[extra dimensions]Hypertext is a popular term with an unclear meaning. In the literal sense the term implies extra dimensions to text; in practice, the term is often applied to computer systems which allow a person to browse a document by gracefully jumping from text block to text block. Yet this `browse' dimension alone is limiting. The argument advanced here is that hypertext is most profitably viewed as a combination of dimensions that extends across large document collections, across collaborative work, and across artificial intelligence. One must start with an understanding of text and from there develop the perspectives that will allow the specification of new hypertext systems for the 1990s.
[links]Text must be linked to other text and to people. The holistic view of hypertext focuses on links: links within a document (microtext), links among documents (macrotext), links among people (grouptext), and dynamic links (expertext). The principles and systems which are relevant to creating and accessing hypertext can be usefully presented under the headings of text, microtext, macrotext, grouptext, and expertext (see Exercise "Integrated").
[definition of text]Hypertext is an extension of text, but to understand hypertext, one must first understand text. Text is a recorded body of information. The terms text and document are synonymous, and while text contains predominantly natural, written language, it may contain graphics.
[history]Six thousand years ago in the cradles of civilization, people were recording information on clay or papyrus. Thousands of years passed before significant libraries were created. The introduction of the printing press in the middle ages took a century to influence the distribution of documents. New technologies for text take root gradually.
[reading and coherence]The words that form a text are the raw material from which a mental representation of the meaning of that text is first constructed. Local coherence is then established across sentence boundaries. Language users must establish coherence quickly. The language user makes preliminary hypotheses about local coherence from the title and first words of the text and from knowledge about global situations.
[readability]Readability means the relative ease with which texts can be read and remembered. Traditional readability models deal primarily with surface variables, such as sentence length. An adequate model of readability must account for the cost of constructing complex memory structures.
[writing process]A good writer must produce a coherent text, but the process of writing is otherwise not necessarily like the process of reading. The three phases of writing a document are exploring, organizing, and encoding. In paper these correspond to notes, outlines, and prose, respectively. Authors like to move freely from one phase to another and are influenced by their own memory, the writing assignment, and the evolving text.
[layout language]To convert the abstract form of a text into a concrete visualization the author needs a layout language. The languages for the specification of layout come in two generic forms. One requires commands embedded within text, while the alternative involves the direct manipulation of the physical appearance with a `What You See Is What You Get' environment.
[writing tools]Tangibility is important to readers, and the tangibility of paper is greater than that of computer screens. While reading is typically easier on paper than on computer, the same does not apply to writing. The features of a computer tool for writing are often more attractive than that of a paper and pen tool. Writing tools have evolved from editors to word processors to outliners to desktop publishers. Outliners help the writer organize thoughts and attach text to the thoughts. Desktop publishing systems facilitate the addition of graphics and the layout of text.
[logical structure]To make electronic information more exchangeable, standards of logical document structure are useful. The Standard Generalized Markup Language is a language for logical document structure and is an international standard for publishing. It is based on the principles of generic encoding of documents and marks a document's logical structure, such as section headings, and not the document's physical presentation.
[definition of microtext]Microtext is text with explicit links among its components. A microtext system provides a computer medium for manipulating the links of microtext and supports authoring and browsing, whereas a macrotext system principally supports searching. The first microtext system, the Augmentation System, was developed in the 1960s, but HyperCard, the first widely-popular, microtext system, did not appear till the late 1980s.
[levels of description]Microtext may be described from various levels, including screen, internal, and logical. What the user sees on the screen is often a combination of windows and buttons, and selection of a button often causes a jump to another window. Inside the computer the information may be stored as files whose markup indicates the actions which the computer program can take on the files, or the information may be stored in some more structured form, such as a relational database. The computer internal representation may also be viewed from a logical perspective. The logical perspective addresses the conceptual units of the microtext and their relations. This logical structure is often described in terms of nodes, links, and attributes.
[independent versus embedded semantic net]The network of microtext may be viewed as a semantic net which models human memory. The nodes and links of the net are labeled with terms that represent concepts. The semantic net of microtext may be independent or embedded. In the independent case, the nodes and links are tagged with terms that represent concepts (per the usual semantic net), but each node or link may point to text blocks. When a user is presented with an independent semantic net microtext, he may traverse the semantic net without seeing a text block. In the embedded case, a text block is at the end of a link. In traversing an embedded semantic net, the user must visit text blocks.
