Musings on my ideal syntax for RDF-in-HTML

Sun Jun 24 2007 at 17:06

Looking again at the bnode comparison of µf, eRDF and RDFa I’ve been thinking about the relative advantages and disadvantages of eRDF and RDFa a bit. The great thing about GRDDL of course, is that if I come up with a syntax of my own that I prefer, I can just write a stylesheet and get on with it. I don’t need to evangelise it, or even document it.

Whether or not it’s worth the extra effort (over using an established syntax like eRDF or RDFa) is another matter of course ;).

I think my dream syntax would mainly be quite similar to eRDF, but enabling a few of the things that RDFa does.

What I like about eRDF

The main thing I like about eRDF is that all visible text and images are described with class-names. The class names, moreover, don’t contain any characters that need to be escaped in CSS. This is actually an extremely clever feature of eRDF. It encourages good separation of style and content; you can start with your content, mark it up with html and eRDF, and then when you come to style it, you have a lot of hooks to style something according to what it is.

I just read a post to public-rdf-in-xhtml-tf by someone concerned about @class being used for semantics instead of presentation. I\'ve seen this argument play out in a few places, and I\'d just like to add, not only is @class for semantics in the html spec, not only is it good practice for maintainability and so on, it\'s good practise for design.

A principle of good design is creating visual similarities between similar types of information; so it\'s a good idea to style, say, e-mail addresses the same way, and if you can use .foaf-mbox as your style hook, then so much easier for you. When I write CSS, semantic class-names make it easy for me to style elements by the type of information they contain, which helps the user visually identify them, and helps make the design consistent and clean. eRDF class names are even better because they come from shared vocabularies, which are easier to remember, understand, and re-use across projects than the ad hoc conventions I\'d use otherwise.

What I like about RDFa

The main things I like about RDFa are:

• You can declare namespaces at any stage in the document (if you’re blogging or something, you don’t need to edit the template whenever you want to use a new vocabulary).
• You can describe resources external to the document (eRDF’s support for this is limited).
• You can choose to have HTML and XML content in your object literals (eRDF only takes the text content).

A mix of eRDF and RDFa ?

So what I want is a syntax that:

• uses class-names to describe all visible text and images
• is element agnostic – (perhaps making the technique useful beyond pure (x)HTML). I feel that the semantics of html elements, and the semantics of RDF triples are qualitatively different, and I\'d rather keep them separate. (In a little more detail, if an eRDF-style classname exists on an element, the object is, in order of preference: the @src, or @value, or child element(s) with @selected or @checked, or @id, or @title, or inner contents of the element).
• gives me control over exactly which triples are generated (I don’t really like eRDF’s generation of rdfs:label triples from anchors)
• lets me declare vocabularies anywhere in the document – not sure how to do this: XML namespacing feels a little wrong if you’re not using the colon separator. Also, as Ian Davis said about creating eRDF:

  I wanted to be mindful that I wasn’t inserting URIs into class attributes but that I was mapping values from class attributes to URIs. Sort of a subtle point. The point is that I’m not trying to change or extend HTML, but add hints about how it can be interpreted as RDF.

  Ian Davis, Re: RDFa reliance on namespace declaration, Fri, 16 Jun 2006 04:23:08 -0700

  An alternative is to allow schema declarations in anchor tags as well as link elements. This is semantically right, but structurally feels a bit messy.
• lets me choose to have HTML/XML content, or strip the tags. Not sure about the best syntax for that – maybe an extension to the class-name syntax, eg: sioc-contents-XML when you want all the tags in the output preserved.
• lets me describe external resources, at least a bit. I don’t want to have to use non-HTML attributes like @about if I don’t want to. I think if descendant elements of an hrefed element have an eRDF-style class-name, the subject would be the @href, rather than the nearest IDed ancestor of the anchor. This is limited (unless you want to make massive portions of your document links), but would allow more semantic annotation of hyper links. Perhaps you could use @about, or owl-sameAs for larger portions of description.
• ideally the html would validate. It’s not the worst thing in the world if it doesn’t in every situation (I would maybe sacrifice validation for the occasional @about), but it would be nice.

What would your ideal RDF-in-HTML syntax be like?

RDF-in-HTML RDFa eRDF