(some loose, still very much unordered rambling...)
Seeing as ISWC2008 is around the corner, and we are in the process of getting the new 'dog food server' (http://data.semanticweb.org) out, my main interest at the moment is in describing scientific conferences and things that have to do with then: mainly people, organisations, papers and talks. While working on that, I had an idea that we should probably think about the way we use the words "vocabulary" and "ontology". What people often mean by those words (questions about communal agreement aside) are terms that are in one namespace. However, looking at the SWC (Semantic Web conference) "ontology" we are using on the dog food server, I'm inclined to change that to "the set of terms that are used for one purpose". So, what I call "SWC ontology" actually comprises terms from a number of different namespaces (foaf, swrc, ical, its own namespace), etc. I'd still like to call it an ontology, because it is what me and other SW conference dog food tzars agreed on using for describing those conferences.
This brings me to another topic: I'm looking for a way to find properties or classes for specific purposes. E.g., I was recently looking for a property that I could use for saying that something is an acronym of something else. After some asking around on the swig IRC, I couldn't really find anything good. Similary, what is a good property for saying what the logo is of something? When I want set about describing some domain of interest X, I don't want to reinvent those properties all the time, but rather find existing ones that fit my purpose, and then build my new ontology for X by picking terms from here and there. The ontology then is a lot of terms from different namespaces, plus some extra ones I probably didn't find anywhere, and, very importantly, some guidelines or best practices of how to use all those terms together. Ontology by example: this is how we describe a person in SWC, this is what a paper looks like, this is an organisation, etc.
There shouldn't be so many different terms that basically mean the same thing. Just take "this is the name of that": rdfs:label, foaf:name, ical:summary, dc:title, dcterms:title, etc. It doesn't really help to use rdfs:subPropertyOf, owl:sameAs or whatnot - hardly anyone is ever using the reasoning that is necessary for this kind of stuff. SPARQL doesn't support it.
Unpopular (?) idea: would it make sense to set up a pool of common terms and call this a "SW base ontology"?! "An ontology which let's you say most of the basic stuff."