The memex is a theoretical system. The memex model has no provision for respecting copy restrictions during reproduction. It assumes that the owner of the memex is the owner of all the data stored within it. There is no mechanism for enforcing the rights of originating hypertext composers and distributors. Memex was a model based on a unified, freely reproducible global data repository.
The model proposed in this paper captures the essence of user interaction in memex with three user classes: author, publisher and reader. The APR model is an attempt to formally define the interface between the readers and the authors. APR is based on this author's experience working within a traditional print publishing company. The mechanisms which support the model have been the subject of research conducted in both academic circles and in a commercial online information setting.
The strategic application of publisher resources should help eliminate the need for an author to have a wide hypertext experience or advanced computer knowledge. The traditional role of a publisher is to provide the author with a range of skills that the author, himself, may not possess.
Authors should write words. Visual artists produce images. The details of grammar, spelling, operating systems, communication links, servers, and network topologies can detract from the basic task at hand: composing art and imparting knowledge.
The only things that an author should have to know are how to submit information for review and where to pick up the check.
If a reader is expected to pay for the data he is consuming, the quality of the data should be fairly high. That is to say, the data must be consistent, timely, comprehensive and relevant to the reader's desires.
Before a reader can purchase data, he must be able to locate and retrieve it. The reader should know (roughly) what he is getting, where he can get it, how he can use it and how much it will cost him.
In the APR model, the publisher is the label applied to every intervening layer between the authors and the readers. The authors and readers interact with different portions of the publishing house. Authors and readers rarely interact directly. Authors interact with editorial staff while readers interact with a publisher's distribution network.
The publisher is responsible for these activities which are common to both print and electronic distribution media:
Publishers may also undertake the creation of search mechanisms, indices and thematic groupings of related material. In the context of electronic publishing via the world wide web, they may also choose to maintain onsite Internet connections and servers.
A finer grained breakdown of the interface between author and reader yields several sub sets within the publishing network including: Internet service providers; bookstores; libraries; online services; ordering departments; warehouse and dispatch staff; editorial staff; proofreaders; layout; trailblazers; non-linear integration specialists; advertising and promotions departments; legal advisors; management and accounts.
The typical delineation in hypertext is to break the data component into two types: consumable content and content relating associations.
For the purposes of this discussion, the basic unit of content for reader consumption is called an electronic page, or epage. An epage is, for an instant in time, a frozen atomic unit supplied for display to the consuming reader. It may be constructed from mixed media and from several independent sub-components prior to delivery, but it must arrive as an integrated unit for display by the reader's information browser.
An epage is analogous to an HTML file after it has been pasted together for display by a browser. An epage has had many labels: an atomic base component [2]; an item [3]; a chunk [4]; a node [5]; and a frame [6].
The content relating association are called links by most hypertext researchers. Links were considered to be the essential feature of the memex [7]. In the world wide web, the link is subordinate to the HTML file. The w3 browsers and the HTTP are only able to render, request and deliver epages.
An ebook can be an interactive, networked resource such as an HTML data set installed under a server. It could also be a standalone bundle for offline browsing like a CD. It may be a mixture of the two.
The epages themselves may contain embedded links in the spirit of HTML. Alternately, links may be applied to an epage from an external organizational layer.
Ebooks are one type of organizational layer into a set of epages. The organizational layer brings the ebook context to the epage. One ebook may closely associate two epages as while another may not connect them at all.
This ebook context has ramifications when an epage is retrieved by a reader. The epage must be constructed to fit the context of whatever ebook the reader is currently consuming.
That is not to imply, however, that every epage must be composed with all the contextual information built in. The context may be applied by the retrieving site and may contain components from different serving sites.
An epage, then, may be served differently on subsequent visits by the same reader during the same session. Network caches can affect the integrity of this type of transaction under the current URL resource identification system.
As it stands, the web has no inherent ability to provide this level of flexibility, though servers can be forced to simulate this behaviour [8]. A scalable solution involving proxies and browsers must be sought.
Another factor which affects the appearance of an epage is what resources the reader is authorized or licensed to access. If an epage contains components drawn from 3 different licensed resources, a reader with access to only one will not be able to access the other two (though he may be enticed to negotiate a license to access them). The author of that epage would receive a view different from that supplied to a reader with access to all three resources.