[appropriateness depends on task and user]The principles of microtext must also cover the psychological or ergonomic issues of human-computer interaction. A novice user wants clear options and simple instructions, while an expert user may place a premium on a powerful command system. For certain tasks and certain users, certain microtext strategies are appropriate. For instance, the task of guiding lay people to the resources of a city via hypertext may benefit from the use of tourist metaphors. An example of a tourist metaphor is an icon of a tour bus which when selected leads the user through a tour of the hypertext.
[popular systems]All of the popular microtext systems are of the embedded, rather than independent, semantic net type. NoteCards, HyperCard, and HyperTies emphasize the connecting of one text block to another. The user selects a highlighted term in one text block in order to see another text block. Guide, KMS, and Emacs-Info follow the model of highlighted terms in one text block pointing to other text blocks, but these microtext systems indirectly provide an independent semantic net. In Guide, KMS, and Emacs-INFO outlines are critically important, and these have an implicit independent semantic net character. An outline is a restricted type of semantic net.
[outlines]The importance of giving users an overview of the network of concepts was not lost to the developers of NoteCards. NoteCards provides special cards with a semantic net whose nodes point to text blocks. The developers of HyperTies have realized the importance of outlines and are incorporating features to facilitate the managing of outlines.
[ clear versus implicit structure]If microtext is to be successful, the ability to translate easily between text and microtext is essential. Two classes of text should be distinguished: clearly-structured and implicitly-structured. The links of clearly-structured text can be readily extracted from the markup commands in the text, while the links of implicitly-structured text must be extracted manually. Examples of clearly-structured text are technical manuals, dictionaries, encyclopedias, and course catalogs. Essays and novels are examples of implicitly-structured text; the logical structure is not suggested by markup commands in the text.
[traversals]Converting text to microtext has immediate economic importance because of the vast amounts of text that could be treated in this way. The reverse process of converting microtext to text will become more important as the amount of available microtext increases. A traversal of the network of a microtext that prints every text block once, and only once, has effectively linearized the text. For a hierarchically structured microtext like Guide or KMS, translation of microtext to text is straightforward. The hierarchy is traversed in a depth-first fashion. If the hierarchy was designed with a sequence in mind, as is often the case, the resulting sequence tends to make sense. For a complex, non-hierarchical network a meaningful traversal is less obvious.
[definition of macrotext]Macrotext is essentially hypertext, with emphasis on the links that exist among many documents, rather than within one document. Typically, many people have contributed documents to macrotext and an institution maintains a macrotext system. Both the interface to the document collection and the links among documents must be maintained. The user is searching for a few documents from a large set. A macrotext system doesn't support the browsing of a single documentthat is a microtext system facility. Traditionally, the term `information storage and retrieval' was used when talking about storing and retrieving many documents.
[history]Vannevar Bush argued as early as the 1930s for the importance of using modern technology to turn text into macrotext. He described a scenario in which one individual dealt with the text of many other individuals by placing connections among the text items. The National Library of Medicine in the United States of America built the first major macrotext system in the mid-1960s; in its first year it stored citations for over 100,000 documents and handled several thousand queries.
[queries and documents]If the language of queries and documents is not unified at some point, a match can not be made between queries and documents. The traditional way of representing the content of documents in a macrotext system is to label each document with a handful of terms from an indexing language. An alternative strategy is to represent a document by the words which occur with significant frequency within the document.
[thesaurus]One of the most popular forms for an indexing language is a thesaurus. While in the lay use the term suggests an alphabetically-sorted list of terms with attached synonyms, in the context of macrotext a thesaurus is far more. A thesaurus is a set of concepts, each represented by synonymous terms, broader concepts, narrower concepts, and related concepts. The broader and narrower concepts form a hierarchical semantic net. This hierarchy helps users find terms to represent their query.
