An object can not be understood if it is only looked at from one angle. When we seek to understand an object we pick it up, move it around, take measurements, and apply scientific tests. We talk about "manipulating" symbols, because like objects, we prefer to understand any collection of symbols from a number of different analytic views. A pilot does not rely only on visual perception but also uses a variety of instruments to provide different windows on the information describing the position of the plane (radar, altimeter, controller's radio, etc.). A hypertext reader should have a similar variety of instruments at her disposal.
The greatest strength of a computer based hyper-textual reading machine is not simply the possibility of hyper-linking texts to other texts (traditional texts use footnotes to achieve the same thing); it is the possibility to provide simultaneous automatic analytical views of a hyper-linked document. A powerful reading machine should provide a collection of modular integrated tools that enable the user to view not only the presented information, but also the context of that information, and meta-information about the containing documents. These tools should aid navigation, memory, and understanding of the text.
Different readers and different texts require different reading styles. A flexible reading machine should allow the user to pick and choose from a variety of modular tools, the tools should function somewhat autonomously (the tools are integrated but also function independently or in any combination). The user should have full control over the layout of these modules' interfaces and should be free to add and remove new modules without arbitrary constraints.
The ability to look at several pieces of information is key to enabling greater understanding of a work. In order to compare information in many traditional electronic text systems, the user has to follow one line of information and commit an impression of it to memory in order to compare it to another line of information. With simultaneous display, two lines of information can be evaluated in the context of each other without burdening the reader's memory; this frees the reader to concentrate on the interpretation of the text more than on the mechanics of reading.
Although not directly part of the semiotic experience of reading, the environment of the reading machine is of some importance to enhancing that experience. The machine should be ergonomic in its hardware and software functionality. The viewing area should be large, the input devices natural to use, the seating and lighting comfortable and conducive to study. Although Bolter suggests that scholars will go through great pains (such as standing at lecterns to read large volumes) to get to their information, I believe that eliminating undue difficulties from the reading environment allows the brain to be less distracted from its fundamental semiotic task.
In the following sections I seek to outline several example components that could be found in a reading machine of the type described above. In conclusion I will attempt to describe a path from current hypertext reading technology to a really useful scholarly reading machine.
In a hypertext system the user must have a way to return to a document once she has departed.
Most systems use a "stack" to implement the "go back" mechanism. This mechanism tracks a single line of followed links by pushing each followed link onto a stack, and popping it back off once the user backtracks. Under this system information about a string of followed links is lost when the user backs up and follows another path. This is a useful system for keeping on-track, it focuses the user, and prevents straying too far from the original document.
The stack system limits the semiotic experience because it forces a "tree" shaped reading of an electronic text. A directed graph (early versions of this idea can be found in the IRIS software) provides a much more detailed view of navigational history. Directed Graph based history shows every link that has been followed, where it linked from, and where it linked to. The result is a diagram with nodes representing pages and connecting arrows representing links. Viewing this graph can itself provide some insight into the organization and context of the text.
A directed graph module could be configured to store various levels of complex histories. After following a large number of links in several documents the graph may become unmanageably complex. The user would be enabled to specify pruning criteria to remove older pages from the scope of the history. Pruning criteria could include or exclude histories from previous sessions, they could reset the graph each time the user leaves a particular collection of documents (by entering a document location directly rather than following a hyper-link), or it simply only keep the last n documents in scope. The module could store all historical out of scope information so that the user could view a history graph of any past usage.
A difficulty in reading a hyper-textual document is that the reader does not know how fruitful it will be to follow a particular link. Without some exploration each individual page of a document exists out of context. This could be remedied by providing a view of the link tree that is rooted at the current page. There are two interesting views that could be used to represent this information.
A flat view of the link tree could be presented like a recursive collapsible table of contents (much like directory listings on some GUI operation systems) or as a directed graph (similar to the above mentioned History views). Use of this view would be analogous to using the table of contents of an actual book.
A "spatial" view of the link tree would represent the information as a three dimension space with icons representing pages at various depths in the link tree as pages further back in the space. This would allow the user to "fly-through" the link tree, changing view point quickly by veering towards one subtree or another. Use of this view would be analogous to flipping through the pages of a book to either get an overall impression of the text, or to find a particular passage. A fly through view is different from "surfing" because it presents each page in spatial relation to other pages, allowing a user to compare various paths through the link tree simultaneously.
Hypertext documents necessarily contains links specified by the author. These links are of primary importance because they are an integral part of the text, without them the text is incomplete. But there is another kind of link, links that put the document into a larger context by providing references.
A Link Set module would allow the user to specify (or download from a third party) reference link sets. These would map words or phrases to hyper-textual references (such as encyclopedia entries, indeed an encyclopedia maker would be a good third party to market such link sets). These link sets would be applied to all documents, providing a readily available context that perhaps the author was not even aware of.
Dynamic networked hypertext systems have the problem that they constantly change. Looking back at a bookmarked document a user may find that the information that was once there has changed or disappeared, indeed the page itself may no longer exist.
At a cost of disk space a History module would allow the user to keep copies of all revisions of a given document that have ever been visited. The module could provide "difference" tools that show the changes between any two revisions of a document.
The document revisions could be stored on the document server, which means every revision of the document that ever existed would be accessible; it could be stored by the client which means only the versions that the user has seen actually exist. Advantages to the server side implementation mean that authors could use it to their advantage by including the time of linkage in their links to other documents, ensuring that links never "break".
There are a number of other features which would also be interesting to explore, many of these are taken from existing hypertext systems:
A relevance spider that keeps track of the kinds of documents you read and attempts to find similar documents.
A document statistics generator that provides a brief overview of the current documents size, number of images, last modification time, time of last visit, and links to previous versions of the document (see Document History above).
A detailed bookmark system that allows the storage of bookmarks with word granularity. Temporary book marks (like fingers holding place in an open book).
An annotation system that allows a user to "mark-up" documents.
Although systems like IRIS are in some ways more advanced than the HTML based WWW, they have not been as widely distributed. The WWW owes much of its success to its simple "stateless" nature. Documents on the WWW are no contained in databases with revision history, targets of links do not have knowledge of the documents that link to them, the document base is distributed and dynamic.
I believe that the modular reading machine described in this paper can bring many of the features of systems like IRIS to the WWW without compromising the simplicity of that medium. Most of the features described are client side features which can be accomplished with "intelligent agents" (personal web crawlers that gather linkage information in the background while the user browses). Client side functionality could be enhanced by special servers which do much of the web walking from a centralized location and make databases available to the client. In any case these specialized clients and servers could derive all of the necessary information from existing HTML documents without modification to the servers that serve those documents.
With the addition of ergonomic user interface hardware, extra disk space, and a fast network connection, this reading machine would make the WWW a medium properly suited to academic reading and understanding. It would finally make the hypertext experience more powerful, more useful, and more efficient than the book experience.