In many respects a trail is analogous to an ebook; i.e., it is a set of epages with an organizational network. However, a trail may extend into many ebooks. These ebooks may not all be under the jurisdiction of the same publisher. In those circumstances, it is reasonable to consider the delivery of trails without the complete set of the epages they are built around.
Shrewd publishers may supply a set of trails along with every ebook they license. Consumers can then be made aware of other products which are available for purchase. The advertisements are built conceptually into the product at a point where they have to most relevance. A trail is advertised when a reader is on one of its thematically related epages.
The trail also provides an alternative mechanism for navigating within the ebook already purchased. It also provides a way for two complementary ebooks from the same publisher to have a utility greater than the sum of their parts. When the two books are co-located they contain the function of each independent ebook. They also have an additional organizational structure which capitalizes on their common themes.
Trails perform another useful purpose for publishers. By using a trail, a publisher can identify which epages from a data pool are to be distributed as a particular ebook. Thus when the publisher `sets his reproducer in action' on the trail [10], the result is an ebook ready for market.
Another advantage of trails is that they help achieve the memex goal of `anticipating the selective needs to be encountered later' [11]. The publisher anticipates a requirement for information relating to the theme of the trail.
The use of a trail is an example of where the epage must be served in context.
In [12], the authors conclude that `tours or path mechanisms... are hard to author and maintain''. The link structures for trails suggested in this paper attempt to alleviate that problem. The trails are not embedded within the epages, rather they are applied as overlays on top of them. They can be manipulated using a special set of tools. This approach allows the epages to evolve separately from the trail.
It is possible to warn a user when he is following a link which will digress from the theme of the trail. In fact, it is possible to constrain the reader so that he cannot digress from the trail at all. This would be an attempt at overcoming the disorientation experienced by many hypertext users [13].
There is a more flexible, and possibly more useful mechanism for constraining the reader. Using the research into distance and metrics from [14] and the notion of depth from [15], we can set a numerical limit on the distance that a reader can digress called the trail width.
The width may be set by the composer of the trail (Bush called this person the trailblazer[16]), the trail publisher or the trail reader. The publisher may set a maximum width for the trail. With sufficient software support, the reader may be able to reduce the trail width even further.
The reader would first set the trail width to the minimum possible. The software would offer the set of epages (still possibly non-linear) which were considered most fundamental by the trail's composer.
The reader would then increase the trail width and review the epages in order to increase their understanding of the information being presented. Eventually, the reader will operate with the maximum possible trail width, wandering freely within the non-linear data structure that is closely associated with the theme of the trail.
This helps the reader build a mental map of the contents of the hypertext by modifying the view and providing a reduced frame of reference. The reader will come to view the links which connect the fundamental relationships differently from the ones which are tangential.
The reader may also be able to regain a sense of completeness which is otherwise difficult to achieve. If a student is to use a hypertext as a replacement for a printed text book, the student must have some way to gauge their progress through the course material.
There are two reasons why this may be necessary: theme scope and commerce. The trail may overlap very superficially with one ebook while it may address the same core theme of another ebook. The trail publisher may have licensed two ebooks from different publishers; one publisher may permit liberal digression while the other may be more restrictive.
Control over maximum trail digression is in the hands of two publishers: the trail publisher and the ebook publisher.
There must be a mechanism to allow an ebook publisher to specify digression limits outside the trail which references it. A trail may be blazed by an uncooperative third party publisher, and the ebook publisher may have no access to the distributed version of the trail.
Trail digression constraints become another impetus for consumers to license services from the publishers. Presumably, the trail forms an advertisement for the portions of the ebook which are not accessible by default. It becomes a subject of further negotiation between the consumer and the ebook publisher.
These types of restrictions are necessary to provide a scalable hypertext creation model. If the author is writing a single ebook and is acting as her own publisher, then an all-or-nothing publication model may be feasible. If, however, the composition team is comprised of scores of authors working on dozens of conceptually distinct products, there must be some mechanism for controlling the reproduction of works for distribution via public access mechanisms.
In order to completely separate an ebook from a set of epages, it is useful to consider the trail boundary. The boundary of a trail of width w is the set of links which, if traversed, would retrieve epages which are farther than w units away from the trail. The set of boundary links require attention by the publisher prior to publishing the ebook.
