Critique of RDF Draft of 1998.2.16
Henry S. Thompson

13 March 1998

1. Purpose and background of this critique

Stable and standardised provision of meta-information about web-transmitted documents is tremendously important, and I am committed to the seeing the RDF project succeed in addressing this need. I'm very concerned that there are two ways in which the current draft fails to do justice to itself and to the thinking that has gone into this effort so far. By way of background, I should say that I've worked from an AI and computational linguistics perspective on knowledge representation language design and implementation off and on since 1976, and have a reasonable background in formal logic as well.

I'll try to set out my concerns below, in hopes they can be addressed before the document goes really public. Can I emphasise at the start, since I don't have time to wrap the criticisms in cotton wool, that my intentions are supportive, not destructive---this stuff is too important to give external critics irrelevant grounds for dismissing it.

2. Problems with levels and vocabulary

I suspect that section 2 is now so old-hat to most of you that you can't really read it anymore, but reading it for the first time from an interested technical perspective, it's extremely difficult to work with, not because the basic ideas aren't clear, but because the vocabulary employed to talk about things is inconsistent and misleading, and the underlying conception of levels of description and/or implementation is itself unclear.

It's not surprising that these difficulties have arisen: describing meta-languages is notoriously tricky and necessarily self-referential. But there is precedent, namely formal logic, which you've drawn on for some of your language, and I think going all the way and using the language of formal logic systematically would vastly improve the coherence and comprehensibility of this part of the document. Note I'm not suggesting that you start the section by saying "RDF is a logic for describing web-transmitted documents", although that's true, because I recognise that the principal effect of such a statement would be to turn off a substantial part of your intended audience.

So before reconstructing the first part of section 2, what is the logic? It's a second-order modal propositional logic, that looks like this:

Note that while RDF properties and RDF strings denote things in the meta-domain of description, RDF resources denote things in the object domain of the WWW.

So for example, the following is a well-formed RDF sentence:

       author("","Lex Holt")) & 
       author("","Jonathan Kilgour"))