1 Introduction
This article presents some research issues for the next generation of hypertext systems. Neither a viable alternative nor a serious competitor to the World Wide Web, hypertext research prototypes are nonetheless vitally important for its future development. To inform the development of the Web, it is on the other hand necessary to consider problems and trends surfacing in the Web community when creating new hypertext systems. The first section introduces some discrepancies of the application of hypertext research to the Web.
We will try to show in the second section why in the Web
we currently see the most problems – and thus the greatest
potential for current research – in the area of hypertext
authoring. The lack of support for authoring links and the focus on
presentation instead of structuring of information is among the reasons for
the non-hypertextual nature of the Web.
The hypertext community has developed a wealth of systems
that allow sophisticated structuring of information. But experience with
these systems has also shown that formal ordering of information is a
difficult and laborious task for users. In section three, we propose the
paradigm of indirect authoring to describe existing and evolving
approaches that might allow a larger community of Web users to construct
hypertexts, without introducing a large cognitive overhead.
Changing the way we read and write cannot be achieved
only by developing new systems that try to replace existing work practice in
the Web. A much more promising goal, also part of hypertext research since
its beginning, is to develop tools that support concrete tasks, thus
creating a diversity of specialized authoring tools that augment the
existing Web. In the fourth section, we try to identify approaches that
support the indirect authoring of hypertexts in specific domains. We also
try to collect some research questions that may help us build better
hypertext systems.
1.1 Hypertext and the Web
[1] Although the Web is far from being primitive from a user's perspective, its technological basis has often been criticized: The Web has no means to mend broken links, uses only unidirectional and generally untyped links, and cannot externally store more meaningful structures.
The history of hypertext is closely connected to the largest – and in some sense most primitive [1] – hypertext system in existence today, the Web (Cailliau and Ashman 1999). If we want to investigate the future use of hypertext, the Web is a good place to start. Looking at its history we may find reasons for both its overwhelming success and its current lack of sophisticated hypertext features.
The Web's success was partly due to NCSA Mosaic and the
novelty of its user interface: the combination of text and graphics in a
single window (Quittner and Slatalla
1998, Fenn and Maurer 1994).
Also, the hypertextual link as a means for integrating different
applications was an integral part of Berners-Lee's initial proposal ( Berners-Lee 1989). The ability
to point at external addresses, and thus to create a web of information,
constituted a critical advantage over the existing Gopher systems ( RFC 1436). The success of the
Web has also been attributed to the simplicity of setting up a server and
providing content. This simplicity was grounded in the intentionally
primitive technological foundation, e.g. the decision to tolerate broken
links immensely simplified the (technical) communication needs between Web
servers and evaded problems of security and trust.
The future of hypertext use is mainly determined by the users of hypertext,
who will decide which system might give them most benefit for their tasks.
Most users are accustomed to the current Web, and are often quite satisfied
with it. Even without trails and sophisticated links, search engines provide
fairly reliable access to information. The majority of computer users have
made contact with the Web, and client applications for the Web are installed
on almost every existing computer. The Web is firmly established –
despite repeated predictions, from its early days on, of its not-too-distant
decline and replacement through more flexible or more powerful systems (Berghel 1995).
Even in a phase when the Web was still taking off, the more structured and hypertext-like approach of the Hyper-G system was competing with the Web without success (Fenn and Maurer 1994). Its presumed benefits over the Web were not enough to persuade users to switch from their recently initiated Web presence to a new infrastructure and a Hyper-G server. While it was difficult to convince users to switch even during that phase, it seems that, today more than ever, no new hypertext system can hope to be a complete replacement for the Web. If a hypertext system can change people's work and recreational activities, it will be the evolving Web.
1.2 A Window to the Web
[2] There is, of course, an increased use of modern technologies in the Web (compare with section 3.1). The point made here is that the time for the adoption of a new technology increases with new versions (e.g. Cascading Style Sheets: CSS1 was almost immediately incorporated in major browsers, CSS2 is not fully supported despite its five years of existence, and CSS3 is taking even longer).
[3] Browsers like Opera, Mozilla and Apple Safari allow new Web pages to be opened in tabbed windows, which means that the same window can be switched between several different buffers containing Web pages. By saving a lot of screen space, this method allows the user to open many different windows at once, and thus directly supports hub-and-spoke navigation.
