SGML Documents: Where Does Quality Go?

Quality control in electronic publications should be one of the major concerns of everyone who is managing a project. Big projects, like digital libraries, try to gather information from a series of different sources: libraries, museums, universities, and other scientific or cultural organizations. Collecting and treating information from several different sources raises very interesting problems, one being the assurance of quality. Quality in electronic publications can be reflected in several forms, from the visual aspects of the interface, to linguistic and literary aspects, to the correctness of data. With SGML we can solve part of the problem, structural/syntactic correctness. SGML provides a nice way to specify the structure of documents keeping a complete separation between structure (syntax) and typesetting. Today there are lots of editors and environments that can assist the user producing well-formed and valid SGML documents (validating their structure). However, current software still gives the user too much freedom. The user has full control of the data being introduced, creating a margin for errors. In this context there are situations where pre-conditions over the information being introduced should be enforced in order to prevent the user from introducing erroneous data; we shall call this process data semantics validation. The idea is to constrain the values of some structural elements of a document according to its final purpose. This way the user (who writes the documents according to that DTD) will not have full control of his data; he will be forced to obey certain domain range limitations or certain information relationships. SGML does not have the necessary constructs to implement this extra validation task. In this paper we will present and discuss ways of associating a constraint language with the SGML model. We will present the steps towards the implementation of that language. In the end, we present a new SGML authoring and processing model which has an extra validation task: semantic validation. Along the paper we will show some case studies that could have their quality improved with this new working scheme.