Re: Review of RDF Based semantics [2nd]

Michael,

That is quick!
>> A few minor comments as follows.
>> - Section 2.1 talks about generalized RDF triples. I am wondering
>> whether it is useful
>>   to restrict the use of literal values as predicates.
>>     
>
> The text currently states that the RDF-Based Semantics *MAY* be applied 
> to graphs with generalized RDF triples, and generalized RDF triples
> may contain literals in predicate position. This does not mean that
> implementations must support such extended RDF graphs, but they will 
> not lead to problems *if* implementations support them.
>
> Technically, there is no problem with having literals in predicate 
> position, since they denote individuals under the RDF-Based Semantics, 
> and, as it is true for all individuals, these data individuals can 
> be properties as well. 
>
> Whether having literals in predicate position is useful or not 
> will certainly depend on the respective application, but such 
> questions are outside the scope of this specification document.
>
> However, the strong relaxation may become relevant to other standards. 
> For example, the same notion of "generalized RDF" is assumed by the 
> following important work-in-progress specifications that have a 
> dependency on the OWL 2 RDF-Based Semantics:
>  
>  * OWL 2 RL (see 2nd paragraph of §4.3 in the Profiles Editor's Draft at
> [1a])
>  * RIF+OWL (see Section 3.1.1 of the current Editor's Draft at [1b])
>
> So I do not intend to do a change.
>
>   
I think I am OK with it. It's just that it is not intuitive to 
understand an assertion like "x 3.1415 y"
even though there is a well defined "meaning" for it.
>   
>> - Section 4.2, I am not exactly clear about 'otherwise IL("s"^^u) is not
>> in LV'
>>  Does it mean that "2.01"^^xsd:integer gets treated like an IRI in OWL
>> 2 FULL?
>>     
>
> If you mean by "treated like an IRI" that the denotation of the 
> literal "2.01"^^xsd:integer is an instance of IR \ LV, 
> then you are actually right.
>
> This is in accordance with Section 5.1 of the RDF Semantics spec [2a]
> (see the third semantic condition there). In addition, 
> Section 5.2 of the original OWL 1 RDF-Compatible Semantics [2b]
> states that "L_I is a mapping from typed literals in V 
> to their denotations in R_I." (rather than LV_I). I have also
> clarified in personal conversations with former members of the 
> RDF Core WG that this is /not/ a bug.
>
> The cited text in the RDF-Based Semantics document is only a summary
> of those parts in the RDF Semantics specification that are actually
> relevant for the definition of an OWL 2 Full interpretation (and
> certain other aspects of the RDF-Based Semantics). So I consider
> a change of these parts of the RDF Semantics to be out of scope 
> for OWL and rather in the responsibility of a future 
> RDF working group.
>
> So I do not intend to do a change.

OK. Actually, that's what I meant.

>> - Section 5, in the paragraph starts with Unscoped variables. "x"
>> denotes an arbitrary individual in the universe.
>>  I am wondering if "element" is a better word than "individual" in this
>> context (mainly
>>  to avoid confusion with owl individual)
>>     
>
> The term "individual" is just what is meant here. The text sais:
>
> [[
> Unscoped variables: If no scope is explicitly given for a variable "x",
> [...]
> then x is unconstrained, which means that x in IR, 
> i.e. "x" denotes an arbitrary individual in the universe
> ]]
>
> The idea behind this convention is that if nothing specific is said about
> a variable x, then x stands for an individual. The reason is that all
> elements of the domain (IR) of an OWL 2 Full interpretation are individuals.
>
> So I do not intend to change this.
>
>   
Hmmm. When an owl user talks about individual, he/she probably means 
those x in his/her ontology
such that x type owl:Thing holds. Here IR is about the interpreted 
domain and we call elements of IR individuals.
Isn't it a bit confusing?