[4] One important reason for the slow development of browser user interfaces might be that producing browsers is not big business anymore. Microsoft Internet Explorer and Netscape Navigator, although they are regarded as commercial applications, are distributed for free and most users have come to expect these important tools to be available without charge. New companies trying to sell their product, and also open-source projects, will be careful not to completely change the browsing paradigm, as they depend on a steady stream of converts from the major browsers. What makes a browser sell are immediately noticeable features such as nice-looking skins, rendering speed, robustness in displaying Web pages or even standards-compliance, not a history mechanism that will be discovered only during longer use.
The Web's huge success story has created distinctive compatibility problems. Although the technical infrastructure is constantly being developed, the adoption of new technology lags behind [2]. In the mass medium the Web has become, publishers are eager to reach the largest possible audience. New features that are incompatible with commonly used Web Browsers are only sparingly used. Thus, for these Web applications the commonly used subset of HTTP ( Fielding et al. 1999) and HTML (Raggett et al. 1999) limits the communication channel.
This is also evident in a lack of innovation in browser
interfaces, where the visualization of and interaction with the primary
interaction tool, the hyperlink, is almost unchanged from the first Mosaic
prototypes ( Weinreich et al.
2001). For reasons of consistency and compatibility, many exciting and
interesting research studies on the augmentation of browsing interfaces
(e.g. Card et al. 1996, Kaasten and Greenberg 2001, Hightower et al. 1998 and
Bouvin et al.
2002) have not been implemented in commercial Web browsers. Some of
these studies tried to support structural browsing, similar to the initial
Hyper-G Project (Fenn and Maurer 1994),
some tried to use a different metaphor from the browser window, e.g. a book
that held individual Web pages. A recurrent theme in research is annotation
tools for the Web, but again, features from the existing research
prototypes, although technically quite stable, have no't become implemented
in standard Web browsers.
Interestingly, the most noticeable exceptions from this conservative stance
surface when the presentation of content is not sufficient. This is true
both for (a) more general needs (e.g. for video or sound): these tools can
reach a high circulation throughout the user community, and (b) small,
well-defined groups with special needs (e.g. musicians looking for sheet
music or chemists searching for the documentation of chemical compounds). In
these special cases, plug-ins extend the functionality of, but do not
replace, the Web browser. Thus, the consistent browsing experience is only
temporarily interrupted while the plug-in is in use.
In recent years, except perhaps for the invention of tabbed browsing [3], there has been no paradigmatic change in the interaction with Web pages, as browser interfaces have remained consistent and almost entirely unchanged [4].
1.3 New Faces in the Window
[5] The Java programming language might have been an alternative way of accessing distributed data through a common language. As programs written in Java could be executed on almost arbitrary platforms, it was proposed that the browser could be replaced by custom-tailored applications for specific purposes. Among the reasons that this did not come true are lack of interoperability and problems with security, but the most important reason might have been inconsistent interfaces.
[6] There is even the notion of the Web browser replacing standard software packages and becoming the only widely used user interface ( Rees 2002).
[7] The unified "Web look" has even influenced the look-and-feel of current operating systems, where you can choose to switch from the traditional double-click to the Web's single-click style. The blue underlined links have also become a recognizable feature and even influenced the design of advertising campaigns.
The monolithic browser was initially only seen as a step towards special-use applications that would allow more flexible interaction [5]. Custom applications with specifically tailored interfaces would replace the standard browser if additional functionality was be required for a certain task ( Wei 1994). But the browser was neither replaced by task-specific tools nor was new hypertext functionality added. Instead, the logic behind presentational HTML changed to accommodate the need for more flexible applications [6]. This allowed the well-known Web browser to be used for all tasks, unifying the interfaces to a degree that was not possible before [7].
In contrast to the slow evolution of Web browser
interfaces, the functionality of Web pages has changed downright since the
beginnings of the Web. This has happened almost wholly on the server side:
dynamically generated content based on user profiles (adaptive hypermedia)
or system state is widely used in applications ranging from portals to
e-commerce to e-learning. Content management systems work behind almost
every news site on the Web, and function-specific tools, such as forums,
Weblogs and collaborative hypertexts (e.g. WikiWikiWebs, cf. 3.3), have changed the publishing process in fundamental
ways.