[query logic]In the simplest query, the user presents a single term to the system, which returns those documents which have been indexed with that term. More generally, index terms are combined with operators, such as AND. If one asks for all the documents that have been indexed with the term `optical disk' AND `encyclopedia', one would get the documents which had been assigned both of the index terms. This notion of operators between terms of the semantic net to support querying is not part of the microtext system armamentarium. A macrotext system will also typically allow a user to request all documents indexed by a term or indexed by any of the terms narrower than that term in the thesaurusthis is another instance of exploiting the hierarchical structure of the indexing language.
[word patterns]To maximally exploit the computer, methods have been developed to index and retrieve documents based on word patterns. A document may be perceived as being about the subject symbolized by a certain word, if that word occurs more frequently in that document than could be expected in a randomly chosen document. To reduce search time, documents may be clustered according to their word patterns. By then clustering the clusters, a hierarchy is created. A search can thus start at the root cluster and proceed in a depth-first manner to the cluster which best matches the query.
[thesaurus presentation]When a very large thesaurus is accessed via a computer screen, the user may need to change the contents of the screen many times to find the terms for a query. As the number of hierarchical levels increases, users take longer and make more errors. Accordingly, parallel modes of presentation should be provided so that users can get quickly and easily to the places they desire. These insights about the user interface for macrotext systems also apply to microtext systems.
[growth in systems]The 1970s witnessed a rapid growth in the availability of macrotext via telephone lines and time-sharing computers. By the late 1970s over 300 macrotext systems were operating, providing access to over 60 million document citations and processing over 5 million queries a year. In the 1980s optical disks became cheap enough to impact substantially on the method of delivering macrotext. Now, an entire macrotext system can be supported by one personal computer with an optical disk.
[MEDLINE history]The National Library of Medicine MEDLINE system which was developed in the 1960s remains one of the salient macrotext systems, and now contains citations to over 6 million biomedical articles. While it remains accessible via telephone lines, optical disk versions of its database are becoming increasingly popular.
[combining micro and macro]One of the latest developments in macrotext is the merging of macrotext with microtext. The IDEX system allows users to search massive document databases through a thesaurus and by author, title, and date of publication. Additionally, once a document is found which seems interesting, the user can stay in the same system with the same interface and proceed to browse the entire document with the Guide microtext system. Several groups are exploring the combination of word-frequency strategies of macrotext with network strategies of microtext. In such systems the user first locates a block of text within a document by searching for a text block that has a certain pattern of words in it. Having found such a text block, the user then browses from that point along the network of the document.
[macrotext to macrotext]Whereas for microtext the connection of text to microtext is a current research problem, methods of automatically or systematically moving texts into macrotext have been well-established. In many ways, macrotext systems have features which microtext systems may eventually emulate. The most fascinating of these features concerns the connection of one body of text to another. The Vocabulary Switching System was developed in the early 1980s to connect the indexing languages of 15 different macrotext systems. This system depends largely on lexical matching of terms. The United Nations, the American National Institutes of Health, and many other organizations have sponsored work to connect indexing languages. No comparable effort has occurred within the microtext arena, but the dream of hypertext is to connect text across more than one document boundary. One can predict that macrotext methods will be explored for connecting one microtext to another.
[definition of grouptext]Grouptext is text which is created or accessed by several people collaboratively. A database is needed to keep track of which author made what changes to the text and when. Collaborators may want to communicate synchronously or asynchronously, and screen sharing is critical to synchronous communication.
[compelling argument]One compelling argument for collaborative writing is that good writing requires an understanding of the reader, and a collaborating author can serve as a reader. Expert writers differ from novice writers in the extent to which they have sophisticated models of the reader. Experiences with children suggest that their writing and reading abilities improve more quickly under collaborative than under solo conditions.
[databases and social mediation]Whereas in traditional databases a primary concern is to lock a record during use by one person so that it cannot be changed by other persons, at the same time, the character of transactions for authoring may call for a different paradigm. Rather than a central server controlling all events, each user can have a copy of the database on his workstation, and his changes to the database may be broadcast from his workstation to a central server which subsequently checks for conflicts. In the case of conflict, the central server notifies those who participated in the conflict, and they then must discuss among themselves how to resolve the conflict. This process of social mediation is contrary to the norm for databases, but supports very rapid access for each user to arbitrary amounts of text.
