Re: Summary and some analysis: New Semantics Initiative

>Pat,
>
>   Technically, given how we are proceeding now with each language 
>having its own model theory, it is not possible to turn off a piece 
>of the graph from the perspective of all languages. It is entirely 
>possible for a new language foo (such as owl), to interpret these 
>and any other triples and provide them truth values.

Sure, of course. Which is fine. The point of the reserved vocabulary 
is so that when they do, (what would have been) their RDF 
interpretation doesn't get in the damned way, right? We only need to 
do this at the RDF level, since RDF is the only syntax that has been 
both proposed as a universal syntactic encoding for the SW and given 
a model theory.

>What we are doing is to make RDF the *weakest*  of the semantic web 
>languages, i.e., it sees the fewest truths in a given graph. Other 
>languages can and will turn on more stuff.

Right, that is taken for granted.

>
>  Even the notion of reserved is defined with respect to a language.

I think we only need to have it with respect to RDF. If OWL needs a 
special vocabulary which it wants reserved ( because its OWL meaning 
isnt going to be what its RDF meaning would be), then the OWL spec. 
can reserve that vocabulary. If OWL-2 wants to use some other set of 
urirefs, for a similar reason, then OWL-2 can reserve that. And so 
on. The only way we would need to take this past RDF would be if 
OWL-2 wanted to use some OWL constructs in a way that was 
incompatible with OWL. I guess I am expecting that by the time we all 
get to OWL-2, people will have learned enough to not try do things 
like that.

>So, we have reserved wrt rdf, entailment wrt rdf, reserved wrt owl, 
>entailment wrt owl

Well, no, we just have one kind of reserved (= RDF MT switched off, 
but maybe given meaning by some other spec). We do have different 
language entailments, inevitably; but at least with Lbase, there is a 
single overarching notion of entailment that incorporates them all.

>... If we want entailment wrt rdf to always mean entailment wrt owl, 
>then we run into paradoxes which is why we have to weasel our way 
>out by weaking entailment wrt rdf which we do by introducing the 
>concept of reserved.

Er...right, though it could be worded more diplomatically.

>We will play out the same drama when we want to layer something on top of owl.

No, I hope not. At least, I hope that we won't ever try to layer on 
top of OWL in the way that OWL is layered on RDF. Surely once we have 
a way of encoding OWL in RDF graphs, nobody is going to want to 
independently encode something else in OWL, rather than RDF, are 
they??

>This is the barest minimum we need to avoid paradoxes. Which brings 
>me to a couple of points that are currently implicit in the RDF 
>spec, but should be made very clear, especially since it is a 
>significant break from version 1 of the specs.
>
>(1) The graph is not the model. It is a piece of syntax.

Right, we have tried to be very clear about that, we even call it the 
'graph syntax'.

>(2) The main thing that is in common between the languages of the 
>semantic web is ... : We need to answer this. They don't use the 
>same model. They do use the same syntax. What else?

Well, since we basically only have two, and since one of them isn't 
fully specified yet, its a bit early to say.

Heres my current understanding of the 'vision'. Each web language has 
two mappings: a meaning-preserving embedding into Lbase (which can be 
used to define its semantics) and an invertible mapping of its syntax 
into RDF graphs (used to transport its expressions around the web 
through RDF engines). To avoid confusion, the following diagram 
should commute:

SWEL expression --(syntax embedding)--> RDF graph
       \                                /
        \                              /
   (SWEL semantic embedding)    (RDF/SWEL semantic embedding)
           \                        /
             \                     /
              \                   /
                 (Lbase axioms)

should commute. This requires that the SWEL/RDF translation into 
Lbase respect the SWEL rather than the bare RDF rules wherever they 
would disagree; and so to keep monotonicity, we would need it to be 
the case that such parts of SWEL/RDF use suitably reserved 
vocabularies to cancel the RDF interpretations that would cause 
trouble.

I don't see that we would ever need to consider the analogous issue 
for anything other than RDF, since other languages which use OWL 
constructs will use them with their OWL meanings (why not?) and 
nobody is suggesting using anything but RDF as a universal syntax 
acid.

>(Please note --- I am not proposing the above. I am merely 
>suggesting that what is hidden in various parts of the spec be made 
>more explicit and clear).

Amen to that.

Pat

>guha
>
>patrick hayes wrote:
>
>>Yes, that was my suggestion to Webont a while ago. That however 
>>seems to inevitably run into this nonmonotonicity issue, since if 
>>some OWL can switch off some RDF content, things can get into a 
>>twist. SO the idea is to only allow the W3C to say when names have 
>>no RDF assertional weight, basically, so that any piece of RDF is 
>>always either on or off, seen from any language perspective.
>>
>>Pat
>>
>>PS delay on the next MT draft is due to a fatal OS crash on my 
>>machine which is screwing me up. SOrry, this has cost me at least a 
>>day.


-- 
---------------------------------------------------------------------
IHMC					(850)322 0319   cell
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax