A Schema Based Approach to HTML Authoring

Marcus Kesseler
Abstract:
This paper presents a novel approach to high-productivity authoring of large, regularly-structured hypertexts. By explicitly representing the objects in the hypertext and the relationships between them in a schema, it is possible to create, manipulate and maintain large hyperdocuments with high efficiency. In the implementation, called the HSDL, all such schema operations are performed on a graphical user interface (GUI). Special attention has been given to the problem of schema evolution. In HSDL, the author can do non-trivial schema update operations even if classes have already been instantiated. The mapping from the schema to HTML, called compilation, is done by a series of programs in the programming language Scheme. These programs, called expanders, are integral part of every schema. Although the default set of expanders will already provide fairly sophisticated HTML layout, users may easily adapt them to suit their special needs. A built-in, schema-aware HTML editor allows the integration of links defined in the schema into the HTML contents of a node.
Keywords:
Authoring Environments, Information Representation and Modelling, Consistency, Integrity, Authoring-in-the-Large, Schema Evolution, Hypertext Linearization

Introduction

Like Garzotto et al. [7], we would like to see the task of producing large hypertexts split into two intertwined but nevertheless clearly distinguishable activities: First, those that deal with the hypertext on a per node basis, which is called Authoring-in-the-Small (AIS), and second, those that deal with structuring decisions on a larger scale, possibly encompassing hundreds of nodes, accordingly called Authoring-in-the-Large (AIL). AIS usually involves dealing with layout oriented problems, or, in HTML, with local structuring decisions. The general idea of schema-based AIL is best conveyed by a four step procedure:

  1. The author models the objects that exist in the target domain, ie, the domain the document is about, and the relationships that exist between these objects. This model is called the class schema.
  2. Using the classes in the class schema as templates, the author proceeds to create instances of the classes and instances of the relationships in the instance schema. These operations are fairly intuitive when performed visually on the graphical user interface of the implementation.
  3. The author then proceeds to fill the as yet empty instances with content. This is where AIS in the conventional sense is done.
  4. In a compilation step, the structural information in the schema is merged with the contents of the instances and gets translated into the target hypertext representation (HTML).

Of course, any approach that forces authors to proceed in such a strictly top-down manner is completely at odds with the way that authoring is actually done. Authoring, which is among the so called complex design activities [12], is an intrinsically opportunistic process. By this, we mean that authors seldom adhere to schematic procedures while creating large documents. Authors seem to randomly switch between contexts and levels (AIS, AIL).
To develop this approach, a great deal of effort has been invested to reconcile the inflexibility inherent in schema-based approaches with the chaotic ways human authors usually go after their business.

The implementation, called HSDL (Hypertext Structure Description Language), evolved from initial experiments with a programming language into a full-blown visual hypertext authoring environment. The Structure Description Language as such, although now invisible to the user, lives on in HSDL as the base for the persistent schema storage mechanism.

The rest of this paper is structured as follows: We give a more detailed overview of the schema and the problems of schema evolution. We then proceed to show how a schema is compiled into HTML. After discussing related work in this area, we finish with our conclusions and future research directions.

The Schema

The schema consists of an intentional description, the class schema, and an extentional description, the instance schema.

The Class Schema

The class schema consists of classes and link-classes between them. A class is a tree-structured composite object. The objects in a class tree are called components. The top component is called the root component. Consider the class Painter in Figure 1:


Figure 1: The painter class

Painter has a root component of the same name, which has the subcomponents Biography and Works. Biography has two further subcomponents, called Childhood and Adult Life respectively.
An arbitrary number of link classes may be defined between any two components or even from a component to itself (see below). Link classes are directed: We distinguish between the source component and the target component of a link class.

The Instance Schema

The instance schema consists of instances and link instances. All instances of a class are exactly alike, that is, classes are used as templates that are always instantiated as a whole. The nodes of an instance are called instance-nodes (or i-nodes for short).


Figure 2: The class and instance schemas representing a gallery

Figure 2 shows an HSDL screen dump which contains the class and instance schemas of a gallery. The gallery consists of three painters and eight of their paintings. The paintings are also classified into four thematic collections. Note that the instances Picasso and van Gogh are not fully visible: the double frame around their root i-nodes indicates that they are imploded.