[synchronous communication]In collaborative work another key ingredient is synchronous communication. The speed of electricity is so fast compared to the speed of human senses that a local network of computers can give every user the sense of sharing the same screen. Such What You See Is What I See screen treatment allows action that would not be possible on paper. Two people can write on the same file at the same time and see what each is doing. Furthermore, other media can be mixed with that of the computer. Video and audio can be carried with the data signals and allow people to see and hear one another simultaneously as they share the workspace of the screen.
[physical proximity]Despite the flexibility of high technology to simulate the sense of being together without being physically close, studies suggest that people find physical proximity to be especially important in creating and maintaining collaborative efforts. New methods are needed to assess the impact of electronic media on work and to determine the conditions under which the computer can help people work together. There are many examples, such as the PicturePhone, where although technologists thought society ready for a new medium or way of work, people preferred the traditional methods.
[annotation]One analog of reading in the grouptext domain is annotation. As a group of people read a document on the computer, they can add notes that explain their reaction to the text. Not only might this help the authors revise the document, it might also help other readers appreciate the various messages which the document contains. Group annotations on paper are difficult to manage, but with the computer many people can make simultaneous comments, which others can elect either to see, or not see. Furthermore, the look and feel of paper can be simulated with modern workstation software that allows the annotator to have a red pen and to write on the document. These annotations are, however, stored separately from the document, and any reader can elect to see them merged in various ways with the document or to ignore them entirely.
[definition of expertext]While links within and among documents are critical to hypertext, its future may depend on links that support complex computations. This computation capability converts hypertext into expertext as expertise is incorporated in the hypertext. The Dynabook microtext system of the 1970s had dynamic links; the user could cause different information to appear within a text block as a function of the traversal followed to that text block. Some of the most famous expert systems of the 1970s, such as MYCIN and INTERNIST, have been converted into systems that combine the facilities of expert systems and of hypertext systems.
[inheritance]The patterns in hypertext semantic networks can be exploited. One example of such a pattern is inheritance. If a node is connected by a hierarchical relation to another node, then the child node can be expected to inherit attributes of the parent node. A different kind of pattern is that exploited by spreading activation. In retrieval, one may want text blocks associated with nodes near another node. The computer can follow the links from a node and collect information from the user, based on the assumption that two connected nodes are likely to have related information which the user may want.
[logic]Logic formalisms can be incorporated into the network representation so as to give hypertext the power of logic systems. If a node corresponds to a predicate and a link to an implication, then when the predicate at the node is true, the implication is activated. Rule-based expert systems are logic systems which can be represented as networks with nodes that are predicates. By associating text with the network, one converts an expert system into an expertext system; a product called Knowledge Pro does just this. While on the one hand, the conversion of hypertext into a logic system provides inferencing capabilities, the logic approach also has its costs. Hypertext is intended to appeal to one's intuition to be simple to create and to access. Logic systems are formal and not intuitive. systems
[procedures]Procedures may be embedded in the links of hypertext. HyperCard is the most prominent example of a hypertext system that supports procedures on the links. The simplest procedure simply says to go to a certain card, but the procedures may be arbitrarily complex. For instance, the procedure may first ask the user about his background and then branch to one card or another, based on the user's response.
[knowledge sources]Expertext systems will ultimately need to reference many different knowledge sources. For instance, one knowledge source may support the parsing of natural language queries. Another may model user backgrounds and modify the interface style based on predictions of what the user wants. Yet, another may observe behavior across the population of users and make modifications to the knowledge of the system in accord with what users seem to want.
[intelligent micro-, macro-, and grouptext]Expertext can be applied to microtext, macrotext, and grouptext. Some computer-assisted instruction programs are examples of intelligent microtext. Intelligent macrotext systems may handle natural language queries and search for documents based on knowledge about those documents. Intelligent grouptext systems have been built to monitor annotations or electronic mail and to notify participants when actions of one person should lead to responses by another.