From handcrafted static HTML (which is still in use today), most users shifted to tools that helped to organize and layout their information. While there is still no consensus on whether to keep coding HTML and CSS by hand or use a graphical tool that might reduce the readability of the underlying source code, the advantages of tools for structuring information are widely acknowledged. Content management tools allow work to be distributed between programmers, graphical artists and content providers. They separate the information from its presentation and from the logic underlying the Web application.
1.4 The Driving Desire to Publish
While the initial Web consisted of only a few servers with a list of personal
favorite sites fitting on a "home page", continuous growth and
commercialization have changed the face of the Web ( Greenstein 2000).
Before business interest in the Web increased to the level we are accustomed
to today, however, it was private users who populated the virtual space (Erickson 1996). These individuals created virtual
identities on the Web, not only by providing selected information, but also
by collecting interesting links and presenting these on their home pages.
These links made it possible to provide pointers and simultaneously convey a
message about the referrer.
A substantial part of the Web is still powered by individuals: apart from the commercial applications that have survived the dot com hysteria into the new millennium, a large part of the Web is strictly non-commercial. This includes government organizations and non-profit organizations, schools and universities and, to a surprisingly large degree, private publications. Indeed, the private desire to produce documents, texts and data for the world to read has probably been one of the driving factors for the success of the Web.
It is not limited to individuals, though.There is also a strong will to publish data in scientific communities. Libraries do not only collect information, but also publish specialized information on the Web (e.g. John 1996). Medical information is published to an increasing extent via the Web ( Curran 2002). At the Hypertext 2003 panel on future visions for hypertext, one of the most popular future visions was the publication on the Web of scientific information, e.g. data from chemical experiments, without being bound by publishers' restrictions concerning copyright and format (Murray-Rust and Rzepa 2004).
1.5 People Do not Read Hypertexts, People Cannot Write Hypertexts
[8] Nelson's original definition of hypertext as non-linear text carries the implication that normal text is linear. As complex arguments in normal text can interweave and refer to one another on several levels, this differentiation is too general (Shneiderman 1989 ).
[9] This is true to different degrees both for literary hypertexts and for mundane Web pages. In closed hypertext systems, as in adventure games, all links are explorable with enough effort. In a system of the scale of the Web this is not true anymore – this is nicely illustrated by numerous copies of the "last page of the internet" on the Web.
Although the Web has often been addressed as a hypertext system, we feel that
users of the Web are only rarely confronted with hypertexts. We define
hypertext as a text being read using hyperlinks that associatively connect
smaller passages of information [8]. Although this is a
less demanding definition than that used in the discussion of hypertext
rhetoric (Moulthrop 1991), still only a small
number of Web pages fit this description ( Miles-Board et al.
2002).
Reading hypertexts is an interactive experience – to advance in a hypertext, a reader has to choose between several alternative continuations – and as such often more work than a linear reading. A linear text specifically (and well) designed for a given reading situation will try to answer questions as they come up in the course of its narration. Hypertext repeatedly allows (and forces) the user to choose what will be displayed next for reading, which may be too much for the reader to decide. Adaptive hypermedia has developed in an attempt to reduce the cognitive overhead and the chance of "getting lost", leaving the user fewer but better suited choices. Still, reading hypertexts may leave the reader with more new questions than answers, as an associative hypertext keeps offering more choices as the reader visits more and more nodes: there is no end [9]. Thus, hypertexts will only be read and used if they offer a substantial benefit for a given task. This benefit has been demonstrated (Perlman 1990) for special applications, such as technical manuals (Marchionini and Shneiderman 1988, Yankelovich et al. 1985), and also to some extent for literary hypertexts (Gee 2001), but for day-to-day use the need for hypertexts seems to be less obvious (for a discussion of possible advantages of hypertext reading see Rouet 1992).
Even more importantly, writing (literary) hypertexts is
hard, so hard that very few people choose it as a profession. Most people
will not change their way of writing easily to accommodate the use of
hyperlinks: consciously writing hypertexts is too much of a change from
their current practice (
Charney 1994). Vannevar Bush's writings (Bush 1945)
have often been interpreted in the sense that writing hypertext is like
making associations and thus a natural task; but hypertext is not "like the
brain" (McKendree et al. 1995). In
many applications, the utility of hypertext has to be demonstrated
first.
