contexts

Guys, mia culpa, I have only just now thoroughly read the minutes from 2 weeks ago [1]  when I was absent but y'all were discussing my idea [2] . We have got to get some things clear here.

Apparently (??) there was a consensus reached there that RDF should be considered to be inherently contextual, and that the non-contextual use (ie the one sanctioned in all W3C SWeb specs so far written, including RDF, RDFS, OWL, SPARQL, OWL2 and RIF) is *not* in fact in conformance with the way things actually are. 

I confess to being somewhat gobsmacked at this point. Because if this really is true, it means that the entire SW pyramid needs to be re-built from the ground up. RDF is then a context logic: it needs a new conceptual model and a new semantics. The way that OWL fits into RDF has to be re-thought. The semantics attached to SPARQL has to be re-written. None of the current entaiment regimes make sense any more. If we go with Antoine's 'modal' approach to describing contexts (which is indeed beguilingly simple) then the very idea that IRIs identify things would have to be re-thought, because it then takes two IRIs to identify anything, one to fix the context in which the other one is understood. (I will not dwell on the obvious infinite recursion that beckons at this point.) 

I know I am a dweller in ivory towers, but I really dont believe that this is right. While a lot of data is time-sensitive and liable to decay, and much of the data out there might be flaky or badly linked using sameAs in a wrong way, etc.., none of this real-world messiness is the same as saying that it is *intended* to be written in a context, or that RDF is, or should be, treated normatively as being a context logic. A context logic, note, is one in which the *intended meaning* if every expression is understood relative to a named context, so that you can't say "2+2=4", you have to also say which context you are saying this in, and it might be false in some other context. So every occurrence of every name (every IRI) potentially has a different meaning in every context. This is entirely a different issue from the fact that real-world RDF is messy. Making RDF into a context logic isnt going to make the RDF data any cleaner or less likely to be flaky; if anything, it just gives people a new field to screw things up in, eg by one persons's time context getting muddled with another person's provenance context, or a chunk of stuff getting mislabelled with the wrong time context label. 

Right now linked data is full of cases where sameAs is being used to link terms in two namespaces, but if we make RDF into a context logic, we will need to do this between different occurrences of the *same* IRI in different graphs. Except that won't work either if the sameAs triples are in yet another graph, which then defines its own context. Sigh. All of this *can* be done; Cyc for example uses a context logic, but it requires using a very complicated kind of global discipline and some elaborate machinery of super-and sub-contexts which impose inheritance constraints on the terms used in them, rather like Java class definitions. Cyc calls them 'microtheories'. It takes weeks of training to learn to use it properly [3]. I really dont think we want to go this way for a Web notation. (And if you think the RDF model theory is complicated, just wait until you see that one.)

One point of my 'quads' proposal in [1] is that it does not make RDF into a different logic, just a slightly larger fragment of good old-fashioned first-order logic, the one that has been around largely unchanged since about 1885, and it uses the 'context' idea in a way that has been commonplace (at least in AI and KR/planning) for about 50 years. Which after thinking about context logics is a huge relief, for me at least. 

Anyway, just a howl from the bench. And the moral is, of course, don't be absent from a telecon just after you make a new proposal. 

Pat

[1]  http://www.w3.org/2011/rdf-wg/meeting/2012-02-22
[2] http://lists.w3.org/Archives/Public/public-rdf-wg/2012Feb/0094.html
[3] http://www.cyc.com/cycdoc/vocab/mt-expansion-vocab.html
------------------------------------------------------------
IHMC                                     (850)434 8903 or (650)494 3973   
40 South Alcaniz St.           (850)202 4416   office
Pensacola                            (850)202 4440   fax
FL 32502                              (850)291 0667   mobile
phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes

Received on Monday, 12 March 2012 22:02:41 UTC