Link Instances Induce Ordered Collections

Large hypertexts are often organized as hierarchies. It is common practice to make the members of a set of nodes linked under a same anchor node (e.g. the sections of chapter, the paintings of Picasso) traversable by previous/next links.
We refer to such a set of nodes connected by previous/next links as an ordered collection, or simply a collection. The anchor node of the hierarchy is called the owner of the collection. In HSDL we generalize the idea of an ordered collection, by specifying that the order within a collection is not a property of the i-nodes themselves, but rather of the link instances leading to them. A set of link instances of the same link class coming from a same source i-node is said to induce a collection of i-nodes. With this definition, it is possible to embed an i-node into an arbitrary number of collections. We move from hierarchies to true networks.
Furthermore, a link instance induces a collection in both directions. The i-nodes that are the targets of a set of links from a same source are said to belong to a target collection. Likewise, the i-nodes that are the sources of a set of links to a same target are said to belong to a source collection.


Figure 3:The Authors & Papers class and instance schemas

Consider the very simple class schema in Figure 3: A set of authors and papers written by them are to be presented as a hypertext. In the instance schema we have three authors (A, B & C) and three papers (1, 2 & 3).


Figure 4: A target and a source collection

The left screen dump of Figure 4 shows the target collection of Author A, i.e., A is author (or co-author) of papers 1, 2 & 3 (HSDL has a display mode in which only the links into or from a selected instance are visible). The right screen dump of Figure 4, shows the source collection of Paper 3, which was written by A & C. Therefore, the link instances of a single link class from Author to Paper naturally induce a collection of papers in the target direction ("Papers by this author") and a collection of authors in the source direction ("Authors of this paper").

Anchors from an owner to a collection usually form an index. By default, HSDL translates indices into the unordered list (<UL>) HTML environment. By manipulating the anchor expanders (see below) it is possible to create a wide variety of layouts for such indices.
Changes to the order of a collection (i.e. sorting) is done by drag-and-drop in the HSDL GUI. Since collections are defined by link instances, reordering always occurs relative to a selected link instance and therefore affects only one of the many collections the instance may be part of.

Since authors may not want to include every possible hierarchy as an explicit index into an HTML node, their generation can easily be toggled in a link-class-properties pop-up menu. The same applies to the generation of previous/next links.

The anchors that aid navigation within the members of a target collection are organized into previous/up/next-triplets (or previous/down/next-triplets for source collections). Consider Figure 5, which shows the compiled HTML result for the Dora Maar instance of Figure 2 in the NCSA Mosaic browser. Dora Maar is a member of three collections: It's a panting by Picasso and has been classified as a Portrait and a Cubistic painting. Such crossing collections are best visualized by color-coding the respective triplets according to their owners. In Figure 5 the up anchor of the first triplet, to Picasso, has been compiled into an up-arrow icon, while in the other two triplets, both to instances of Thematic Collection, have been compiled into the owner's name. For simple changes to triplets styles HSDL offers pop-up menus (e.g. Void, Arrow, Name, Name+Arrow, Color). For users wishing to go further, expander programming offers the necessary degrees of freedom.

It should be emphasized, that Figure 5 has no edited contents whatsoever. All visible elements are HSDL defaults. The "Painted by" label is entered by the user during link class definition and therefore produces this layout in all instances of Painting.