It is also an unproven claim that hypertext technology
will have an effect on our cognitive capacity, often made in analogy with
the evolution from oral to literate cultures; and there is also no consensus
on whether a possible effect will be beneficial (Nadin
1997). It is doubtful that we will develop general, natural
abilities for writing hypertext in the near future. Writing hypertexts is
and will remain a difficult task. Thus, the question of appropriate tools
comes into focus.
2 Hypertext Authoring
[10] Playful in the beginning (e.g. the large variety of color bullets that came immediately into use), using the full possibilities of the new medium (3D fonts, animations), the design of Web pages has generally changed into something more subtle and better to read.
[11] For an example of what can be done with CSS, see http://www.csszengarden.com/ or http://www.w3.org/Style/CSS/.
In the early days of the Web, a simple-to-setup server, a text editor and
some pictures were all that was needed to create one's own hypertext. This
simplicity ?, together with a compelling new user interface (a collage of
graphics and text in a single window!) (Quittner and Slatalla 1998),
probably accounts for the immediate and overwhelming success of the Web.
While the design focus of HTML had been the structuring of data (Berners-Lee et al. 1999, Cailliau and Gillies 2000), it
soon became obvious that the user's preference was that of creating the
desired look and feel of their documents [10]. On the other
hand, commercial applications called for the interoperability of
data.
Technological developments of recent years have dealt with these concerns.
CSS ( Bos et al. 1998)
allows sophisticated styling of Web pages [11], XML ( Bray
et al. 2000) is employed as a universal tool for
information exchange. On the downside, authoring for the Web is not that
simple anymore: the new powers come at a price. The complexity could be
reduced with appropriate support, but the commonly employed tools do not
offer much help.
In this section we will try to clarify why this is an opportunity for hypertext research, instead of searching for a new killer application that could replace the Web as we know it today. Enabling larger communities of users to create hypertexts seems a much more concrete and promising task for the future.
2.1 Shortcomings of Actual Tools
An informal survey of 35 publicly available and widely used HTML editors conducted in spring 2003, including both WYSIWYG and source code editors, free and commercially sold software, indicated that linking support in these applications can only be considered very poor ( Obendorf 2003b). While some of the low-level editors listed all available attributes for anchors, they provided no help in finding meaningful values for these. The high-level site management tools were not much better: although it would be possible to sift through the Web site automatically for structural and textual data to provide help in filling the metadata gaps ( Weinreich and Lamersdorf 2000), the support for link creation was at best limited to finding target documents.
The results of these shortcomings are immediately visible
when looking at the use of attributes in HTML Web pages: in another informal
study ( DuCharme
2002), almost no metadata were found to be present, link types and even
link titles were not used (use in link tags: <0.1% for types,
<0.4% for titles, respectively). In a further study looking for
associative links on the Web (Miles-Board et
al. 2002), the authors found less than 0.075% (!) of their
random sample of Web pages to match their criteria for associative
linking.
2.2 Structuring versus Authoring
Authoring has always been a driving force behind the idea of hypertext. The hypertext visionaries Vannevar Bush, Douglas Engelbart and Theodor Nelson, all saw the structuring of information as one, perhaps the most important, application of hypertext.
Bush advocated specialist-created trails of information
(Bush 1945). His prototypical user was a scholar, creating a
historical account of medieval England. Thus, he focuses on creating a trail
through existing information, although he occasionally "inserts a comment of
his own" or writes an analysis that he then stores in the memex device.
Engelbart's main focus was the creation and manipulation of information
organized in easily accessible hierarchical structures ( Engelbart
1962). His prototypical user was a "knowledge worker". Extending Bush's
vision, Engelbart's user built an argument that, on top of its hierarchical
structure, allowed links to nodes and substructures. The system supported
different views on the data, but the hierarchy served as the basic element.
Nelson (1987), in contrast, stressed the freedom
of authors to represent the interdependence of concepts, thus creating
complex interlinked structures.
The notion of structure has been an important landmark in hypertext
discussion ever since. This discourse has often been led from the reader's
point of view. Authoring for everyday users, however, often has the function
of creating documents that, simply stated, can be read and printed by
others. The focus of attention is on published content; the publication form
is also an important factor. This is evident in the (mis)use of HTML tags to
layout documents instead of structuring them. The question of structure is
often of secondary importance to these users.
