When writing RDF (in N3 normaly) by hand, the # methd is very easy, and works well. You naturally write in files of manageble size which are easily delivered over the net. The thing after the # is the local identifier in the file, and you can add also som identifiers for other related things which are naturally described in the same file.

When a large set of identifiers are generated automaticaly, then the 303 method may make sense, but the # method will work too, even if every uri ends “#it”, which looks odd. It is faster from the client’s poiunt of view.

For example, the tabulator, which, like many things, looks up the ontology for each predicate or class term it comes across, has to look up every single Dublin Cote term (which has no #) like dc:title separately just to find each redirects to the same ontology file.

• Berlin the city (the thing itself; the concept)

• A document describing Berlin

Only cognitive scientists or cultural historians would bother to divide the first one into two distinct entities. I’m not one of those, and therefore don’t bother to make the distinction, and believe that the way I’ve presented it is perfectly clear for the intended target audience.

If you have a good term for “those URIs that do not identify documents but other things, like cities or people or software projects“, a term that is understandable to non-experts, then please share. The best I could come up with was “concept URI”. And if you offer “entity URI” or “non-information resource URI“, then I shall personally lock you up in an ivory tower.

After introducing a distinction between berlin-the-concept and a-document-about-berlin, you say “something that is located in Germany and something that mentions the class Feature”.

Do you really mean that concept of Berlin is located in Germany?

In actual fact there are 3 or more entities:

• Berlin the city (a physical entity)
• The geonames document record concerning (presumably) the city of Berlin
• The concept of Berlin

We can probably dispense with the last entity unless we are cognitive scientists or cultural historians.

Once you correct this then your analysis of the problem is a good one

Summary: the 303 approach is new, but the problem is so boring and dumb that every month another nitwit picks it up to discuss it endlessly.

Face it: you cannot identify concepts, because concepts are only perceived by humans and do per se not exist. Philosophy helps here: you only perceive, and you use language to communicate facts about perceptions. So there is no uri for the document or for the concept. Think of words. You are reading my words at the moment, but is there an identifier for identifier besides identifier? If URIs are like words, then it depends how you use them in social behavior. If you decide that “http://ws.geonames.org/rdf?geonameId=2950159″ identifies berlin and enough people copy this, then its the identifier for Berlin for these people, like the “alt-names” in the document at this uri.

What you can do is use topic maps resource-identifiers or what we used in the PIMO language, occurrenceRefs, see here: http://ontologies.opendfki.de/repos/ontologies/pim/pimo-language.rdfs

see also: http://www.dfki.uni-kl.de/~sauermann/2006/01-pimo-report/pimOntologyLanguageReport.html#ch-IdentificationIssues