Re: MT for imports (was: Re: Imports Proposal)

>Pat,
>
>I can live with this the proposal. It seems like we have to go to a lot
>of trouble to make this fit into RDF

The problem isn't particularly to do with RDF. What we are bumping up 
against here is the idea that web languages express propositions, 
it's something very basic.

>, but we made our bed, I guess we
>have to lie in it. I'd like to hear what others (especially Peter),
>think of this idea.
>
>I have a few comments on the wording, please see below...
>
>Jeff
>
>
>pat hayes wrote:
>>
>>  >Pat,
>>  >
>>  >The proposal we were discussing on the telecon today is at:
>>  >
>>  >http://lists.w3.org/Archives/Public/www-webont-wg/2002Nov/0004.html
>>  >
>>  >I hope you can come up with wording that you find suitable, while not
>>  >changing the basic idea considerably.
>>
>>  OK, I will write it in terms of triples, hope that is OK with you.
>>
>>  -----------
>>  Short version.
>>
>>  aaa owl:imports bbb .
>>
>>  means that the document aaa asserts the content of the document bbb.
>>  If this occurs in document aaa itself, then it means something like
>>  'I assert bbb', ie it means the same there as if you had copied the
>>  OWL content of bbb and inserted that into aaa in place of the imports
>>  triple. Call this an "I-import triple". Since this rule applies to
>>  any owl:imports assertions in bbb, and so on, the assertion of a
>>  document amounts to the assertion of the imports closure of the
>>  document. Therefore, a document containing an I-import triple entails
>>  anything which is entailed by the imports closure of the document
>>  referred to be the object, in the usual sense of 'entails'.
>>
>>  This does not, however, mean that a reasoning engine is obliged to
>>  actually perform any fetching or inserting operations when it comes
>>  across an imports statement; the use of the term 'assert' here can be
>>  understood as meaning something like 'assent to' or 'agree with'.
>
>I object to singling out imports for this kind of statement.

The only reason I did so was because it has often been given this 
kind of imperative reading (eg in many of our earlier drafts) and I 
wanted to emphasize that this wasn't being said here. It was more a 
herding in than a singling out, if you take my meaning. I agree with 
what you say below. But I think a note is needed, perhaps in a less 
in-your-face position, because many readers are going to arrive with 
a presumption that imports is basically an imperative, and they need 
to get that cleaned out of their heads to avoid misunderstanding.

>This is
>true for every construct in the language. For example, a reasoner is not
>OBLIGED to compute the subsumption hierarchy given a set of complex
>class descriptions, but if it doesn't then it cannot be considered a
>complete (as in sound and complete) reasoner for the language.

True.  But do you want to say that a reasoner faced with an imports 
and wanting to be complete *must* perform a file transfer? I wouldn't 
even go that far. Who knows, it might have some other way of knowing 
what the content of the imported RDF amounts to.

Part of the merit of a model theory is that it purges the last whiff 
of imperativeness from the stables.

>We should
>make clear the difference between obligation and entailment in general
>for the language.

Indeed.

>
>>  If bbb does not identify an OWL document, the 'imports' statement is
>>  equivalent to importing an empty graph, ie it is vacuously true, ie
>>  meaningless; the graph means the same as the subgraph got by erasing
>>  it.
>>
>>  If aaa contains
>>  bbb owl:imports ccc .
>>  where bbb does not identify aaa, then this does not entail any
>>  consequences of ccc (or its closure.) The difference is somewhat
>>  analogous to that between
>>  (bbb implies ccc)
>>  and
>>  (bbb and (bbb implies ccc))
>>
>>  -----------
>>  Long version.
>>
>>  The truth conditions for owl:imports are non-standard in two ways,
>>  both of which require us to modify the semantics, but in ways that
>>  are orthogonal to the other semantic issues, fortunately.
>  >
>>  First, they are given for a particular token of owl:imports in a
>>  particular document, and may be different for other tokens of the
>>  same triple in other documents. This is actually not possible in a
>>  conventional semantic theory, but we will tackle it directly by
>>  altering the semantic rules. An interpretation is usually defined in
>>  terms of a mapping from a vocabulary, ie a set of names. We will
>>  instead consider this to be a mapping from a set of name *tokens*,
>>  where we will say that for all tokens not in an imports triple, that
>>  all tokens of a given name map to the same meaning. This reproduces
>>  the usual semantics for the rest of the language but allows us to
>>  distinguish one token of owl:imports from another. (To sum up, one
>>  could say that the meaning of owl:imports is *essentially indexical*
>>  (http://csli-publications.stanford.edu/site/1575862697.html) )
>>
>>  Second, they use urirefs to refer to documents, in ways that go
>>  beyond the standard semantic rules for RDF but which have been argued
>  > for by many eminent authorities. So let us follow those authorities
>>  and say that any token of an absolute HTTP URI in an imports triple
>>  must always be interpreted as denoting the document which would be
>>  retrieved by using the HTTP protocol on the WWWeb, so that for
>>  example I(http://www.coginst.uwf.edu/~phayes) is required to be a
>>  certain document with a picture of an idiotic grin on it, in all
>>  interpretations. This leaves open the issue of how to define
>>  'document', but we need not do that, since the only documents we need
>>  to worry about are OWL documents that define an OWL/RDF graph; so we
>>  will treat all such HTTP URIs as denoting an OWL/RDF graph, and if
>>  there isn't any OWL in the document, or if you get a 404 error, then
>>  that is the empty graph.
>>
>>  However, since the result of an HTTP operation depends on the state
>>  of the WWWeb, the truth of any particular owl:imports token may vary
>>  depending on the state of the web. Inferences made from any such
>>  triple should therefore be considered to incorporate an implicit
>>  reference to the state of the WWWeb at the time the inference was
>>  made. Users SHOULD use some means to distinguish state-dependent
>>  inferences from non-state-dependent inferences. Owl:imports is the
>>  only item in the OWL namespace that requires this special treatment.
>
>I would think that any inferences on the Semantic Web should incorporate
>an implicit reference to the state of the Web at the time the inference
>was made.