Structure may in fact be an obstacle for authors. Although missing structure,
as might be observed in the Web, may mean that a reader has greater problems
finding the desired information, demanding the provision of structure may
discourage potential authors. While the process of accessing information is
quite fragile and users may quickly turn to another information source if
they cannot find what they are looking for, the process of authoring is even
more vulnerable: if the cognitive overhead introduced by structuring draws
attention away from the task at hand, the products of writing will differ
strongly.
Thus, for systems aiming at producing hypertexts, discrete support of the authoring process, and reducing breakdowns, is critical. Widespread use of hypertext as a communications media, as called for by Bernstein (1999), cannot be answered only by developing more powerful hypertext systems that allow the sophisticated structuring of information. Instead, writing hypertexts must be made easier.
2.3 Simplifying the Creation of Hypertexts
In a technical sense, authoring for the Web is becoming more and more complex
due to the new technologies that become gradually more widely used. This is
to different degrees equally true for Web design (CSS), structuring
information (XLink, HTML 4.0 predefined semantic types) and the
implementation of advanced hypertext features (through a dynamic,
server-side implementation or the use of a public service).
[12] There are tools that specifically support the authoring of private home pages. However, their functionality is often so limited that no more than the equivalent of a business card can be produced without employing more powerful tools.
Today, available tools are often general-purpose applications. Whether the task is to create a private home page [12], an online magazine, or an e-commerce site, the tool should provide support to all kinds of authoring projects. On the other hand, we found that the current tools lack appropriate support even for linking documents, let alone typing or providing metadata. We believe it is necessary to pay close attention to the particular task that can be supported by adding functionality. The design goal should be that of producing minimally complex tools with just the required functionality; the tool used to create hypertexts on the Web should relate directly to the user's activity (Bodker 1991).
3 The Promise of Indirect Authoring of Hypertexts
It has been a repeated finding in hypertext research that authoring
hypertext, and especially authoring relations and structure, is a very
difficult task (Nanard and Nanard
1993,
Shipman and Marshall 1999). The cognitive overload created by
structuring, and the misunderstandings that can arise when typing systems
are used, prevent many authors from creating hypertexts. On the other hand,
there is also a general consensus that hypertexts are desirable ( Bernstein 1999). Rather than proposing a
novel, general solution for authoring hypertext on the Web, we would like to
promote a new approach to supporting the authoring of hypertext. Instead of
explicitly supporting structural features of hypertext, thus placing the
focus on system functionality, we would like to name some approaches that
focus on the use of hypertext by users.
[13] A quote from the Hypertext 2003 discussion was "Hypertext is what is in the mind (between the user and the interface)". This holds true only for readers of a hypertext. A writer can well produce a hypertext without having the interlinking connections in her mind before starting to write. So the hypertext evolves with and through writing, even if the author does not consciously focus on the interdependencies.
An obvious solution is to make the authoring easy by distracting the user from the fact that she is authoring a hypertext [13], a complex interlinked network of nodes. To enable the user to master the complex task of building a hypertext, she must not be required to map the entire hypertext space in advance. As the main idea is that links are built through work processes, this could be called implicit linking. Good examples of approaches building on this idea are the generic link ( Carr et al. 1995) and spatial hypertexts (Marshall and Shipman 1995).
As an example for existing and emerging hypertexts, three
established and three emerging approaches to authoring hypertexts are
presented here in more detail.
3.1 The Generic Link
In its early days, with its very limited number of possible sites, the Web
could be made accessible by "home pages", link lists that mentioned
interesting resources for a particular topic. Today's Web could not be
imagined without search engines. One use for search engines is to make up
for non-existing associative links, finding information that is only
textually mentioned in Web pages. Thus, this first approach to hypertext
authoring is not to create hypertext at all.
[14] These include Mozilla and its derivatives, Opera, the Apple Safari browser and also Microsoft Internet Explorer with the Google toolbar installed.
An increasing number of browsers [14] allow the selection of specific words in a Web page and, through a context menu, the automatic initiation of a query. This feature closely matches the simplest form of a generic link ( Hall et al. 1992): thus, this well-known feature of hypertext systems has entered the Web without attracting much attention. In contrast, an earlier attempt to introduce the generic link feature – this time controlled only by a single company, Microsoft – unleashed a storm of protest throughout the user community ( Hughes and Carr 2002, Kaminsky 2001). It should also be noted that an important feature of the generic link is missing from the Web: permanent external storage. Although search engines can find useful information, they will not find all available information, and the result set will vary at different times.
