W3C WebOnt

Fourth Meeting of the W3C Web Ontology Working Group

7-8 October 2002
hosted by Hewlett Packard Company in Bristol, UK

RESOLVED: to thank the hosts Jeremy Carroll, Dave Lanfear, Jan Ward, Dave Cartwright, Mike Wonham of HP. with applause.

edited by Dan Connolly , based on notes from various scribes
$Revision: 1.30 $ of $Date: 2002/10/23 18:14:49 $ by $Author: connolly $


proposed agenda: WOWG: agenda Bristol ftf Sep 30 2002 Sep 30 2002

note: issue numbers are taken from Web Ontology Issue Status

Monday, Oct 7

Logistics, agenda amendments

Attendance: present:

  1. Stephen Buswell
  2. jeremy Carroll
  3. Dan Connolly
  4. Jonathan Dale
  5. Mike Dean
  6. Jos De Roo
  7. Pat Hayes
  8. Jim Hendler
  9. Ian Horrocks
  10. Nick Gibbins
  11. Rudiger Klein
  12. Deb McGuinness
  13. Libby Miller
  14. Enrico Motta
  15. Peter Patel-Schneider
  16. Guus Schreiber
  17. Michael Smith
  18. John Stanton
  19. Herman ter Horst
  20. Evan Wallace
  21. Chris Welty
  22. Frank van Harmelen
  23. Raphael Volz

regrets: Obrst/Mitre, Horan/Sun, Hellman/Unicorn, Thompson/Unisys, Heflin [joined in part], Stein

Brian McBride also joined for one session.

see also:

Issue 5.3: semantic layering

notes by Herman ter Horst Philips Research

The discussion centered on the three approaches to owl semantics that are described in Pat's document: weak owl / large owl / fast owl

The discussion ended with a straw poll on the question Which one is preferable: large owl or fast owl?

The discussion, chronologically:

Jim: for the entire f2f, whoever is not chairing (Jim/Guus)
is participating.

Jim: overview sheet of semantic layering
large/weak OWL
owl-dl: fast owl
owl lite

pat: the following two questions are orthogonal: 
owl-owllite   and   embedding in rdf
the issue is: whether to  restrict the language to support
efficient reasoners

ian: I do not entirely agree with pat on orthogonality;
classes as instances not orthogonal

frank: is the large owl option gone in last version?
pat: I would certainly put it there.
pat: the only argument for weak owl is the belief
that large owl is not mathematically OK
from the user point of view very little utility
in weak owl

guus/pat: weak the safest option 
could be a consensus place
frank: least constraining design
peter: name separation not in weak owl

jeremy: complete implementation?
peter: syntactic exclusions in fast owl 

guus: I see two proposals
we should make large owl not version 1
1) fast owl semantics normative semantics
2) normative weak / name separation

Dan: example entailment.
john type student
john type teacher
john type /intersection student teacher/
some proposals satisfy this, some don't 
involves comprehension principles

pat: restriction class not named;
entailments involving restriction classes not in weak owl
nothing breaks by assuming these classes exist

ian: daml+oil semantics in terms of primitives
led to difficulties
earlier attempts: dark triples
now: side conditions on models

mike: I get lost here: we know what semantics is,
but we have to do layering
pat: strict layering fast owl on rdf
the weaker versions can also do the layering
only not with efficient reasoners

dan: testcase
:x statecodede ks
:x population 10000
:y statecode kas
:y statebird lark.
can infer  :z statcode ks
population 10000
statebird lark
(not complete) this affects layering

jeremy: no decisions so far on semantics
heart of problem: owl:class rdf:type owl:class
theory of classes without foundation

pat: central question: should we think of owl as a separate language
or as an extension of rdf?

jim: summary of two options:
large vs. fast: is large doable?
large vs. fast: which is normative?

ian: technical work shows that axiomatic sem. of daml+oil
is seriously broken

frank asks pat: is large owl sem techn ok? pat: yes
ian disagrees

mike: we need clear concise, well understandable description of semantics
large more difficult than weak
pat: not true, only a small extension of the definition

chris: weak/large: people not enough accustomed
pat: much literature

dan: want typical example that shows/illustrates what we do

dan: the whole purpose of W3C-spec work is to match expectations
of users

jim: recognize dan's point. Do not ask: which research paper is better,
but does it help to achieve our requirements

jeremy: I support large owl
any current implementation would  be incomplete
comparable to situation with XSLT
fast owl is most difficult to explain- because of differences between
e.g. owl:class and rdf:class.