The publisher has several alternatives for dealing with boundary links: replace them with vanilla text; replace the destination set with a class of link which supports followup contact; or let them dangle and hope for the best.
Time plays an important part in some information resources while it plays less of a role in others. Complex temporal information sets stagnate if they are allowed to remain frozen for too long; they go stale.
When the author is the publisher, the fluidity of an epage is at the whim of a single individual. An advantage of the APR model in a collaborative working environment is that absent workers can be back-filled by other staff. This helps ensure that highly fluid data retains its relevance.
This restriction can be alleviated with ebooks. The reader can pay for just that portion of the ebook which has changed since the last edition he purchased. The publisher is able to supply the appropriate `patches'.
The cost of manufacturing standalone errata is relatively low compared to an identical print volume. There is minimal extra cost associated with releasing both a new edition of an ebook and the self-contained errata in parallel. They are both products of the same production process.
This is not without its complications. Fluidity causes problems.
Suppose a reader has made significant extensions or, worse, modification to the original edition supplied by the publisher. If the errata is confined only to individual epages, update may be relatively straightforward, if interactive. If, however, the new edition involves restructuring the organizational layer, the update process becomes much more complex. This author can propose no solution that does not require significant human intervention.
Another complication with multiple editions relates to trails. Recall that trails could be combined with ebooks when they were bundled for market distribution. Eventually, the trail may come into contact with a later edition of an ebook known to contain relevant passages.
A solution for this would be for an ebook to carry with it a newer copy of the trail which could accommodate the older editions of known ebooks as well as any customizations applied by the reader after delivery of the original. This could easily cause an exponential growth in the amount of secondary information which must be bundled with each subsequent product.
Another approach would be for the trail to be extended into an arbitrary ebook. One mechanism for natural language processing which could be employed for this task was presented as a mechanism for automatically suggesting links in [17]. Another alternative is a concept-based retrieval system [18].
The LHT source files are able to express entire ebooks, epages, links and trails.
If an ebook is encoded along with all its epages in a single file, the linear ordering between adjacent epages is reflected in the resulting, compiled ebook. LHT shatters the mapping of an epage to a single file. An epage may be authored within several different source files.
An epage can identify not only the links which emanated from within its content, but it can also suggest epages and general themes (ebooks, trails) which might link to it. A link can be created which has a visible component on an epage other than the one within which it was composed. It may be composed within a source epage, within a destination epage, or outside both sets.
The language was implemented along with a set of translators which `compiled' the LHT source into a data structure for replay by a browser of the author's devising [20]. The prototype translator has been modified to produce HTML encoded ebooks. The compiler analogy equates HTML to machine code.
One of the abilities that LHT possesses that HTML lacks is the ability to embed codes in the source which, once compiled, produces epages which are served differently in different contexts. The composer of the LHT epage can specify sub-atomic components which are visible in one context and invisible in others.
A concept like this one was integrated with the Hypertext Abstract Machine (HAM) [21].
LHT can also express a link which should be dropped in (parachuted) into another epage. These links may be created when an epage is retrieved, or at the point when the epage is compiled into the ebook. Parachute links are used to implement the organizational component of a trail. They are applied only when relevant to the context.
The Dexter model [22] formalizes the idea that a link is not subordinate to an epage; rather, the two are both specific cases of a component. The Dexter model can accommodate parachute links in the spirit of LHT. A Dexter link can have more than two end points [23]. It can also support a more flexible variety of directions [24]. In contrast, a HTML link is a one-way link routed within the source page.
LHT and the associated software can be used to provide a two tiered approach to information delivery. The authors and editors control the essential content of the epages but the publisher can apply a standard stylistic look-and-feel to every epage that is produced. The publisher can also apply a standard organizational layer to every ebook produced.
Using LHT, it is possible for a publisher to develop the organizational layer within an ebook such as an encyclopedia in advance of receiving many of the authors contributions. While the structural coordinator of such a project cannot anticipate every relationship that the researchers will uncover, high level development can occur asynchronously with the supply of the contributions. Epages can be linked before they are composed.
The LHT server also provides a mechanism to enforce commerce. The first time a new reader retrieves any epage from an ebook he is funneled through a toll booth. Once license has been granted, the requested epage is supplied with prominent links to relevant background information about the ebook(s) and trail(s) which contain the epage.
The task of maintaining context is grafted as an extension of the server. The context could be easier to maintain as a shared record between the server, the browser software and any intervening proxy/caches.