3.2 Associative Hypertext by Keywords: WikiWikiWebs
[15] The first Wiki (http://c2.com/ppr) played an important role for the people behind the pattern movement in software technology.
[16] Some of the nodes in a WikiWikiWeb are often adopted by individual authors, who carefully look after "their" nodes and function as both experts and editors for their favorite topics.
The success story of WikiWikiWebs started with a group of people who were very much interested in developing their ideas into a common theory and structuring a complex topic [15] (Leuf and Cunningham 2001). This was supported by using a novel and innovative form of authoring in these systems. First, everyone could edit everything; the idea of collaboration was present in a pure and unrestricted form. Second, and of more consequence to this discussion, a web of links was created by the introduction of a concept called WikiNames: the spelling decided whether something was a keyword and linked the keyword to its definition. As with the generic link, a hyperlink was automatically created every time a concept was mentioned in the text. Furthermore, a definition node was created and later filled with information for each WikiName, which encouraged the creation of an "organically grown" hypertext [16].
Together, these two simple, yet powerful concepts allow
even authors who are completely unaware of the notion of hypertext to create
a complex network of ideas, definitions and discussions.
3.3 Collage: Spatial Hypertexts
Forcing users to create a formal representation along with their texts can prevent the use of hypertext systems (Shipman and Marshall 1999), or lead to "misuse" of pre-defined type systems (Nanard and Nanard 1993). The class of spatial hypertext systems was developed as a consequence. In spatial hypertext, e.g. VKB ( Shipman et al. 2001) and Tinderbox (Bernstein 2003), everything is implicit. A spatial hypertext is shaped by creating small notes and positioning these in a two-dimensional space. Often, attributes can be assigned to these notes, and notes can be put inside others. There is no defined meaning for relative spatial positions, color or even hierarchical inclusion.
Nonetheless, these systems are useful for authors who can
develop their own system of codes and relations while writing (or leave it
to intuition and later define the level of formalization). It is also
possible to extract very simple structures automatically from spatial
hypertext and publish the resulting hypertext on the Web ( Shipman et al. 1995). Although the preservation
of context can be realized with hyperlinks or even transpointing windows (
Nelson
1995), it is also possible to transfer some spatial characteristics into
the result rendered on the Web. Simple co-representation on the same Web
page can also create hypertexts (Bernstein 2003)
as the spatial and graphical layout conveys structure in a strong sense (Tufte 1990).
3.4 Composition, Transclusion and Annotation
Hypertext can be created by manipulation of text. When this happens in a process where information is both consumed (read) and produced, the newly produced information is still closely related to the information sources used. Preserving these relations in semantic types will produce a hypertext where the history of ideas and arguments is explicitly represented ( Uren et al. 2003). The source texts are part of the composition of the resulting hypertext.
Closely related are approaches where only the links to
the original documents are preserved during the collection of material.
(Nelson 1987) coined the term "transclusion" for this technique. Hunter
Gatherer ( schraefel and Zhu 2001) is an
example of a simple tool implementing the transclusion approach: by
collecting "components" from the Web, a "collection" can be created and
later edited.
Annotations are another approach employed working with
material available on the Web. Most people naturally create annotations as
they read. There have been successful efforts to support paper-based reading
techniques on digital devices ( Schilit et al. 1998), but
still there is no annotation tool in use in the context of the Web.
Annotations differ from composition and transclusion in their private nature
( Marshall and
Brush 2002). If personal annotations are to be incorporated into a
form of digital library (
Marshall 1997), it is important to put equal value supporting a
natural reading process and on the extraction of useful information (
Obendorf 2003). Directly connected to both user tasks and resources
in the Web, annotations could form the basis for authoring hypertexts.
3.5 An indirect authoring paradigm
Although the approaches mentioned in the preceding paragraphs differ in many respects, they have a common characteristic: authoring takes place in the context of a concrete task. While the text is directly authored, the hypertext, consisting of (typed) links or enriching the text with other forms of attributes, is created indirectly. Depending on the type of interaction of the author with the system, hypertextual functionality is created for a prospective reader. Indirect authoring can take forms ranging from those that need no special action of the author (generic links) to those that require nonformal representations (spatial hypertext) or trivial conventions (WikiWikiWebs) to process models describing and supporting a specific authoring process (incremental formalization of personal annotations or spatial hypertexts).
