- From: Richard Newman <r.newman@reading.ac.uk>
- Date: Sat, 23 Sep 2006 00:54:49 +0100
- To: g.tummarello@gmail.com
- Cc: semantic-web@w3.org
> Isnt that as ambiguous as any human readable description, includign > the comments which describe the terms of ontologies? Possibly. The difference is that the tdb URIs are used to describe a resource by its relation to another (mostly non-RDF) URL, which is arguably less precise; the only distinctive part of the URI is the HTML document it refers to, and there are many-to-many relationships between the referents and the signs, to borrow semiotics terminology. > I am a fan of tag uri! > we support them in DBin (the first thing you get when starting up is a > proposed tag uri for you as a user). But they are to be used if you > want to make sure there is no possibility of collision. Talking about > pets, beers YOU brew etc. For any other (and most?) identification > purposes you really want to "collide" with others.. It's a tradeoff, of course, between allowing 'good' accidental collision and avoiding 'bad' inadvertent collision. Completely unique URIs will never cause erroneous collisions, but they don't allow useful accidents. I suspect that the right thing to do is not to rely on colliding URIs at all, but instead to use IFPs, CIFPs, or rules to infer identity. IMO the useful collisions are mostly limited to 'native' web resources (mailto: URIs, web resources themselves, SIP endpoints) and not the real-world things to which ontologies refer. > If you only mint your own then put a bunch of "sameAs" would be the > same.. are we not allows to talk about URIs we dont own? Talking about > and minting sound the same to me .. The difference between these (using the same URIs versus using different URIs and relating them) is the ability to control the inference of identity. I could only infer identity based on trust relationships, say, or operate in a strict mode where I don't merge data at all. When I refer to someone else's URI, and it's owned by them, I am implicitly agreeing that I'm referring to the same thing that they are referring to, and they can choose what that is because they own the domain. (The point of the ontology comments is to try to unambiguously convey the referent to me.) When two people mint a tdb URI, and they collide, there is a risk that they are each referring to a different concept with the same identifier. There is no way of deciding who is wrong, or even detecting the problem. >> Not necessarily, but it's recommended that dereferencing a URI >> returns an 'authoritative' description of that resource. > > when the URL is hostd by someone who "owns the idea" .. +1 tdb The problem is that the authoritative description is not RDF, so one can't be sure precisely what it's a description of. It allows for useful collision at the human level (we probably have some idea of what http://slashdot.org/ is referring to), but I daresay it's not enough for machines. ("Slashdot contains 2 million comments, is composed of 200,000 users, started in 1994, started in 1999, has a profit margin of $30k, and 6 million rows stored in an Oracle database? That's confusing!") >> > the semantic of it however allows one the >> > decide whether to ignore the date and smush on that.. or do >> > whatever else, also according to the "type". >> >> URIs are meant to be opaque. Better would be to use named graphs to >> provide a context so that the date, user etc. become explicit. > > From my understandign the "uri opacity" in the SW is somehow an > extension to a principle which is sound sound limited to HTTP only.. ... > Example.. if its a ISBN uri it could show you a page from your > favorite bookstore with prieces and comments about that book. If its a > URL.. well.. just invoke the system browser, if it is a tag URI, since > the specs say "put your email" then it could show a button which says > "email the minter and ask what he meant :-) " :) For something like an ISBN in a book application, then sure -- introspect away. I think that's specialized and rigorous enough. In a more general sense, though, there are several reasons to keep discrete pieces of data separate within the model, rather than within the URI. For instance, in this case I'd have to put conditionals throughout my URI handling code to spot these URIs and act on, or ignore, the date; I might have to do that for tdbs, tags, various URN info schemes... I generally feel that if another approach exists it's better to take it. -R
Received on Friday, 22 September 2006 23:55:33 UTC