Vincent Quint
INRIA Rhône-Alpes, Grenoble, France
Documents available on the World-Wide Web can roughly be separated into two categories:
Anchor documents are mainly composed of a collection of links pointing to other documents. Their role is to provide readers with a short piece of information and to give them access to other documents. They contain mainly URLs, surrounded by some pictures and text. Their structure is very simple.
Information documents contain more text and provide much richer information. They are in some way self-contained, even if they refer to other documents. They are also much longer and are structured, at least in sections, subsections, etc.
The HTML DTDs fit the needs of the first category, but they pose problems with the second category:
Information documents are represented in different formalisms (PostScript, dvi, ASCII, etc.) and cannot be read or browsed with the same tools and in the same way as other documents.
When they are represented in HTML, reading them is not easy, as they are either to long or divided into a number of pieces.
No comfortable tools are available for authoring such documents and publishing them on the Web.
For solving these problems, one could use two different types of document representations on the Web: a simple formalism, HTML (including HTML 3, with a rich representation of tables and equations), and a more powerful one, providing high level features for handling structured documents.
Taking the logical structure of documents into account is an efficient way of allowing a number of manipulations, for both the writer and the reader of documents. SGML is a good candidate for the second representation. It is able to represent very diverse documents, by using a specific DTD for each type of document, and a number of tools are (and will be) available for manipulating such documents. It also allows simple conversion to HTML. Although converting a document from HTML to any SGML DTD remains a complex problem, there exist solutions for converting some HTML documents to some DTDs. Thus, the two representations will not be incompatible and information documents will have the same status as anchor documents.