Re: Conformance was Re: apologies and TEST update

At 23:04 +0100 1/6/03, Jeremy Carroll wrote:
>I think your message was a good argument for *having* conformance statements.
>Pretty much I read you as saying working out whether a document is in OWL DL
>or not is hard work, and your team does not want to implement it. If many OWL
>systems take that attitude then they won't interoperate with OWL DL reasoners
>and we have wasted some of our time.

jeremy, you missed my point completely.  I'm simply saying that for a 
non-reasoner tool to be Owl ?X compliant, it doesn't seem to me that 
THAT TOOL has to be responsible for determining whether the document 
is in the ?X class.

For example, we have a crawler that will wander through documents and 
grab certain stuff and build a new document,  the document it creates 
is in OWL Lite.
(that tool doesn't count as Owl Lite compliant because it cannot read 
OWL Lite documents, but I could live with that as odd as it seems to 
   It hands that document to another tool, which wants to claim Owl 
Lite compliance (for example we'll probably try to claim RIC is at 
least Owl Lite compliant).  That document can read all OWL Lite 
documents, can "appropriately process" (whatever that means) what it 
finds, and can support name separation (and more) - which our tool 
  Thus, I'm not arguing for a second definition of Lite, or for 
anything other than saying that I don't think Owl Lite conformance 
depends on the tools recognizing what is NOT in Owl lite as you wrote 
the requirement.
>My understanding of WG politics is that there are three or four camps e.g.:
>    a DL camp
>    an RDF camp
>    some practical ontologists
>    the scruffies
>the language we have come up with *must* have enough to satisfy these camps,
>and *should* take the best each camp has to offer.

I agree

>The way the layering works, with the resolution of the semantic layering
>issue, is to give the DL camp what they wanted and the RDF camp what they
>wanted, but at different levels. For this to be an advantage for the user,
>rather than just a pain in the neck, we need to be sure that moving between
>the levels is straightforward. Unless all OWL tools can tell you whether your
>current file is in OWL Lite or not, and whether it is in OWL DL or not, then
>this simply will not be the case.

I don't agree with you, and none of the non-reasoner tools we build 
(all of which would likely want to be called OWL Compliant at either 
Lite, DL or Full depending on tool) can do this.  What they can do is 
properly handle a document that someone else (human or automatic) 
says is an appropriate document.  So I don't see why every tool has 
to do this redundant work

>We could alternatively reopen 5.3; either in the limited way that I have
>suggested*, trying to weaken the syntactic constraints on OWL Lite and OWL DL
>further than Peter has yet conceded; or, in a more wholesale way, saying that
>we fundamentally don't buy this story.
>As an editor, particular a test editor, I see my main role as keeping the WG
>honest. Either we have various levels, and we have both an abstract syntax
>and an RDF graph, and we expect tools to implement these levels and users to
>use them; or we don't.. The conformance statements are there partly to test
>our commitment to what the other documents are saying.

again, I disagree not at all with your goals, but with your specific 
implementation of them

>I am particular suspicious of my natural allies in the WG who seem to be
>taking the attitude of, well OWL Full is the level my tool will work at, and
>I will do my version of OWL Lite that is just a restricted vocabulary, and
>Peter and Ian can play over there in the OWL DL playpen.

not sure I understand, but I wouldn't expect tools to be called OWL 
Lite compliant if they do inappropriate things given an Owl Lite 
document.  Given an OWL Full document, I don't see why I should 
expect them to do appropriate things at all - be nice if they barf on 
it in some sense (error messages would be appropriate), but I don't 
see that they should have to do that to prove compliance.

>In line comments:
>Jim Hendler wrote:
>>>   This section uses the language of [RFC 2117]. (sic)
>>>A language tool without a reasoning capability MAY claim OWL Lite
>>>if it can:
>>>+ Accept all OWL Lite constructs.
>>that I can live with
>>>+ Distinguish RDF/XML documents that conform with OWL Lite from those that
>>>not. Specifically, it MUST identify those RDF/XML documents that can be
>>>formed by applying the mapping from the abstract syntax for OWL Lite as
>>>defined in [OWL Abstract Syntax and Semantics].
>>this doesn't make sense to me and I would oppose it. Why should an
>>OWL Lite tool think of OWL Full as anything other than an arbitrary
>>RDF document?  If I take my tool and feed it something in OWL DL or
>>Full, it will likely have some heuristics to notice and reject things
>>not in the  sublanguage (i.e. our current tools could recognize that
>>owl:union being applied to something is not in Lite) but we treat
>>things from OWL Full as if they're RDF.  THat is, if someone says
>>:foo a owl:class;
>>    rdfs:comment "This is a owl class";
>>    dc:creator "hendler";
>>    :nodeID "37a".
>  >
>>   :nodeID a rdfs:property.
>>then we have put a property on a class (violating class as instance),
>>but recognizing which is legal between rdf:comment, dc:creator, and
>>:nodeID isn't easy (in fact, I'm not actually sure - does putting an
>>rdfs:comment on a owl:class violate class as instance? 
>I don't think so, there are these things called annotations in the abstract
>syntax which seem to offer limited support for such things.
>As far as I understand, literal valued properties and URI valued properties
>are permitted, but the URIs used are not used elsewhere in the ontology... (I
>might not have that right).
>>  What about a
>>dc:creator?) - if this was bar:nodeID the checkign gets even more
>The checking *is* ugly, if you don't like it we should reopen 5.3.
>The checking is not impossible, or even difficult, at least not by a
>computational complexity measure. (I suspect that in practice it is linear).