Indirect authoring can be taken as a starting point for
thinking about new hypertext applications. Users will seldom create
hypertexts for the sake of creating hypertexts but they might want to use
hypertext created for their specific needs. It is thus necessary to enable
users to create hypertexts without interrupting their normal work. A system
that provides only minimal features but offers these without high cost will
have a better chance of being used than a sophisticated system that needs
the user to devote much time and effort to be able to use it.
4 Some Consequences for Future Hypertext Research
[17] Web Services allow access to a distributed service provided via standard Web protocols (using HTTP and XML) for arbitrary clients. They can therefore be understood as another take against the unifying Web browser and towards special-use applications. In consequence, the semantic Web might open opportunities for custom tools to access data published on the Web. These tools might augment Web browsers and, for special applications (e.g. reading newsfeed), even replace them. On a smaller scale, this is already reality, as Web Services offer the same services normally encapsulated in Web interfaces via a defined interface that can be used by custom Web clients (e.g. Web search via Google).
This article has argued that while Web clients and link interfaces have changed only marginally since the "birth" of the Web, applications available on the server side are changing rapidly, as is the content accessible through the Web. New tasks served through the Web demonstrate the need for custom features. Due to the huge success of the Web and its dependence on a unified interface (cf. section 1.3), no single system can hope to replace the Web in the near future.
Even the semantic Web, which is often seen as the Next
Big Thing, is subject to the competition of today's Web. This is visible in
the efforts undertaken to create portals powered by semantic Web techniques,
but usable with a standard Web browser (e.g. Quan
et al. 2003), and the recent emphasis on Web Services by
the World Wide Web Consortium [17] as the first step
towards a Semantic Web (e.g. Berners-Lee 2003). This denys even the
existence of a new killer application as initially suggested (Berners-Lee et al. 2001) and instead
advocates the integration of existing applications via the semantic
Web.
Thus, it seems that in the immediate future, the Web
browser will be the window through which hypertext on the Web is accessible,
thus limiting the user's options for interactivity. Still, only a fragment
of the interactivity possible in Web browsers is put into use by today's Web
pages, which cannot generally be considered hypertextual in nature (cf. section 1.5).
4.1 Developing a Diversity of Specialized Authoring Tools
Even today, a number of specialized tools for writing hypertext employing different modes of editing are in use. Consequently, there will be no single killer application for everyone, but rather evolving generations of tools for content creation. New forms of hypertext on the Web call for new tools, and for each task the optimal mode of support for indirect authoring will differ. Systems that are successfully finding their niche in the ecosystem of the Web, such as the aforementioned Weblogs and WikiWikiWebs, demonstrate that limiting the cognitive overhead for authors to a minimum enables users to create individually useful hypertexts in the Web.
A promising approach to developing applications for
authoring hypertext is by prototyping specialized applications for
well-defined user groups. Developing for well-defined stakeholders and
generalizing the results for a larger audience has long been a successful
practice in application development (Cooper 1999, Mayhew 1999). There is a growing number of systems
that follow this approach: (Miles-Board, Carr et al. 2003) developed a tool for managers
writing reports; as they review and revise reports, their activity is
recorded and used to create descriptive metadata. Miles-Board, Deveril et al. (2003)
created a hypertext tool supporting the writing of dance critique, paying
special attention to the linear nature of dance performances. Gronbaek et al. (2003) created a system that
allowed architects to associate data with real-world objects, such as
samples of building materials, and
Uren et al. (2003) try to support the arguments employed in
scholarly critique. In many of these projects indirect authoring and a
variant of the linking-by-interacting approach has been implemented
successfully.
Indirect authoring can also be applied to other existing
hypertext paradigms. For example, the work of Shipman et al. (1995) and Shipman and McCall
(1999) on spatial hypertext and incremental formalization shows how
spatial organization that is intuitively created by an author can be
transferred into more formal representations, thereby creating a hypertext.