Hypertext is traditionally based on author-reader interaction which doesn't scale to support commercial collaborative hypertext initiatives.
HTML is not a natural language within which to compose. The HTML pages are well developed while the linkages are weak. Links must be accommodated as first class objects.
Trails with constrained digression can alleviate a significant disorientation problem that readers often experience. They can also help publishers prepare ebooks for distribution via a variety of protocols. Ebook editions complicate the matter of trails.
Trail following and trail blazing [26] both require formal attention by the w3o when it reviews the capabilities of the next generation of browsers. Digression and trail width must be supported by the browser or a proxy if a trail is to include epages from a mixed set of ebooks.
Research into a trail data construct which respects varying local widths is necessary .
LHT shatters the tight coupling of the organizational layer with the epage. This coupling pervades HTML. The separation of epages from the links permits an easy mechanism for building and maintaining trails. The tight coupling between the source epage and the one-way link will limit the ability to introduce trails to HTML.
Another significant enhancement to linking is its type. A HTML link can glean some type identifiers from the protocol for data retrieval and the mechanism for interpreting the returned epage. These type apply more to the epage being retrieved than to the link itself. A link has type of its own.
LHT supports link traversability by user class as an explicit link type. The work of Trigg [27] and other researchers (eg. [28], [29] and [30]) have extended the concept of links through types. There is a potential benefit afforded by link type which requires further investigation.
Epages must be displayed by the browser in context. A mechanism to serve the same epage differently depending on the context of the links is necessary. The context is built from variables at the server and the browser; it is a state shared between them. Contexts must be composed before they can be served.
The current implementation of the caching proxy is a hurdle to both commerce and context in the world wide web. Without context elements embedded in the URL, two users served from the same cache may experience epage crosstalk. If two users request the same licensed epage from the same caching proxy, the originating server may not license each user individually.
Trail width and constrained digression extends beyond the epages which form the backbone of the trail. The appearance of the epages served during trail following is a function of trail, ebook, server and browser side context. Without context, these types of trails cannot be accommodated using HTML.
The trail boundary may help control ebook production and selective replication. The boundary which consists of a set of links requires more study. Trails and their boundaries might be used to define open and closed sets within the hypertext which, in turn, may lead to a meaningful definition of the topology of a hypertext.
Fluctuation and fluidity in hypertext is a serious problem. Delivery of editions of ebooks and trails illustrates some of the difficulties which can arise when data goes out of sync. More research into data fluidity and decay is warranted.
2. Halasz, F., and Schwartz, M., 1994, ``The Dexter Hypertext Reference Model'', Communications of the ACM, Volume 37(2), pp 30,39. February 1994.
3. Bush, V., ``Memex Revisited'', Science is Not Enough, William Morrow & Company. New York, U.S.A., 1967.
4. Trigg, R., and Weiser, M., 1986, ``TEXTNET: A Network-Based Approach to Text Handling'', ACM Transactions on Office Information Systems, Vol 4(1), pp 1,23. January 1986.
5. Campbell, B. and Goodman, J., 1988, ``HAM: A General Purpose Hypertext Abstract Machine'', Communications of the ACM, Volume 31(7), pp 856,861. July 1988.
6. Akscyn, et. al., 1988, ``KMS: A Distributed Hypermedia System for Managing Knowledge in Organizations'', Communications of the ACM, Volume 31(7), pp 820,835. July 1988.
7. Bush, V., ``As We May Think'', Atlantic, Volume 176, pp 101, 108. July 1945.
8. Cuff, L., 1995, ``Publishing Compiled Hypertext with LHT and the World-wide Web''. Masters Thesis (in prep. -- title in flux.) Memorial University of Newfoundland, St. John's, Newfoundland, Canada.
9. Bush, V., ``As We May Think'', Atlantic, Volume 176, pp 101, 108. July 1945.
10. Bush, V., ``As We May Think'', Atlantic, Volume 176, pp 101, 108. July 1945.
11. Bush, V., ``Memex Revisited'', Science is Not Enough, William Morrow & Company. New York, U.S.A., 1967.
12. Botafogo, R., et. al., 1992, ``Structural Analysis of Hypertexts: Identifying Hierarchies and Useful Metrics'', ACM Transactions Information Systems, Vol. 10(2), pp. 142-180. April 1992.
