Arenas and Libkin define the background and requirements of a normal form of Extensible Markup Language (XML), with the goal of converting arbitrary XML into well-formed XML. Since my own work is with Extensible Business Reporting Language (XBRL) processor implementation, this background, and their algorithm for normalizing XML documents, is of high interest to me. Recent database implementations for storing normal-form XML data--both those implemented specifically for XML, such as Apache.org’s XIndice, and by XML features added to popular commercial heavyweight relational databases--make this technology timely and of wide interest.

The authors begin by demonstrating the need to normalize XML, based on XML data structures that have redundancies of data in different areas of the structure. They relate that to relational functional dependency. Extending from relational to XML functional dependencies, the authors then represent XML tree structured data into tuples (XBRL also does this). Functional dependencies are defined using tree tuples. Finally, an XML normal form (XNF) is defined.

Anticipating massive Web databases of poorly organized data, the authors’ goal is to provide principles for good design, and algorithms for producing that design, sort of an XML schema reengineering. My take on this process is different: I see the need for this technology in the opposite direction, not for the rescue of Web “stuff,” but for the upcoming new generations of databases and software that will all be based on XML technologies. Mainstream business processing now considers XML and XML derivatives for automated interchange between systems, and as core infrastructural technology.

Physically, this paper seems to be about the right length (12 pages). The presentation is consistent with the authors’ ideas, and the text was easy to read. The latest reference is from one year before publication.

This paper demonstrates that XML structure can be converted to a normal form, and that data can be losslessly converted into the resulting structure. In my work with XBRL, there are more expressed semantics than just the structure of data. Source fact values expressed in XML are given context, measure, role restrictions, and semantics, defined in XML linking language (XLink) linkbases. Applications of the authors’ theories, if extendable, would be very useful.

I would recommend this paper to researchers and developers of XML and XML-derived data structuring and data storing technology.