Publication tools could transform informal representations into hypertexts
suitable for publication. They could utilize the user's experience with
copy-and-paste or drag-and-drop techniques, introducing direct manipulation
techniques to the authoring of hypertext (compare with section 2.1).
4.2 Some Open Questions for Indirect Authoring Tools
[18] There has been some work on cognitive models underlying the writing process (Smith et al. 1987) and the role of external representation (Neuwirth and Kaufer 1989). The former defines a model where a text is produced from a network of thoughts via a hierarchy (as a table of contents) to the linear narrative, which is then consumed by the reader to again build a hierarchy and a network of thoughts. The latter focuses on the production of summaries and states that hypertext systems should be more flexible and support more than one hierarchy. There is, however, no empirical evidence for these models, and in some respect they contradict findings from cognitive psychology ( Charney 1994).
Careful investigations of authoring processes are needed [18] to determine which hypertext features will be useful in certain situations and how the creation of structure can be supported by custom tools. Open issues include questions such as:
-
Which level of support can you provide with your system?
-
How flexible must it be?
It is also still unresolved why certain hypertext features are immediately incorporated into commercial software while others only surface repeatedly in research prototypes.
An important characteristic of many hypertext systems is
a higher degree of interactivity than possible with standard Web pages; the
user cannot only passively consume information but actively contribute to it
(even if only by personal annotation). Under the label of the "Writable Web"
this has been a focal point of hypertext research. But while the Web as a
whole still lacks the means to store user-customized information, as was
included in the original Web proposal ( Berners-Lee 1989), many server-side applications make
good use of user's contributions entered not through a custom interface but
through the Web browser.
A general shortcoming of existing authoring tools,
however, is the closed world in which hyperlinks are authored (much like in
historic standalone hypertext systems). Citing or linking across boundaries
is not well supported, even though the universal hyperlink has been the main
reason for the success of the Web. The boundaries are not only between
hypertext systems and static Web pages; pointing to external targets often
breaks the internal linking mechanism. For example, the implicit linking
mechanism of WikiWikiWebs breaks when external links are included in a Wiki
page, even if they point to other WikiWikiWebs. Problems like this should
not burden the author. It is necessary to develop standards that allow
access to metadata of systems with similar functionality.
5 Conclusion
[19] It is still an open question how to generalize the existing solutions to a larger scope: the linking problem surfaces again when the boundary to the real world is considered. Connecting virtual objects to real meaning has been one of the initial objectives of the semantic Web. Ossenbruggen et al. (2001) show that hypertext research is an important source of experiences and can also inform the development of the semantic Web. A potential semantic Web will face similar problems to those discussed here: "How can authors make use of ontologies?" or "How can they understand and control metadata on their own work?" If authoring hypertexts is still to be possible for users in a metadata-rich environment like the semantic Web, the need for indirect authoring tools is acute.
The authoring of hypertexts is a difficult task for users. This paper shows
that current authoring tools for the Web are not very supportive of the
construction of hypertexts. The cognitive overhead that is introduced when
the user needs to input a formal representation that collides with his
immediate task-dependent needs has to be minimized. Indirect authoring
describes a shared characteristic of very different approaches experimenting
with authoring modes that allow the user to concentrate on meaningful units
of information for his immediate task. While the user interacts directly
with the text (and possibly a graphical representation), the structure of
the hypertext follows from an interplay of user actions and pre-defined
conventions.
The notion of indirect authoring can be used as a perspective to identify systems, both in the Web and from hypertext research, that are of immediate use to specific user groups. Varying, extending and combining these systems might produce interesting results. Together, a diversity of task-specific authoring tools might provide a larger number of users with the means to produce hypertexts in the Web.
The difficulty encountered by authors today when they
create metadata for hypertexts points to the risk that adaptive hypermedia
and the semantic Web will be initiatives that fit only certain, well-defined
communities because of the skills involved [19].
Extending these communities through authoring support tools that
minimize the user's overhead will be vital to the success of these
technologies.
Acknowledgements
As this article started as a result of informal discussions at Hypertext 2003, thanks go to all who took part in them. The author also thanks the anonymous reviewers for their help in focusing this article. Matthias Finck, Horst Oberquelle and Harald Weinreich provided criticism and helpful comments on an earlier draft. Finally, Bettina von Stockfleth helped to significantly improve the readability of this text. I would also like to thank you, the reader, and invite you to share your thoughts.