Figure 5: The Dora Maar instance in the NCSA Mosaic browser

Non-Hierarchical Links

Obviously, not all links in an hypertext will be hierarchical. One of the main characteristics of hypertext is the ability to explicitly represent associations between any two nodes in a network, completely disregarding any overall structure the nodes may be embedded into.
With the slight restriction that every link is an instance of some link class, it is possible to create such links without introducing new concepts into HSDL.
Suppose that Cezanne's biography mentions that he was influenced by van Gogh. Clearly, one would like to make the words "van Gogh" into a link to the node representing van Gogh. In Figure 2 the class Painter has a link class called Influenced by (link class names are not visible in the screen dump) from Biography to Painter. What at first looks like a puzzling link class from a class to itself, works exactly how one would expect on the instance level: As shown in the instance schema of Figure 2, it is possible to link Cezanne's biography to van Gogh.
HTML anchors representing link instances are placed at default positions before or after the HTML contents of an i-node, depending on the current set of expanders. The built-in HSDL HTML-Editor provides the means to override such default placements and insert anchors anywhere into the HTML contents of an i-node. Link anchors can be copied or grabbed: A copied anchor will still have a duplicate at the default position. A grabbed anchor will be removed from it. Whole indices can also be copied or grabbed.

Note that some authors (e.g. Conklin [5] and Garzotto et al. [7]) make a clear distinction between organizational links, which span hierarchies, and referential links, which allow random linking of nodes.
In HSDL this distinction becomes somewhat blurred. At first sight, there really seems to be a great difference between the links from a chapter to its sections and a link expressing that van Gogh influenced Cezanne. On closer inspection, we find that the Painted by links from a painter to his or her paintings, which though seeming to carry much more meaning than an index of sections in a chapter, span a completely analogous structure. Now, if our gallery had, say, six painters that were influenced by van Gogh, even the instances of this link class, which seemed to carry a lot of semantics, would start looking like a collection (and hence hierarchy) of painters. Therefore, as seen from the HSDL formalism, there is no difference between referential and organizational links. Their seemingly different semantic weight is in the eye of the beholder.

Referential Integrity and Fast Prototyping

The HSDL implementation enforces strict referential integrity on the class and on the instance schemas. Therefore, it is impossible to create a dangling link or to make parts of the schema inaccessible (no orphans). Referential integrity is enforced by following three simple rules:
  1. When a new class is created, a link class linking it into an existing class has to be created along with it.
  2. When deleting a class, also delete all link classes leading into or out of it.
  3. Class or link class deletions are not allowed if they make parts of the schema inaccessible.
The same rules apply to the instance schema. Deletions in the class schema are cascaded to the instance schema, that is, when a class schema object is deleted, all its instances are also deleted. If cascading a legal class schema deletion leads to an inconsistent instance schema, the class schema deletion is declared illegal and undone.

To efficiently enforce rules 2 and 3, the HSDL software architecture includes a Change Manager. Aided by an undo-stack, which records all schema changes that occur during an editing session, the Change Manager implements the transaction concept for schema update operations. Therefore, destructive operations that lead to a referentially inconsistent schema are simply rolled back.
The pop operation on the undo-stack is also available to the user, thereby providing full undo capabilities within an editing session.
Given that referential integrity is ensured, it is possible to compile the schema into a working hypertext at any stage during development. This has great advantages for fast prototyping, since an author can generate hypertext skeletons very fast. Such skeletons of HTML nodes, though devoid of edited content, contain default anchors generated by HSDL and are therefore fully navigatable. They are of great help to check if the design and structuring decisions represented in a schema really mirror the author's intention on the hypertext level. Experienced HSDL users can create fully functional HTML skeletons consisting of hundreds of nodes and links in a couple of minutes.

Heterogeneous Collections, Guided Tours and Multi-Link-Classes

As yet, there is one important feature missing in the expressiveness of HSDL: The ability to create heterogeneous collections, that is, collections composed of instances of different classes, which are needed in two main modelling situations:
Fine grained modelling:
Suppose we want to model the canonical "Article" hierarchy of sections and subsections. As such, it presents no problem. But suppose we want to go further and say that a subsection is composed of a random sequence of paragraphs, figures or tables. With the HSDL indices presented so far, it is not possible to insert a link to a figure between two paragraphs. We would have an index of all paragraphs, followed by an index of all figures and an index of all tables. Although we could explicitely copy or grab the respective link instances into the i-node contents, thereby "welding" them to fixed locations, this unfortunately renders the link instances inaccessible to the reordering operators provided by the HSDL GUI.
Guided tours:
Which are sequencial paths through random subsets of i-nodes of an instance schema.
To tackle this problem, we extend the definition of link classes as connecting a set of source classes to a set of target classes. A link class where the source or the target class set has a cardinality larger than one is called a multi-link-class. A multi-link-class defined from {A, B} to {C, D, E} allows for link instances from any instance of A or B to any instance of C, D or E.

