Re: LANG: first sketch

Frank, Ian

some input on teh first sketch of teh language.  Sorry for the late 
reply but I have been away from my email for a while.

The document says

>1. OWL should have a frame-style part (for easy idiomatic usage) as 
>well as a DAML+OIL/DL-style axiomatic part (for high expressiveness)
>2. The frame-part would consist of (roughly):
>… Datatypes, cardinalities, oneOf, conjunction
>… Inverse, transitivity
>… disjointness axioms, (in)equalities of names
>1. The axiomatic part would be close to DAML+OIL expressiveness and 
>include (again roughly):
>… quantifications (hasClass, hasValue, toClass), negation, 
>disjunction, qualified cardinalities
>… general subsumption axioms with complex expressions on both side 
>of the subclassOf symbol.


I agree that distinguishing between a frame part and an axiomatic 
part is a good way to structure teh language.  However, both RDF and 
DAML+OIL distinguish between a frame (resource) specification  part 
and a property specification part, so I wonder why not to maintain 
it. In other words, we could have three parts to the language

Frame Model
Property Model
Axiom model


I am also not sure about the rationale of the proposed split between 
the axiomatic part and teh frame part.  In a sense this may not 
matter, if we consider OWL in its totality.  However, you can imagine 
a lightweight use of OWL which only relies on basic frame and 
property descriptions, and therefore we may want to get the 
modularization correct

Most frame languages I know have the following constructs

Classes
Instances
Template Slots with values, range, & cardinality restrictions

Indeed it is difficult to talk about frame languages unless they 
include these aspects.  In your model, ranges are part of the 
axiomatic model,  while cardinalities are in the frame model, which 
is a bit confusing. My suggestion would be to simply take the set of 
property restrictions in daml+oil and add them to the frame model. 
Moreover, the frame part in your model includes  Inverse, and 
transitivity,  which in DAML are property elements and are not 
normally to be found in frame models.

So,  I  would reformulate the strawman proposal as follows;

<<<
Frame Spec
Classes
subclass-of links
Property restrictions (teh whole daml+oil set could simply be reused)
datatypes
enumeration (class same as enumeration set)


Property Spec
Again, here we could simply reuse what already exists in daml+oil and 
simply add a transitivity contruct - I don't see anything 
particularly hard or controversial in this part of daml+oil except 
that I would remove the redundancy between between equivalentTo and 
samePropertyas and just have one construct


Axiom Spec
disjointness axioms, (in)equalities of names, negation, disjunction, 
subsumption, etc..
>>>

The above is essentially a reformulation of daml+oil according to the 
80/20 rule.  On this initial basis, as you suggest, we could add 
other idioms, on the basis of experience and requirements. Indeed the 
80% bit (frames and properties) seems to me to be non-controversial 
and if we can quickly agree on it, then we can concentrate on the 20% 
bit, which is far less straightforward.

Enrico


















>
>SUMMARY
>1. OWL should have a frame-style part (for easy idiomatic usage) as 
>well as a DAML+OIL/DL-style axiomatic part (for high expressiveness)


Yes, agree with this distinction as a way to design the language.

>2. The frame-part would consist of (roughly):
>… Datatypes, cardinalities, oneOf, conjunction
>… Inverse, transitivity
>… disjointness axioms, (in)equalities of names


Frame part shoudl have the following:

