- From: John Black <JohnBlack@kashori.com>
- Date: Fri, 22 Dec 2006 09:49:34 -0500
- To: "Harry Halpin" <hhalpin@ibiblio.org>, "Joshua Tauberer" <jt@occams.info>
- Cc: <semantic-web@w3.org>
Ok. You are suggesting something like this. A set of statements in a context will be represented by a named graph[1]. And the URI of the named graph will, when http-dereferenced return an instance of a StatementContext class. And this context individual will have a property of utteredBySpeaker. And the individual in that slot will be the referent of a blank node in the enclosed named graph. And an application, with a custom, non-standard function, can then rewrite the graph to replace the blank node with the value of the utteredBySpeaker property of the context. After that, processing can continue acording to standards. @prefix contexts < http://www.kashori.com/ontology/contexts.owl# > . @prefix ex < http://example.com# > . contexts:Anouncement { _:I ex:lifeStyleChange ex:married . } The name of the graph when deferenced will retrieve an instance of StatementContext. contexts:Announcement a contexts:StatementContext ; contexts:utteredBySpeaker contexts:JohnBlack . Now a contexts-aware application could rewrite the first graph, substituting the speaker contexts:JohnBlack for the blank node _:I because in the context of http://www.kashori.com/ontology/contexts.owl#Announcement _:I refers to contexts:JohnBlack I'm putting more about this here: http://kashori.com/wikiPim/SemanticWebStatementContexts 1.http://www.w3.org/2004/03/trix/ cheers, John Black www.kashori.com Harry Halpin wrote, > > I think the methodology of contexts (which could also be thought of as a > variation of named graphs, where the URI of the "name" names the > context) that Joshua Tauberer has suggested might be the best for the > time being. > > In essence, part of the problem is that "I" and "you" and other > indexicals are actually "variables" that, as Joshua put it, then given > in a context, map to individuals. Since there is no notion of a variable > in RDF as it is (although there is in N3, and one would assume, whatever > RIF turns out to be), well, then this is going to be hard to model > coherently given current standards. It would, actually, be a good > use-case for including variables in the SW layer cake. > > One aside - the treatment of deixis in formal semantics is hard (I'd > recommend looking at Barwise and Perry's work in situation theory for > one way to do it, and Guha's work on this is also excellent) and since > by its very nature, formal semantics is trying to achieve some degree of > context-independence, it's always going to be messy. > > However, I'd like to make an aside - in the Semantic Web community, > we're always striving for globally sharing information, and this means > in practice elimating some of the context in order to share - and this > will always do some violence to the subject matter being modelled, so no > matter how hard we try to get context independence there's always going > to be some level of brushing context under the carpet :) > > Furthermore, if we're going to use "context", we're going to need some > way of putting it up front, and my bet is variables, possibly with > "names" is the way to go. > > Joshua Tauberer wrote: >> John Black wrote: >> >>> Are the following URI allowable according to web and semantic >>> web standards? Are they ambiguous? Are they useful? In each case, >>> the referent would depend on the context of the use[1] of the URI. >>> http://kashori.com/ontology/indexicals.owl#I >>> http://kashori.com/ontology/indexicals.owl#you >>> http://kashori.com/ontology/indexicals.owl#this >>> http://kashori.com/ontology/indexicals.owl#it >>> http://kashori.com/ontology/indexicals.owl#here >>> http://kashori.com/ontology/indexicals.owl#there >>> http://kashori.com/ontology/indexicals.owl#now >>> >> >> I think one of the major benefits of RDF is that applications completely >> oblivious to any schemas can still at least make one conclusion about >> URIs, which is that if two documents use the same URI, they are >> referring to the same thing. Context-dependent URIs break this. It >> might be fine in some cases, but in general it would break the principle >> that <indexicals.owl#I> refers to the very same thing no matter where it >> appears. >> >> >>> Therefore they would only be of use if they could be embedded in a >>> structure that specified a context. >>> >> >> That would be fine, except afaik the only way to publish a triple in an >> embedded context is with N3 formulas. You can't, for instance, create >> an embedded context in RDF/XML. (It wouldn't be sufficient to add a >> triple to an RDF/XML document to say "hey, this is one of those >> documents where you interpret <indexicals.owl#I> as ME" because it >> doesn't solve the problem that oblivious applications won't know to do >> that.) >> >> But, there is a way to get around these problems, which is to use some >> indirection. Actually, that actually parallels the real-world side of >> things. That is, we're not all named "I" (in the sense that a URI is a >> name for something). "I" is really a function from a context to an >> individual. The closest thing to a function in RDF is a predicate, so >> you could do this (in N3): >> >> <> indexicals:isBeingReadBy _:you . >> <> indexicals:isBeingReadAtLocation _:here . >> (then you go on to make assertions about _:you and _:here.) >> >> where <> is the URI for the document itself, and _:you and _:here are >> bnodes. Like this, an application isn't going to make the mistake that >> the same person is the "I" of every document. Here it says "someone is >> reading the document", and just when the applications understands >> indexicals:isBeingReadBy it can say "Oh, and _:you is actually me!". >> >> (I'm not positive that that actually solves every problem, but it seems >> to be a step in the right direction.) >> >> > > > -- > -harry > > Harry Halpin, University of Edinburgh > http://www.ibiblio.org/hhalpin 6B522426 > >
Received on Friday, 22 December 2006 14:54:18 UTC