Figure 6 shows the "Article" schema with a multi-link-class from {Subsection} to {Paragraph, Figure, Table}.


Figure 6: The article schema with a multi-link-class

Linearizing Hypertexts along Link Classes

The fine-grained model of the "Article" hierarchy down to single paragraphs, as outlined above (Figure 5), raises a new problem: A single paragraph is usually too small an information unit to be presented as a hypertext node of its own. Ideally, one would like to lump many paragraphs together into a larger unit. Taking this idea one step further, it would be nice to linearize the whole hypertext into a single, large document, mainly to make it accessible in printed form.

To cater for this need, HSDL link classes and link instances have an embedding mode. When activated, the location of the anchor in the source i-node is replaced with the whole content of the target i-node. Setting the link classes of some "main" hierarchy of a schema into embedding mode, will linerize the whole hypertext. Note that other, non-embedding, links remain fully functional within the single "super" node (at least in HTML). HSDL checks for and prevents attempts to create circular embeddings.

Schema Evolution

As mentioned above, authoring is an opportunistic process. Therefore, to be acceptable, a schema-based approach has to provide a wide range of operators that allow authors to revise schema design decisions. In the database field this problem is known as schema evolution [3]. In HSDL the following schema evolution operations are supported:
Add a component to a class:
All existing instances of the augmented class are extended by a corresponding node.
Reposition a component in a class:
All existing instances are restructured accordingly.
Delete a component:
The corresponding nodes are also removed from all instances. Links defined on the deleted component are moved up to the deleted component's parent.
Re-class an instance:
An instance can be made to change class. If the instance is source or target to link instances for which no equivalent link classes exist in the new class, corresponding new link classes will be inserted automatically.
Re-anchor a link instance:
The source and/or target i-nodes of a link instance can be moved to new i-nodes. If no link class exists between the components corresponding to the changed i-node, a new link class will be inserted automatically.
Transitive derivation of link classes:
A pair of transitive link classes can be used to derive a new link class (e.g. A->B and B->C ==> A->C). On the instance schema new link instances corresponding to the inner product between the existing link instances are inserted.
Triangular derivation of link classes:
From A->B and A->C derive B->C or C->B, depending on the user's choice. Again, on the instance schema new link instances corresponding to the inner product between the existing link instances are inserted.
Split a class into two new classes:
Any component of a class, except the root, may be selected to become a class of its own. Any subcomponents of the selected component as well as any link classes into or from these components are also carried over to the new class. A link class between the new class and the source class is created automatically. All instances are also split accordingly.
Merge two classes into a new class:
If, first, two classes have exactly the same number of instances and, second, there is a link class between the two classes, the link instances of which form a bijective (one-to-one) mapping between the instances, then it is possible to merge the two classes along this link class (target moves into source). If the target component is not a root component, the target class will continue existing but the target component will migrate over to the source class.

Compilation into HTML

The translation of a schema into HTML is controlled by a series of programs in the programming language Scheme [1]. These programs, called expanders, can be attached to any object in an HSDL schema. As an example, consider the default expander that generates the visible title of an HTML node:
  (define (title node) 
     (emit "<H1 ALIGN=CENTER><B>" 
           (name (class node)) ": " (name (instance node)) 
           "</B></H1>"))
The n-ary function emit writes all its arguments into the HTML file corresponding to the node that is currently being compiled. The function name generally returns the name of an HSDL object as a string. The functions instance and class respectively return the instance the node belongs to and the class that the node is an instance of.

Applying the expander above to the root node of the instance Picasso of class Painter would generate the HTML title:

   <H1 ALIGN=CENTER><B>Painter: Picasso</B></H1>

Expander Shadowing

Since expanders can be attached to any HSDL object, it is easily possible to customize the appearance of certain classes or even instances. If, for example, we attach the following (re)definition of the title expander to the class Painting
   (define (title node) 
      (emit "<H3>" (name (instance node)) "</H3>"))
then this new expander will shadow the global one defined above, and the title of the painting The Sunflowers will be expanded into:
   <H3>The Sunflowers</H3>
Instead of shadowing, we could also say that unless an expander is locally defined, it will be inherited from the higher schema levels.