[software_engineering]Software engineering environments are a major application area for all of hypertext. For one example of microtext, the definition of a subroutine in a program may be linked to the places in the program where the subroutine is called. The program may be connected with the requirements document and the user's manual and with other programs that serve related purposesthis is macrotext. Since groups of people inevitably write, read, and modify the documents of the software life cycle, grouptext is needed. In each of these areas, knowledge can be incorporated into the system to try to increase the productivity of the software engineers. Since software engineers naturally use computers as tools anyhow, the opportunities for improving those tools by incorporating expertext capabilities is ripe.
[society]The model of an individual and his place in the world is relevant to an understanding of hypertext. The processes of an individual creating and accessing text occur within a social context. The economic analysis of hypertext also depends on such a model of the individual and his role in society.
[influence as copying strings]Text is a fundamental medium of influence. To make precise the notion of influence, assume that each person is described by a set of bit strings and each bit can be assigned a unique tag. When a string is copied, all the tags associated with its bits are also copied. Production of a document involves the copying of strings from the person into the document. When another person accesses or reads a document, the strings are copied from the document into that person, and influence occurs (see Exercise "Influence"). People who want influence will read documents that will further their own influence.
[writing for an audience]A writer must build on the structures in text which his audience is prepared to understand. Radical departures from the expected structure are unlikely to be understood. Similarly, the impact which the text is expected to have on its audience must be one which the audience was almost ready to bear anyhowotherwise, the writer won't succeed. A good writer must understand his audience.
[creating and accessing]Dealing with text invokes enormously complex psychological and social processes. In a simplified model of cognition, a person uses a process bank and a memory bank. A task which motivates text creation or access also primes memory with a network of relevant concepts.
[hypertext needs special access cues]In parsing and organizing text a person immediately needs local coherence cues. As access continues, structures that provide global coherence become important. Accessing text on paper is already a resource-consuming task. If having to deal with hypertext requires new cognitive resources and doesn't remove any of the old need for resources, then people will be unlikely to use hypertext. If hypertext searching, browsing, and reading are to be feasible, the reader must easily gain helpful cues about coherence at the local and global levelscues which wouldn't be available with text alone.
[authoring assistance]In generating and organizing a network of concepts and text, the creator also accesses that text to test how well it satisfies the task. The global structures generated in writing are not fixed properties of the memory system, but are generated on demand in some particular task context. People need help in combining locally coherent structures into globally coherent structures. By serving as an intelligent database machine, the computer may be able to provide some of this help. Furthermore, if the computer can facilitate collaboration among people, then the most potent aid to good writing will have been stimulated.
[history of mass production]Although the printing press of the 1400s is often viewed as beginning the revolution in publishing and as marking the advent of literacy, the size of libraries did not substantially change for many decades after the invention of the printing press. It was only after the methods and markets for the mass production of books were established that libraries grew substantially in size and literacy became more prevalent. The implications of this history for hypertext are that the existence of a technology does not necessarily imply its wide-spread availability.
[cost versus benefit questions]The power of hypertext comes at a price [Scac89]. The cost can be expressed in the terms of the technological and social resources that get displaced to accommodate the use of hypertext. Society has made an enormous investment in paper processing techniques. If companies abandon paper in favor of computer media, do they lose their and society's investment in paper media? A business must make substantial new investments in equipment and training to make hypertext work. Will the profits derived from hypertext more than compensate for the new start-up costs?
[coordination economies]Hypertext and electronic mail may help a group work together, but the potential of hypertext systems as coordinating mechanisms depends on the continuity of staff participation. If one person of the group does not use the new technology, then an additional channel of communication must be retained for that one person. For instance, if the best way to reach people who don't use the computer network is through a paper memo, then the cost of preparing such a memo and waiting for it to reach its audience by the normal paper routes must still be borne.
[downward mobility]Professional enrichment may emerge from mastery of hypertext techniques. However, as with text processing systems, technical staff may broaden their skill base downward through the assimilation of more clerical responsibilities. Hypertext skills may enable increased workloads but little or no new upward mobility. For instance, the professional who previously wrote a report in draft on paper and then asked the secretary to polish and distribute it, may, with the new computer resources, also feel obliged to polish the document and distribute it electronically. What has the professional gained from being able to polish and distribute so easily?