13. Trigg, R., and Weiser, M., 1986, ``TEXTNET: A Network-Based Approach to Text Handling'', ACM Transactions on Office Information Systems, Vol 4(1), pp 1,23. January 1986.
14. Rivlin, E., Botafogo, R., Schneiderman, B., 1994, ``Navigating in Hyperspace: Designing a Structure-Based Toolbox'', Communications of the ACM, Volume 37(2), pp 87,96. February 1994.
15. Botafogo, R., et. al., 1992, ``Structural Analysis of Hypertexts: Identifying Hierarchies and Useful Metrics'', ACM Transactions on Information Systems, Vol. 10(2), pp. 142-180. April 1992.
16. Bush, V., ``As We May Think'', Atlantic, Volume 176, pp 101, 108. July 1945.
17. Bernstein, Mark, 1989, ``An Apprentice That Discovers Hypertext Links'' reference lost for the moment. Very Likely from EP90, Cambridge University Press. pp 212,223.
18. Arents, H., and Bogaerts, W., 1993, ``Concept-Based Retrieval of Hypermedia Information: From Term Indexing to S Hyperindexing'', Information Processing and Management, Vol. 29(3), pp 373,386.
19. Cuff, L., 1995, ``Publishing Compiled Hypertext with LHT and the World-wide Web''. Masters Thesis (in prep. -- title in flux.) Memorial University of Newfoundland, St. John's, Newfoundland, Canada.
20. Cuff, L., 1993, ``Trailblazer: a DOS browser for hypertext''. Report for Software Engineering masters course, Memorial University of Newfoundland, St. John's, Newfoundland, Canada.
21. Campbell, B. and Goodman, J., 1988, ``HAM: A General Purpose Hypertext Abstract Machine'', Communications of the ACM, Volume 31(7), pp 856,861. July 1988.
22. Halasz, F., and Schwartz, M., 1994, ``The Dexter Hypertext Reference Model'', Communications of the ACM, Volume 37(2), pp 30,39. February 1994.
23. Leggett, J., and Schnase, J., 1994, ``View with Open Eyes'', Communications of the ACM, Volume 37(2), pp 76,86. February 1994.
24. Grønbaek, K., and Trigg, R., 1994, ``Design Issues for a Dexter-Based Hypermedia System'', Communications of the ACM, Volume 37(2), pp 40,49. February 1994.
25. Bush, V., ``As We May Think'', Atlantic, Volume 176, pp 101, 108. July 1945.
26. Bush, V., ``Memex Revisited'', Science is Not Enough, William Morrow & Company. New York, U.S.A., 1967.
27. Trigg, R., and Weiser, M., 1986, ``TEXTNET: A Network-Based Approach to Text Handling'', ACM Transactions on Office Information Systems, Vol 4(1), pp 1,23. January 1986.
28. Garzotto, et. al., 1993, ``HDM -- A Model-Based Approach to Hypertext Application Design'', ACM Transactions on Information Systems, Vol. 11(1), pp 1,26. January 1993.
29. Snaprud, M., Kaindl, H., 1994, ``Types and Inheritance in Hypertext'', International Journal of Human-Computer Studies., Vol. 41, pp 223,241.
30. Halasz, F., and Schwartz, M., 1994, ``The Dexter Hypertext Reference Model'', Communications of the ACM, Volume 37(2), pp 30,39. February 1994.
Leslie Cuff is enrolled in the Master of Science program in the Department of Computer Science at Memorial University of Newfoundland where he has developed the LHT language and translator. In addition to time spent writing scientific software and working with The Enterprise Network, he was an image technician and photographer for the Encyclopedia of Newfoundland and Labrador, Volumes 4 and 5.
He resides in St. John's, Newfoundland and Labrador, Canada (a mere 10 kilometres from Cape Spear, the most easterly point in North America) with his wife and son. He is a birdwatcher and an all season outdoors explorer. He attended www-2 at Chicago in October 1994. He participated in w3collab at Boston in September 1995. He is a member of the w3o working group for treating links as first class objects.
He is involved in a project to commericialize the LHT multimedia processing kit with a consortium of Newfoundland based media producers, information disseminators and telework agencies.
He can be contacted at lez@cs.mun.ca or les_cuff@porthole.entnet.nf.ca via email.