburgh: ACM Press. Rada, R. & Lunin, L. F. (1989). Perspectives on hypertext: Introduction and Overview, Journal of the American Society for Information Science, 40, 159-163. Raymond, D. R. & Tompa, F.W. (1988). Hypertext and the Oxford English Dictionary. Communications of the ACM, 31, 871-877. Scacchi, W. (1989). On the power of domain-specific hypertext environments. 40, 184-190. Schnase, J. L., Leggett, J., Kacmar, C., & Boyle, C. (1988). A comparison of hypertext systems (Tech. Rep. No. 88-017) College Station, TX: Department of Computer Science, Texas A&M University. Stotts, P. D. & Furuta, R. (1989). Petri-net-based hypertext: Document structure with browsing semantics. ACM Transactions on Information Systems, 7 3-29.

Stotts, P.D. & Furuta,

R. (1990). Temporal hyperprogramming,

Journal of Visual Languages and Computing,

SCIENCE-March

1992

1, 237-253.

Walker, J. H. (1987). Document Examiner: Delivery interface for hypertext documents. In F. Halasz (Ed.), Proceedings of Hypertext ‘87 (pp. 307-323). Chapel Hill, NC: ACM Press. Walker, J. H. & Bryan, R. E. (1988). Supporting document development with Concordia. Computer, 21, 48-60.

Weyer, S. & Borning, A. (1985). A prototype electronic encyclopedia. ACM Transactions of Office Information Systems, 3, 63-88. Yankelovich, N., Haan, B., Meyrowitz, N., & Drucker, S. (1988). Intermedia: The concept and construction of a seamless information environment. Computer, 21, 81-96.

JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATIONSCIENCE--March

1992

191

f., ‘:i..

c:

‘5. , ‘; ’ fi : li ii (’ >I. i ‘, i ,((:: il. ‘i,( (,‘” 8‘ j::si, : F’ ‘: (I! i :i ::j), (( L:,, $ +; 2’:: ,: ,&(. ‘,, ( : : >,“8: ,, (‘(< c‘:n::_l ;,;g: b 0 I_,:2 ” ,:l “:i,,: ‘S,i ? i+;: F;,,‘i.!: ,e, ?.‘,” ‘:,.; :::: ~: ~(8 :: 1,’