[domain-specific help]The combination of a hypertext system with a particular application provides a framework for organizing information processing work in a systematic manner. Standardized document contents and structures support the cataloging and reuse of information. The links of hypertext support the organization of work and information. This leads in turn to improved communication and productivity. A domain-specific hypertext system incorporates information about the domain and is tailored to help a particular group of users with a particular class of text-related tasks. This kind of facility has not been adequately available in pre-hypertext technology.
[novel functionality]In the late 1960s and early 1970s, programs were developed which could generate teaching materials. Part of the teaching took the form of the student answering problems set by the program. The problems were not prestored by the course designer but generated by the program. The teaching of mathematics is particularly well suited to such generative programs, since the student can generate new examples by varying the parameters of a mathematical model [Dick79]. This kind of hypertext goes beyond what a textbook can offer and illustrates a domain-specific application.
[extend to videos]The extension of hypertext to hypermedia increases the marketability. The average person is now more likely to watch television than to read a book. But computer technology for manipulating videos may build on the principles of hypertext. For instance, the libraries of videos which are now readily available, could be stored on computers, and people could search the computer database for the videos they require. Furthermore, links could be built into the video; by interacting with a computerized video projector, the viewer could jump from one part of the video to another, as might a browser of hypertext.
[information explosion]The amount of text in the world is growing rapidly. A Second World War fighter aircraft had 1,000 pages of documentation, a Korean War aircraft had 10,000 pages of documentation, and a Vietnam War aircraft had 100,000 pages of documentation. Yet, more dramatic than the growth in the size of individual documents is the growth in the total number of documents. Each individual with a desktop publishing system can now create text which looks as if it came from a professional publisherand in high-technology societies every professional has access to desktop publishing systems. Unfortunately, while creating new text may be easy, making it understandable and available remains difficult. New structures and methods are needed to deal with the information explosion.
[selection and Vannevar Bush]Vannevar Bush argued that knowledge that could not be selected was lost [Bush45]. His thoughts about selection were so key to his thinking that the following two paragraphs appeared verbatim in each of his published papers about his now famous memex (memory extended) system. The real heart of the matter of selection, however, goes deeper than a lag in the adoption of mechanisms by libraries, or a lack of development of devices for their use. Our ineptitude in getting at the record is largely caused by the artificiality of systems of indexing. When data of any sort are placed in storage, they are filed alphabetically or numerically, and information is found (when it is) by tracing down from subclass to subclass. It can be in only one place, unless duplicates are used; one has to have rules as to which path will locate it, and the rules are cumbersome. Having found one item, moreover, one has to emerge from the system and re-enter on a new path. The human mind does not work that way. It operates by association. With one item in its grasp, it snaps instantly to the next that is suggested by the association of thoughts, in accordance with some intricate web of trails carried by the cells of the brain. It has other characteristics, of course; trails that are not frequently followed are prone to fade, items are not fully permanent, memory is transitory. Yet the speed of action, the intricacy of trails, the detail of mental pictures, is awe-inspiring beyond all else in nature. Bush was explicitly thinking about a personal machine in which an individual would record associations of significance to that individual.
[idiosyncratic associations]How do different readers take advantage of the associations that the author or other readers have made? Early objections to memex anticipated this problem of idiosyncratic associations. In 1945 Bush received a letter from John Weakland containing the following questions: [Nyce89] (1) Wouldn't the fact that association patterns are thoroughly individual make a general use of the memex difficult? (2) How would the tremendous bulk of information already recorded be made usable, especially for a searcher who wants to branch into lines of thought and knowledge that are quite new and unfamiliar to him? This kind of problem anticipates one of the greatest problems of hypertext. If an individual doesn't take advantage of existing associations, then how can access to new literature be provided? If the associations of others are to be followed, then what conventions or rules will allow these associations to be commonly understood?
[classification systems needed]While Bush's dream was to connect the world's literature, his approach to this suffers from some naivety. He felt that the existing classification systems were too cumbersome for people to use. Yet, his proposal that each person build a network of connections among documents would require each person to find documents and connect them. The advantage to a standard classification and its usage in indexing text is that people looking for text can learn about the classification and then find text which they have not previously seen.