To start translation into HTML the following function is called in turn for each node of the current schema:

   (define (generate node) 
      (header node)
      (body node)
      (footer node)
      (for-each generate (included-children node))
      (if (not included?) (signature)))

Compilation is incremental, that is, after an initial full compilation only changed instances are translated into HTML files.
Note that expander programming is not necessary as long as the default expanders built into the system produce acceptable HTML structures (and hence layout). Only users wishing to adapt the structure to their own needs will have to deal with expander programming.

Mapping I-Node Trees to Files

Some hypertext systems support composite nodes, that is, nodes that have internal addresses which can be directly jumped to. Since HTML is also in this category, we are offered the possibility of mapping composite HSDL instances to a smaller number of HTML files than the instance has i-nodes. Consider the class Painter of Figure 1: in a hypertext system without composites, each component would be mapped into its own node (or file). In systems which do support composite nodes, we might insert all five components into a single node, separating them by appropriate visual cues (rulers, titles, etc).
To support this feature, HSDL components and i-nodes have an inclusion status, which may be set to included or excluded. If all i-nodes of a painter are included, HSDL will linearize the tree depth-first and generate a single composite HTML file. If they are all excluded, HSDL would generate five HTML files. Since inclusion is a property of each individual node, any mapping from a composite node to one or more HTML files is possible. In every case appropriate navigation is ensured by the insertion of the necessary link anchors.
By default, the i-nodes of an instance inherit the inclusion status from the respective components. Therefore, it is possible to change the mapping of all instances of a class by modifying the inclusion status on the class schema. On the other hand, the user can override the inherited status for individual i-nodes.
HTML nodes are saved in a directory tree. The names of the directories and HTML files try to reflect the schema as much as possible. For example, the root i-node of the instance van Gogh of class Painter in the schema Gallery is written into the HTML file gallery/painter/vangogh.htm.

HTML Anchor Generation

During compilation links have to be translated into HTML anchors. To this end, the HTML string ultimately building the anchor is divided into seven parts:
  1. The anchor prefix html-prefix, which, to generate the index of a collection, would be typically set to the string "<LI>" so that it will appear as a list item in a corresponding HTML environment.
    To create such an HTML environment we distinghish between the first html-prefix in an index and all the following ones. For example, to start an HTML enumeration with a header displaying the name of the link's target class in "<H4>" size, the first prefix is set to the string returned by the evaluation of
    (string-append "<H4>"
                   (name (class (target link)))
                   "</H4><OL><LI>"))
  2. The HTML anchor start tag and attribute ("<A HREF=\"").

  3. The URL (generated by HSDL).

  4. The SGML close tag (">").

  5. The anchor label html-label, which in indices is set to the string returned by the evaluation of (name (target link))) so that the name of the corresponding node will be inserted. Of course, any legal HTML anchor label is permissible. The ubiquitous previous/next links between members of collections or hierarchies will usually have iconic labels, often images of arrows. To do this in HSDL, the link label expander is simply set to something like "<IMG SRC=\"images/arrow-lt.gif\">".

  6. The HTML anchor end tag ("</A>").

  7. The anchor trailer html-trailer, which will usually be used to set punctuation marks after labels. For example, setting the prefix to the empty string "" and the trailer to ", " will produce a comma separated list of anchors in an index.
    In analogy to the special first prefix in a collection, the last trailer in a collection is different from all preceeding ones. It is therefore easily possible to close HTML environments. The last trailer in an enumerated list will then consist of the string "</OL>".