no, we shouldn't open the issue, we should change your definition of 
compliance to mean the same thing it means in other languages and 

>>In short, I don't see how I can be a tool without a "reasoning
>>capability" if I have to do reasoning to determine whether a document
>>fits the restrictions, and while I think the notion that mapping the
>>abstract syntax should be good enough, I'm not convinced it is,
>>because of the potential reasoning chains that can cause the problems
>>(i.e. X subclass Y subclass Z chains).
>Any computer program must do some reasoning!!!
>A reasoning component is something that can do reasoning in accordance with
>the OWL semantics ...
>>>+ Provide support for name separation corresponding to the separation of the
>>>domain of discourse.
>>I'm assuming by this you mean that if I try to create something of
>>one type, when it's been declared as another, the system should
>>produce an error.  If so, I could live with it.  Of course, testing
>>it exhaustively will be arbitrarily hard as there are some cases
>>(like the ones above) where determining the set of all legal types
>>may be hard.  If we mean assuming types are separated, this maintains
>>that, then I can live with it.  If it means "can determine if types
>>are separated, then this is a subcase of the 2nd restriction, which I
>I am not quite sure I get the difference you are pointing at.
>In as much as I do, I guess I am in the second case.
>- an OWL tool needs to be able to warn you when you use a name as both a
>datatypeproperty and an individual and hence stray into OWL Full; and if it
>is fed a document that has such a URI it needs to be able to detect that it
>is not an OWL DL document (which may or may not involve detecting this
>particular problem).

yes, that is what I meant too - i.e. if my system reads something is 
a datatype, and you try to drag and drop it into the filler of an 
object type property, my system will issue and error message.  But, 
this is assuming it was given a document in which the types were 
correctly separated in the first place.  If not, I don't know what my 
tool should do (nor should I - it was handed an inappropriate 

>  >>A language tool without a reasoning capability MAY claim OWL DL conformance
>>>it is OWL Lite conformant and it can:
>>>+ Accept all OWL constructs.
>>>+ Distinguish RDF/XML documents that conform with OWL DL from those that do
>>>not. Specifically, it MUST identify those RDF/XML documents that can be
>>>formed by applying the mapping from the abstract syntax for OWL DL.
>>sam problem as 2 above
>>>Language tools without a reasoning capability MAY claim OWL Full conformance
>>>if they are OWL DL conformant and they allow name separation support to be
>>>switched off. Specifically, the following possibilities SHOULD be supported.
>>>+ Classes, datatypes and properties SHOULD be useable as instance data.
>>>+ Classes and datatypes SHOULD be useable as properties.
>>>+ Data valued properties SHOULD be useable as individual valued properties.
>>>+ Subclasses and subproperties of classes and properties in the RDF, RDFS
>>>OWL namespaces SHOULD be allowed.
>>this one I'm more okay with as it means an OWL Full tool cannot have
>>certain restriciton.  Of course, this assumes we've changed the
>>previous definition.  If not, I could live with deleting both the OWL
>>DL and OWL Lite restrictions, and using this by saying ... Accepts
>>both RDF and RDFS graphs as defined in <cites> and the following
>>properties SHOULD be supported:
>>>   Reasoning components MAY claim complete OWL DL conformance [or complete
>>>Lite conformance] if they provide complete reasoning over OWL DL [or OWL
>  >>Lite]. i.e. A conformant complete OWL DL [Lite] reasoner MUST find proofs
>>>all OWL DL [Lite] inferences. A conformant complete OWL DL [Lite] reasoner
>>>MAY find proofs for any OWL Full inference. A system which claims complete
>>>OWL DL conformance MUST also be OWL DL conformant.A system which claims
>>>complete OWL Lite conformance MUST also be OWL Lite conformant.
>>I won't comment since I don't intend to build reasoning components
>>per se, and will let those who do decide this (modulo the comments
>>above w/respect to conformance)
>See my comments to shimuzu on this  para. A rephrasing for clarity.
>>>A document MAY be described as an OWL Lite document if:
>>>+ It does not use any constructs not in OWL Lite.
>>>+ It corresponds to some document in the abstract syntax for OWL Lite using
>>>the mapping defined in [OWL Abstract Syntax and Semantics].
>>>A document MAY be described as an OWL DL document if:
>>>+ It does not use any constructs not in OWL DL.
>>>+ It corresponds to some document in the abstract syntax for OWL DL using
>>>mapping defined in [OWL Abstract Syntax and Semantics].
>>these I actually like - I do think we can define conformant
>>documents.  However, there are some wordings that need to be more
>>careful - for example, is rdfs:comment "in OWL Lite"?  i.e. OWL Lite
>>as defined in...
>That's not *my* problem. It needs to be clear where OWL Lite is defined, and I
>will link to that. Ditto OWL DL.
>>also, why not for OWL Full?  We could specify this is any document
>>that uses RDF and RDFS, whcih would help address the question we're
>>often asked as to relation between these...
>We finally agree on something. I thought that was a waste of breath, but if
>the WG wants it I certainly would not oppose.

Professor James Hendler
Director, Semantic Web and Agent Technologies	  301-405-2696
Maryland Information and Network Dynamics Lab.	  301-405-6707 (Fax)
Univ of Maryland, College Park, MD 20742	  240-731-3822 (Cell)

Received on Monday, 6 January 2003 19:08:44 UTC