[guidance on link types]The general link in hypertext is unlabeled and doesn't give the browser enough information about what to expect. It is like a goto in computer programming and creates a similar `lost in space' or `spaghetti tangle' feeling in the reader. Hierarchical links seem particularly useful as people have heuristics for anticipating hierarchical structures. Other link structures or navigation aids are needed. Furthermore, guidance is needed as to the types of links which are appropriate for a given type of text and a particular type of audience.
[limit the number of link types in one text]A novice hypertext author may find himself introducing many different types of links and nodes. Experience suggests, however, that one document should contain a relatively small number of different styles otherwise, readers will be confused. Consider by analogy the case of typesetting. When authors start to use computerized document formatters to drive laser printers, the authors tend to use many different fonts and character sizes. Later they realize that when many typesetting changes are introduced on one page, the page looks fragmented.
[virtual structures]In authoring, a global structure is not given a priori but evolves dynamically. With current hypertext tools, however, authors may stick to inadequate structures because the effort to change the initial global structure is too large [Fisc88]. NoteCards, for instance, requires its users to segment ideas into labeled cards that are placed in fileboxes. However, a user in the early stages of writing may not know what the labels should be. Virtual links make hypertext more flexible. A link could, for example, specify the source extensionally but the destination intensionally. Thus one could link `Claim X' to the `node containing the strongest evidence for Claim X' [Hala88].
[new representations are needed]Hypertext, in its current form, is a scheme of representation largely short of supporting processes. As a consequence, the work needed to take advantage of its representation features must be expended by its users, through some division of labor between creator and accessor. Processes must be developed whereby the computer can assume a larger role in transforming the information of the author into forms that are usable by readers. processes
[guidance]One of the founders of hypertext has said that despite the positive claims made by various groups about hypertext, By and large the hypothesis remains unproven that, with little guidance, people can construct really good trails, really good webs that help them and help other readers. I think we still need to test that hypothesis in a major way [Dam88]. The question is not whether links can be helpful, but what guidance is needed in the creation and access of links. A hypertext network may look like a complex map, which people need help in navigating. This guidance could come in the form of knowledgeable-assistance, although the necessary knowledge is only partly identified [Fris90]. Some knowledge has global relevance and applies to all documents; for example, outlines should have a handful of elements under one heading. Some knowledge pertains to the structure of a class of documents; for instance, a budget report must include a section that summarizes inflow and outflow of cash. All document tools should contain knowledge of global import, while tools for dealing with one class of documents should have knowledge about that class.
[artificial intelligence versus hypertext]Artificial intelligence and hypertext have complementary strengths and weaknesses. One of the major shortcomings of artificial intelligence is that people have difficulty building knowledge bases, though, once built, these are supposed to be easy to use. A major difficulty with hypertext is that text bases are difficult for readers to browse. People create massive text bases more easily than massive knowledge bases. If people created text bases which had some knowledge base features, would access be easier? Would the extra effort of adding artificial intelligence features to the text be rewarded by an increased impact on the audience of the resulting text plus knowledge base?
[links]Links are the essence of hypertext. In microtext the links are within a document. Links among documents via an intermediary representation constitute macrotext. Links among people and among versions of text are essential to grouptext. Enlivening the links makes expertext. What are the requirements for a hypertext system that supports these links?
[tailored to user and situation]A hypertext interface should be responsive to users who have varying levels of ability. The user interface should be modeless and never report an unrecoverable error. The number of items from which to choose at any given juncture should be small. Different perspectives should be available for different situations.
[extensible like Emacs and HyperCard]Hypertext systems should have extension facilities analogous to those currently found in Emacs and Hypercard [Hala88]. Without those facilities, many potential users will discover that the problems of dealing with hypertext systems outweigh the benefits. Both Emacs and HyperCard are built around an interpreter for a general-purpose programming language that also has special features for handling hypertext objects. Moreover, the language has a kind of scalability, as simple things can be done using single commands, yet complex programs can also be written.
[Intermedia strengths and weaknesses]One of the most powerful, commercially available hypertext system is Intermedia. What are its strengths and weaknesses? Intermedia's modeless interface supports flexible manipulation of objects and links. The system supports interactions as on paper and also beyond those which one could get on paper. While discussion, authoring, annotation, searching, browsing, and reading can be done on the system, the semantic support for these activities is not specific. Intermedia also does not address the problems of translating text into hypertext or hypertext into text.