Classes
Instances
Template Slots with range, cardinality

>1. The axiomatic part would be close to DAML+OIL expressiveness and 
>include (again roughly):
>… quantifications (hasClass, hasValue, toClass), negation, 
>disjunction, qualified cardinalities
>… general subsumption axioms with complex expressions on both side 
>of the subclassOf symbol.
>
>LANGUAGE FEATURES, NOT SYNTAX:
>The focus group should discuss first which 
>features/expressiveness/constructs OWL should have. This discussion 
>should be separate from what syntactic form each of these should 
>have, or even the specific syntactic form of the language as a 
>whole. Syntax can be designed last, not first. Actually, we expect 
>that a single syntax won't do. We will need at least two: one 
>RDF-based syntax (for machines) and one presentation syntax readable 
>for humans (taking lesson from OIL), and probably also an XML-DTD 
>based syntax.
>DISTINGUISH LANGUAGE CONSTRUCTIONS FROM THE UNDERLYING LOGIC.
>The language constructions should provide good support for idiomatic 
>expressions that occur often in practice; the underlying logic 
>consists of (typically much smaller) atomic constructions with nice 
>mathematical properties. Of course, the idioms should be 
>translatable into the logic.
>Examples of such idiom are:
>… a frame with slots
>… slots that are optional or required
>… slots that are multivalued or singlevalued
>… local range restriction on slots
>… unique properties, uniquely-identifying properties ("unambiguousProperty")
>… etc
>Translation of these idioms into the logic might result in e.g.
>… a non-optional multivalued slot with a specified range corresponds 
>to hasClass + toClass
>… a non-optional singlevalued slot with a specified range 
>corresponds to hasClass + cardinality=1,
>… etc.
>OWL should have a stronger idiomatic vocabulary than DAML+OIL (which 
>already has some, but not much)
>LOW THRESHOLD, HIGH CEILING
>On the one hand, we want to make a language that is easy to use for 
>most, on the other hand we don't want to limit the sophistication of 
>some. This could be achieved by a better separation of things that 
>are easy from things that are hard.
>Note: easy/hard refers to "easy to understand by our target group", 
>perhaps also to "easy to implement in a variety of tools" (editors, 
>storage, browser/visualisors), and least of all to "low complexity 
>class for a reasoner") In fact it is quite possible that the "easy" 
>part of the language might formally have the same expressive power 
>and complexity as the full language.
>A first stab:
>The following are considered "easy":
>… Datatypes, cardinalities, oneOf, conjunction
>… Inverse, transitivity
>… disjointness axioms, (in)equalities of names
>The following are considered "hard":
>… quantifications (hasClass, hasValue, toClass), negation, 
>disjunction, qualified cardinalities
>… general subsumption axioms with complex expressions on both side 
>of the subclassOf symbol
>
>FRAMES VS. AXIOMS
>One way of obtaining "low threshold, high ceiling" is to have two 
>parts to the language: a frame-based part, and an axiom-based part 
>(for want of a better term).
>… In the frame-based part, expressions are limited to the idiom and 
>the easy constructions, e.g., no quantification (besides of course 
>implicitly through idiomatic expressions), no boolean connectives 
>(besides implicitly through idiomatic expressions)
>… In the axiomatic part, people would be free to use DL-style axioms 
>with arbitrary class-expressions on either side of a subsumption 
>symbol. The expressiveness of the frame-style part of the language 
>may be less than that of the axiomatic part, but probably not much 
>less.
>
>RELATION TO DAML+OIL
>The above proposal takes DAML+OIL as its basis (as per our charter). 
>The expressiveness of OWL will be close (if not: equal) to DAML+OIL. 
>The aove proposal extends DAML+OIL in order to provide a lower 
>step-in threshold (which is perceived to be too high for DAML+OIL). 
>Another extension is to provide a human-readable presentation syntax 
>besides an RDF/XML-based syntax.
>NEXT STEPS TO TAKE:
>… collect idiom for inclusion in the language
>This should be based on
>… experience with DAML+OIL
>… our use-cases
>… experience with earlier ontology/KR-languages
>… deal with other requirements/objectives
>The above focusses mostly on the ontological expressiveness of the 
>language, and not really at requirements like tagging, importing 
>ontologies, etc. These should (of course) be dealt with. (but hey, 
>this was only a first step:-)
>



>From the agenda for the 7 March teleconf:
>
>>2) LANG: OWL lite (van Harmelen, 30 min)
>>
>>  - proposal will be posted by Frank either Wed late or Thu
>>    morning, European time.
>
>
>This document is now available at:
>http://www.cs.vu.nl/~frankh/spool/OWL-first-sketch.html
>
>Frank.
>    ----
>
>Frank.van.Harmelen@cs.vu.nl              http://www.cs.vu.nl/~frankh
>Department of AI, Faculty of Sciences,  Vrije Universiteit Amsterdam
>de Boelelaan 1081a, 1081HV Amsterdam, The Netherlands
>tel (+31)-20-444 7731 fax (+31)-84-221 4294