>> - In Section 5, does it make sense to add a semantic condition saying
>> that literal values
>>  cannot be used to denote a class? It is hard for me to see the meaning
>> of "x rdf:type 3.1415"
>>     
>
> This is similar to the discussion about literals in predicate position
> above.
> A literal is just a name for a certain data value, which is an individual
> under the RDF-Based Semantics. And every individual can have a class
> extension,
> hence can be a class. 
>
> Further, this is not really a discussion related to specific aspects of 
> the OWL 2 RDF-Based Semantics, but already holds for the semantics of RDFS, 
> and should therefore be treated by an RDF working group. Note, however, that
> what you want to disallow is currently perfectly allowed (consistent) in
> RDFS.
>
> Apart from this, I would consider this a "big change". First, the semantic 
> condition would be of a form that is different from all other semantic 
> conditions in OWL 2 Full (and OWL 1 Full), because it would have to talk 
> about vocabulary elements instead of domain elements. Second, I would expect
> that such a semantic condition may have non-obvious side effects on the
> semantics, and I would need to very carefully check. Given that we plan to 
> vote the document into LC *today*, there is no time anymore for such deep 
> analysis.
>
> So I am not going to do a change.
>
>   

OK. I am fine with it. Again, I think it is not intuitive (or less 
intuitive compared to "x owl:sameAs 3.1415" :))

>> - In Section 6, second last paragraph, "IEXT(I(owl:topObjectProperty)) =
>> IR x IR"...
>>  for which there are no corresponding domain and range axiomatic
>> triples.
>>  Why?
>>     
>
> This is a reasonable question! The text was even confusing, as I see now. 
> I have changed the text, please check!
>
> DIFF:
> <http://www.w3.org/2007/OWL/wiki/index.php?title=RDF-Based_Semantics&diff=21
> 611&oldid=21279>
>
>   

I don't quite understand it. The same kind of argument applies to the 
pair of owl:complementOf and owl:disjointWith, right?

>> - Section 7 gives a very interesting example on DL entailment is not
>> Full entailment and how to
>>   fix it through syntactic changes. It is very useful without a doubt.
>>
>>  I am wondering that in addition, can we describe, at a high level,
>> what can be meaningfully modeled/expressed
>>  using OWL 2 FULL but not with OWL 2 DL. After all, users of OWL 2 FULL
>> care more about
>>  using the additional expressivity than aligning inference with DL.
>>     
>
> Section 7 is the successor to that part of the original OWL 1 RDF-Compatible
>
> Semantics document, which covers the correspondence theorem [3a] and its 
> proof [3b]. Its purpose is to show that the two semantics of OWL 2 are 
> strongly aligned, as this was an overall design goal of both OWL 1 
> and OWL 2. What "strongly aligned" means is expressed by the actual 
> correspondence theorem in Section 7.2. The elaborate example in Section 7.1 
> shows that the two semantics are not identical, but motivates the actual 
> form of the correspondence theorem and its proof in Section 7.3. 
> To summarize, there is some justification that the content of Section 7 
> exists in our OWL 2 specification, regardless of whether OWL 2 Full 
> users are concerned about alignment with OWL 2 DL or not.
>
> On the other hand, in my eyes, questions concerning meaningful 
> OWL 2 Full modeling should be treated by research papers and text 
> books, and not by the formal specification of the language itself. And 
> it is far from clear to me how such a text should look like. And 
> given that we vote on LC publication today, it will not be 
> possible anymore to create such a text.
>
> So I am not intending to do a change.
>
> [3a] <http://www.w3.org/TR/owl-semantics/rdfs.html#5.4>
> [3b] <http://www.w3.org/TR/owl-semantics/proofs.html>
>
>   

I know this is a bit too much to ask. I tried :)  We need a nf & 
rationale for OWL 2 FULL :)

>> - The Proof of the Balancing Lemma. It seems that the algorithm
>> described should terminate. Can we
>>   state it explicitly?
>>     
>
> Ok. Can you please have a look whether my changes are to your satisfaction?
>
> DIFF:
> <http://www.w3.org/2007/OWL/wiki/index.php?title=RDF-Based_Semantics&diff=21
> 618&oldid=21611>
>
>   

Looks good. Thank you.

>> - Table 8.2, if C in IC, then exists z in IR s.t. <z, c> in
>> IEXT(I(owl:complementOf))
>>  Should it be "z in IC" instead? I am asking because it seems to me
>> that this "z" can participate
>>  other comprehension conditions including owl:intersectionOf.
>>     
>
> No, "z in IR" is intentional and sufficient. That the corresponding 
> individual is a class follows from Table 5.2, by the entry
> about "owl:complementOf, after the comprehension condition has "fired".
>
>   
I see it now. Thanks.

Cheers,

Zhe

Received on Wednesday, 8 April 2009 19:10:28 UTC