[text to hypertext]Before the potential of hypertext can be realized adequately, tools to facilitate the conversion of text to hypertext must be robust and generally available. A marked-up text only explicitly specifies some of the links important for hypertext. A text-to-hypertext conversion tool should help people add links that weren't explicit in the text. One such conversion tool uses, at least, two windows for displaying chunks of text and suggesting links between those chunks. Natural language processing tools could suggest the meaning of blocks of text and the links between blocks. While natural language processing is a notoriously difficult computational problem, the requirements for text to hypertext conversion are much more relaxed than those for many other natural language processing applications.
[semantic net]The conceptual structure and function of hypertext might be seen from the semantic net perspective. Semantic nets are simple enough to be intuitively understandable, and yet sophisticated enough to support certain aspects of hypertext. The semantic net representation may be seen as instance of an object-oriented representation and, within the broader framework of object-oriented representations, all of hypertext can be handled.
[writing]The writing process should allow notes, paragraphs, semantic nets, and outlines to be manipulated. Textual notes can be expanded into paragraphs and indexed in a semantic net, which can be traversed to create an outline. A document can be automatically generated from the outline and paragraphs, and conversely, a document with markings for section headings can be automatically translated into paragraphs and an outline.
[creation phases]A hypertext authoring system should support discussing, writing, and annotating. Particularly, for discussion and annotation, the date and author of a text block should be recorded and displayed. The network structure for the discussion and annotation phases should be handled similarly to that for the writing phase.
[access through semantic net]For reading, browsing, and searching many access points should be supported by a hypertext system. The system must respond to queries for objects based on authorship and date. The semantic network which represents the text must be eminently accessible. Users must be able to pose queries of logical combinations of terms from the semantic net. When a new document enters the system, that document's semantic net must be automatically connected with the system's existing semantic net so that retrieval can go across all documents.
[collaboration]The first generation hypertext systems included at least some support for medium to large teams of workers sharing a common computer network. Oddly enough, many of the systems of the 1980s were designed for single users and do not support collaborative work. As text is a social phenomenon, hypertext should deal with groups of people.
[evaluations]In microtext, links among small blocks of text within a document are explicitly available to users. Macrotext is a collection of documents with links among the documents. Grouptext connects people and the portions of the document which the people are modifying. Expertext has augmented links which support intelligent computations. Measures of recall and precision are well accepted as reflecting the search effectiveness of a macrotext system. No formulas for evaluating text, microtext, grouptext, or expertext are widely accepted. Perhaps this imbalance is because macrotext systems are the most mature of the computerized hypertext systems, but until rigorous evaluation standards are available for microtext, grouptext, and expertext, their developments as scientific disciplines will be stunted.
[dynamic]The hypertext systems of the future will support all of the following: an individual writing or reading a single document, a group creating or accessing a few documents, institutions creating massive libraries, and dynamic text that adapts to input. A paper document can't respond to one reader differently than it does to any other reader. But a hypertext system can collect information from the user, and generate new presentations that suit the user.
[hardware and multimedia]An image may carry a valuable message, and a video may carry more. With the improvements in electronic storage and networking, the incorporation of audio and video into hypertext has become commercially viable. One is now able to watch a show and interact with it as one would with a hypertext. Links from one part of the show go to other parts, and the viewer decides which path to take. Given the society's reliance on television, the potential influence of hypermedia is awesome.
[future]The future of hypertext depends on developments in hardware, information science, human-computer interaction, and artificial intelligence. Hardware developments determine what is physically possible. The sciences of information, human-computer interaction, and artificial intelligence provide powerful theories and software tools for hypertext. To the extent that people can control the tools, they will want to use them.
[integrated]Exercise "Integrated": Assume that an integrated microtext, macrotext, grouptext, expertext system is available. Describe possible everyday uses of this hypertext system. (30 minutes).
[influence]Exercise "Influence": A person may be described by a set of bit strings. Production of a document involves the copying of strings from a person into the document. For another person to be influenced by a document, the strings from the document must be copied into that person. Describe conditions under which three people might produce documents and be influenced by them so that one document has perpetual influence. (90 minutes).