To compile a link into an HTML anchor the function expand-anchor is called with the link as parameter:
    (define (expand-anchor link)
      (emit (html-prefix link)   ;; 1 Anchor prefix
	    "<A HREF=\""         ;; 2 HTML start tag 
            (html-href link)     ;; 3 Link target address
            "\">"                ;; 4 SGML close tag
            (html-label link)    ;; 5 Anchor label
            "</A>"               ;; 6 HTML end tag
            (html-trailer link)  ;; 7 Anchor trailer
            ))

Related Work

HDM

Garzotto et al. have shown the importance and the feasibility of using a schema-based approach to AIL in several applications of their Hypertext Document Model (HDM) [7].
The HDM schema, though, includes a much larger number of concepts than the HSDL schema. In HDM there often are many ways to model identical access structures. The lack of orthogonality can significantly raise the complexity of the implementation, especially with regard to schema evolution problems.
It is unclear, if HDM has been implemented up to the GUI level and if users can adapt hypertext generation to their own needs. HDM is used to generate ToolBook (Asymetrix Corp.) applications [2]. To our knowledge HDM has not yet been ported to generate HTML.

MacWeb

MacWeb, by Nanard & Nanard [9] [10] , also uses a class schema (called a Web of Types) and an instance schema (the MainWeb). It has a GUI (for MacOS) and a scripting language (WebTalk). MacWeb generates an internal hypertext format browsable with the MacWeb viewer.
MacWeb aims to provide the means to capture knowledge about the target domain that will be used at runtime to generate documents dynamically. Dynamic documents are adapted or even created at runtime under the control of the author's WebTalk scripts.
Such an approach could also be applied to HTML. Instead of WebTalk scripts one would have to generate programs for whatever is the scripting language of the respective WWW server. Portability would be strongly restricted, since it is not (yet) possible to intertwine HTML with code (although HotJava's Java language surely is an interesting possibility in this context).
Notice that in such approaches the problems of hypertext authoring are augmented by the problems of programming (debugging, runtime errors, etc). Programming expanders in HSDL is on a completely different level, since they are used solely to generate HTML. After compilation, an HSDL-generated hypertext is a standard HTML application.
Schema updates are much more complex if they are to transform the hypertext structure and the semantics of the attached scripts consistently.
To our knowledge MacWeb has not yet been ported to generate HTML.

Text to HTML Translators

As yet, the only tools supporting AIL in the HTML world were text to HTML translators. Such systems exist for a variety of text formats. Two of the most popular are WebMaker [13] (originally from CERN, now available from Harlequin Inc.) and LaTeX2HTML [6] by Nikos Drakos.
Text to HTML translators start from a linear document and translate the ubiquitous part/chapter/section hierarchy into a corresponding HTML hierarchy. Only anchors supporting navigation within this hierarchy are generated automatically (e.g. previous, next, up, contents). References within the text (e.g. citations) are also translated over to functional HTML anchors, but such links cannot be counted as automatically generated, they are merely translated. It is possible to embed HTML anchors into the text, but such links are created individually and the author has no global view onto the current net topology. Therefore, it is probably an arduous task to embed nodes into multiple hierarchies.
One of the advantages of the text to HTML approach, is that a high quality paper version of the same hypertext will by definition always be available. In HSDL we follow the reverse route to achieve the same effect: We start from a network of nodes and produce a linear paper version by specifying the linearization hierarchy afterwards. Complex schemas will often have a whole range of possible linearization hierarchies. The text to hypertext approach does offer this level of flexibility.

Conclusion and Future Work

We believe that the HSDL schema-based approach to HTML authoring of large hypertexts offers the following advantages: Several medium-scale projects employing HSDL are currently under way. Most of them are commercial applications revolving around technical documentation and business-to-business WWW marketing. We are looking forward to report on our and our users' experiences in the coming months.

