Re: Semantic markup of documents:

> In my view, it is more of a problem to nail 
> what meta information/relationships we want to describe 
> rather than how.

I agree it's more important.
 
> I think that concerning navigation *within a page* 
> sticking to structural markup and meaningful text, 
> in particular link descriptions (anchors' content)
> should do the job. 

where does "structural markup" stop and "metadata" start is the 
issue: what's data to one application is meta to another.
 
> If a page is so complicated that it needs meta 
> information about the purpose of links, there is 
> something fundamentally wrong with the page.

how about the example of attributes added to table to allow for better 
navigation: this is within a page.

> I think that the main usability problems are in navigating 
> within a website, that is, between pages.

where does frame navigation or form navigation fit ?

> This may be addressed by using LINK elements
> that describe relations between the source document 
> and other documents. User agents may display/process them in 
> some way. (I avoid the important question 
> what are the relationships we want to 
> describe, because I don't know the answer)

the HTML4 spec already defines a list of such relationship
(http://www.w3.org/TR/REC-html40/types.html#h-6.12)
 
> Using LINK is not in contradiction of using RDF.

agreed, to the extent that they can both be used together without
breaking anything

> They are complementing each other.

also agreed to the extent that one (LINK) is a much simpler way of
doing something the other supports (or will) more exhaustively.

> The RDF draft suggests to refer to an external RDF 
> document using <LINK rel=meta>. 

yeah, but that's really a last resort hack (to paliate poor xml
include facilities in html) and shouldn't be used as an argument