Re: semantics update

From: pat hayes <phayes@ihmc.us>
Subject: Re: semantics update
Date: Wed, 30 Jul 2003 01:24:15 -0500

> >From: pat hayes <phayes@ihmc.us>
> >Subject: semantics update
> >Date: Mon, 28 Jul 2003 23:29:18 -0500
> >
> >  > http://www.ihmc.us/users/phayes/RDF_Semant_Edit_Weak.html
> >>
> >>  now reflects the post-Beckett/PFPS/Carroll editing and all subsequent
> >>  decisions. It has a slightly different look/feel (colors, table
> >>  titles etc) , updated references (though if anyone has any complaints
> >>  about any of those please feel free to correct me) and some more
> >>  anchors and internal links. Ive made the text links slightly visible
> >>  by messing with the background color, but if people don't like this
> >>  its easy to tweak it in some other way.
> >>
> >>  Text changes since the last version are in red. They include
> >>  rewriting of the definition of 'merge', some minor rewordings to
> >>  clarify meanings more carefully, and putting back the definition of
> >>  'vocabulary entailment' which had gotten lost somewhere (its now
> >>  section 2.1 and has several links to it) and references to blank node
> >>  *identifiers* in the statement of the rules (suggested by Dave). Also
> >>  the dire warning about rdf:value (section 3.2.4) has been made less
> >>  dire, also suggested by Dave.
> >>
> >>  Significant changes are that XMLiteral values are stated explicitly
> >>  to be distinct from character strings (defn of RDF interpretation,
> >>  section 3), and the equivalence between plain literals and xsd:string
> >>  typed literals is noted explicitly and an inference rule provided
> >>  (end of section 7.4). The wording of the Lbase translation has been
> >>  slightly altered to fit that last change also.
> >>
> >>  The change list has been rewritten and is at the end.
> >>
> >>  Pat
> >>
> >>  PS. Peter, I believe this now addresses all your concerns.
> >
> >It may be that the changes do address all my concerns, I don't have time to
> >check just now, and may not for at least a week.
> >
> >
> >However, during the quick check I just made I found some remaining
> >concerns.  The first thing I checked was the list of post-last-call
> >changes.  I noticed that several changes that result in changes to RDF(S)
> >entailments are not mentioned as substantial changes.
> >
> >The change making LV = ICEXT(I(rdfs:Literal)) means that
> >	ex:foo ex:rel "a" .
> >rdfs-entails
> >	ex:foo ex:rel _:x .
> >	_:x rdf:type rdfs:Literal .
> >whereas it does not in the last call semantics.  This was pointed out in
> >http://lists.w3.org/Archives/Public/www-rdf-comments/2003JanMar/0085.html.
> >
> >The change to datatyping in RDF makes many changes to D-entailments.  In
> >the last call semantics, there were few D-entailments, as there was no way
> >to impose connections between typed literals and datatypes.  For example,
> >	ex:foo ex:rel "1"^^xsd:decimal .
> >did not xsd-entail
> >	ex:foo ex:rel "01"^^xsd:decimal .
> >in the last call semantics because xsd-interpretations did not require that
> >I(xsd:decimal) be the xsd:decimal datatype.
> 
> I view both of these as ERRORS in the last call document which have 
> now been fixed, in both cases in close consultation with you, for 
> which I am grateful.  So I am not sure what your point is here. Are 
> you saying that you want us to revert to the situation in the LC doc?

Of course not.  However, I believe that they need to be documented as
changes that could be seen by users.

> >The change requiring non-emtpy datatypes, although it technically does not
> >affect any entailments, changes the permissable set of RDF datatypes, and
> >thus forms a significant change to the RDF datatyping design.
> 
> Do you , honestly now, think of this as a SIGNIFICANT change? Can you 
> cite any example that has been seriously proposed of a datatype with 
> empty domains? Again, I regard the omission of this condition simply 
> as an editorial oversight: frankly, I never even thought about the 
> possibility of an empty dataype until you pointed out that it was not 
> explicitly prohibited and, if ever defined, would produce strange 
> behavior in the semantics. So I prohibited it.

Again, this is a change to the specification that impacts its users.  I
believe that it thus needs to be documented as a change.  

> >Without a comprehensive list of such changes, I do not view the RDF
> >Semantics document as complete.
> >
> >
> >Problems arise in the description of other changes.  I can't
> >imagine how the significant changes to the mapping to Lbase can be listed
> >under ``The following changes do not effect [sic] the technical content.''
> 
> Perhaps that could be better phrased. I should have said 'normative 
> technical content'
> 
> >The change to lists doesn't affect any entailments that I can see.  In
> >fact, the it doesn't change anything at all with respect to the semantics,
> >even the set of RDF graphs that are the result of RDF/XML parsing.
> 
> That is true, I should have listed that in the previous section.
> 
> >The change to datatypes is not needed for compatability with OWL.  It is
> >instead needed because the last-call treatment of datatypes didn't work
> >right.
> 
> 
> It was the need to support OWL identity reasoning which brought this 
> to the fore, but OK.

I disagree even with this claim.  Before the changes to datatypes, there
were expected D-entailments that just didn't follow.  Perhaps OWL depended
more on these changes being made, but I would hope that the changes were
made to fix the problems with D-entailments, not just to support OWL
identity reasoning.

> >Peter F. Patel-Schneider
> >Bell Labs Research
> >Lucent Technologies
> >
> >
> >PS:  I also noticed a typo in 4.3 -  Rdf -> Rdfs
> 
> thanks.
> 
> OK, the version at my URI now has the change list rephrased more in 
> line with your remarks above (and the typo fixed). Textual changes 
> marked in red; some of the material has been moved to other headings 
> also. Apart from the typo, the main text is unaltered.
> 
> Pat

peter

Received on Wednesday, 30 July 2003 07:52:55 UTC