Future Work

The interfaces between the internal HSDL modules (formalism, GUI, persistent storage, compiler, etc) have been carefully crafted to provide a maximum of orthogonality. The multiple inheritance capability of the implementation language CLOS (see below) has been essential to achieve this goal. We see the current HSDL implementation as the kernel of a system extensible in various directions:
Support for CSCW (Computer Supported Cooperative Work):
Obviously, the edition of large hypertexts will often be the task of a whole group of authors. Hence, we have to deal with the synchronization problems typical of CSCW. We believe that the easiest way to achieve this is to use a database as the persistent storage and the database locking mechanisms to enforce consistency.
Versioning:
The need for effective version management for large bodies of information is de rigueur for any system claiming AIL capabilities. Like for CSCW, using a database as persistent storage should provide a firm foundation for the extension of HSDL with version management.
Improved Instance Schema Editor:
The current instance schema editor (see Figure 2) is not really well suited for AIL. A screen containing hundreds of interconnected items will always look clogged, no matter how cleverly they are arranged.
In the short run we will be implementing an instance browser inspired on the class browsers usually found in object oriented programming environments (e.g. Smalltalk, CLOS).
As a medium term goal, we think that the article on fisheye views by Sarkar and Brown [11] offers some intriguing perspectives well worth exploring.

Appendix: Miscellaneous Implementation Details

Persistent Schema Storage:
The schema is stored in an SGML file, tagged with a custom HSDL DTD.
Implementation Language/Platform:
HSDL was implemented in CLOS, the Common Lisp Object System, in Allegro Common Lisp for Windows (ACL/W 2.0) by Franz Inc. HSDL has been tested under Windows 3.1, Windows95 and Windows NT.
Performance:
In a typical application, the compiler will generate approximately 10 HTML files per second when running on a 90MHz Pentium processor (8MB RAM).
This paper has been produced with HSDL. A meta-hypertext describing the corresponding schema and some further HSDL features can be found at http://faui80.informatik.uni-erlangen.de/IMMD8/staff/Kesseler/www4meta/home/ahyperte.htm .

References

1. Harold Abelson and G. J. Sussman. Structure and Interpretation of Computer Programs. MIT Press, Cambridge, 1985.

2. Andrea Caloini. Matching Hypertext Models to Hypertext Systems: a Compilative Approach. In Lucarella et al. [8], pages 91--101.

3. Jay Banerjee, Won Kim, Hyoung-Joo Kim, and Henry F. Korth. Semantics and implementation of schema evolution in object-oriented databases. In Proceedings of the ACM SIGMOD Conference, 1987.

4. Emily Berk and Joseph Devlin, editors. Hypertext/Hypermedia Handbook. Software Engineering Series. MGraw-Hill Publishing Company, 1991.

5. J. Conklin. Hypertext: An introduction and survey. IEEE Computer, 20(9):17-41, 1987.

6. Nikos Drakos. From text to hypertext: A post-hoc rationalisation of LaTeX2HTML. 1994. (Available via WWW UCSTRI Computer Science Index: http://www.cs.indiana.edu/ctr/search).

7. Franca Garzotto, Paolo Paolini, Daniel Schwabe, and Mark Bernstein. Tools for Designing Hyperdocuments, pages 179--207. In Berk and Devlin [4], 1991.

8. D. Lucarella, J. Nanard, M. Nanard, and P. Paolini, editors. ACM Echt 92: Proceedings of the ACM Conference on Hypertext. ACM Press, 1992.

9. Jocelyne Nanard and Marc Nanard. Using Structured Types to incorporate Knowledge in Hypertext. In Proceedings of the Hypertext '91 Conference, pages 329-342, 1991.

10. Jocelyne Nanard and Marc Nanard. Should anchors be typed too? an experiment with MacWeb. In Proceedings of the Hypertext '93 Conference, pages 51--62, 1993.

11. Manojit Sarkar and Marc H. Brown. Graphical fisheye views. Communications of the ACM, 37(12):73--84, December 1994.

12. Norbert Streitz, Jörg Haake, Jörg Hannemann, Andreas Lemke, Wolfgang Schuler, Helge Schütt, and Manfred Thüring. SEPIA: A Cooperative Hypermedia Authoring Environment. In Lucarella et al. [8], pages 11-22.

13. The Harlequin Group Limited. WebMaker Users's Manual, 1995.

About the Author

Marcus Kesseler
http://faui80.informatik.uni-erlangen.de/IMMD8/staff/Kesseler/kesseler.html.

University of Erlangen-Nürnberg
IMMD VIII - Artificial Intelligence
Am Weichselgarten 9 / D-91058 Erlangen / Germany
email: kesseler@immd8.informatik.uni-erlangen.de