Well, that might be nice, but in fact that has been considered for 
RDF(S) and explicitly rejected at this stage. The RDF(S) - and hence 
the DAML+OIL and OWL - semantics are quite explicit about it; they do 
*not* take account of time-varying data, changes to assertions or any 
of the other 'real' complexities like this. This is one of the 
reasons why reification was explicitly taken out of the RDF model 
theory, and they do not refer to the state of the web, even 
implicitly. So if we incorporate something into the language which 
does explicitly have this kind of formal semantic condition, then we 
are breaking important new ground here, and we ought to announce this 
very clearly.

I really think this is not a routine bread-and-butter thing we are 
doing with 'imports'; it boldly goes where no web language has gone 
before. We ought to make a big fuss about it if we decide to do it.

>Imports is not special in this aspect.

Yes, it is, at least with the semantics given here. Its formal 
meaning changes when the web changes. Nothing else in any web 
language so far does this. Erasing some RDF ceases to assert the 
content of the RDF, but it doesn't *change* the content. The erased 
RDF still means what it meant when it was being asserted, and if you 
re-assert it again then it will still mean what it did before. But 
some imports somewhere else that refers to my RDF will switch 
meanings, invisibly, without changing its outward form at all. Its 
like a notice on a shop-front that points to the next building and 
says 'next door is a grocer', whose truth changes depending on what 
is built next door, without anyone doing anything to the notice 
itself.

>The Web changes, this
>is a fact. The triples on a single page can change just as easily as
>pages can come and go (and quite likely, much more frequently than
>that). I can choose to put a bunch of triples in one page, or distribute
>them among a hundred pages. If I delete a set of triples from the single
>page, then this is equivalent to removing one of the distributed
>documents from the Web. Should not these two types of changes be treated
>equally in the semantics?

Well, no, see above. That is, theres a real difference between 
changes in assertion and changes in meaning (a difference that 
imports gets muddled, by the way.)

But in any case, so far NONE of the semantics refers to ANY kind of 
change. We are breaking new ground here. This is not a lightweight 
area to get involved in. It gets us into issues of temporal 
reference, common time ontologies, distinctions between transaction 
time and reference time, truth maintenance strategies, nonmonotonic 
justification reasoning (nobody told me it had changed, so I assumed 
that it hadn't) indexicality, all kinds of stuff. Im not saying that 
imports necessarily involves all this, but the whole area of handling 
changing data does.  Just blandly assuming that we have been in this 
world all the time isn't being realistic, seems to me. We really 
havn't.

BTW, owl:imports actually makes all this more complicated, by setting 
up time-dependent relationships between the content of different 
documents. Consider the following sequence of events, all at 
different times in the order given;
A publishes P
B publishes Q which imports P
C reads and archives Q
A changes P
C draws a conclusion.
Now, whose fault is it that C's conclusion is no longer up-to-date? 
The question is rhetorical: I can see good arguments all possible 
ways. I mean only to emphasize that the importing makes it harder to 
keep track of the content of changing information sources, not easier.
What happens next is
B updates Q so it no longer imports P
A changes P
C archives Q
B changes Q so it re-imports P
....

>If I have a document d with a set of triples t
>that has an entailment p, and then I modify d so that it consists of
>triples t', it may no longer be true to say d entails p. There is
>clearly an implicit reference to the state of the Web there.

All true, but beside the point. The key word is 'implicit'. Its so 
implicit that its not even mentioned. You might as well say that 
because OWL can be written down on paper, there is an implicit 
reference to the ink. The fact is that although we all know this 
stuff happens, nobody so far has made any attempt to incorporate it 
into the semantics of the languages.

Pat

-- 
---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola              			(850)202 4440   fax
FL 32501           				(850)291 0667    cell
phayes@ai.uwf.edu	          http://www.coginst.uwf.edu/~phayes
s.pam@ai.uwf.edu   for spam

Received on Tuesday, 12 November 2002 12:18:16 UTC