Jos: agree with  Pat and Jeremy: uncomfortable with owl:class etc.
we have rdfs:resource, rdfs:property, rdfs:class and we should not
add other things for essentially the same things

Raphael: question to jeremy - what practical use do you have of an 
incomplete reasoner?

Pat: I see no problem with language with incomplete reasoner, such
that a subset has complete reasoner

Ian: you can't use FOL

dan: we have to manage expectations between users and implementations

jim: straw poll
who is in favor of large owl, who is in favor of fast owl?
large: 4
fast: 7

dan: I want examples
don't talk in generalities

jim: the compelling reason for DL is that DAML+OIL used it

chris: large owl has not been on the table for a year
jim: with the document on owl semantics pat reacted to requirements from the wg
and dan's model theory

no decisions/actions at this time; discussion continued in a later session.


notes inIRC, starting 13:27:43Z

no actions/decisions at this time; discussion continued in wrap-up session

Issue 5.6: owl:imports Issue 5.14: ontology versioning

background: Heflin's message with the two proposals

notes inIRC, starting 15:05:15

PROPOSED: to include daml:imports syntax as in the reference WD, to provide a specification of the set of triples to be included in the test of semantic entailment. To have a separate class of tests for this

no decisions/actions at this time; discussion continued in an imports breakout

Tuesday, Oct 8

Semantic Layering

(continued from earlier session on semantic layering)

discussion notes inIRC log, starting at 08:51:48

Frank summarized the break-out: Consensus on semantic layering.

RESOLVED: to close the layering issue (5.3) as described in Consensus on semantic layering, provided the 2 technical bits of work can be done.

ACTION Peter Patel-Schneider/Pat Hayes: draft OWL semantics, including the "2 technical bits" [should also address issues 4.6, 5.9, 5.22]

ACTION JeremyC: produce test cases.

OWL Lite

discussion notes inIRC log, starting at 09:42:20

Volz presented, based on...

ACTION Volz: send OWL Lite in Datalog to W3C for the meeting record

ACTION Ian Horrocks: send www-webont-wg a brief description of results of rules work on DAML.

DebM presents summary of comments

ACTION Deb McGuiness: send OWL Lite issues presentation materials to W3C for the meeting record (done 16Oct)

RESOLVED: to close 5.2, endorsing existing an owl lite language subset and test class.

RESOLVED: to close 5.16 (again); to stay with minCardinality etc. as currently in WDs, with an editorial note suggesting presentation syntax names hasExactlyOne/hasAtMostOne/hasAtLeastOne/hasNone, and a link to rationale
by consensus; abstaining: Hendler, DeRoo, McGuiness, Schrieber, Stanton. Enrico concurs.

ACTION Chris Welty: write rationale re 5.16 (sent 10Oct)

RESOLVED: to close 5.15 as per the current WD, including someValuesFrom, allValuesFrom
by consensus. ABSTAINING: Volz, MikeS, Frank, Nick, Chris, Jeremy

ACTION DebM: raise hasValue issue.

ACTION Frank: collect suggestions for replacement name (for owl lite, large, and fast)

Uniform treatment of data values

discussion notes inIRC log, starting at 11:37:57Z

RESOLVED: to close 5.1 per Consensus on semantic layering, noting stateCode test works in large owl.

ACTION Mike Dean: update reference to remove constraint that inversefunctional is objectproperty.

ACTION Dan Connolly: test case

ACTION Jos DeRoo: test case

ACTION Pat Hayes: be sure upcoming semantics draft addresses 5.22

PROPOSED: re 5.20 Should OWL provide synonyms for RDF and RDFS objects? no, owl should not have synonyms; owl:Class is not a synonym. things like owl:subPropertyOf should go away. subPropertyOf: goes. Literal goes. owl:Property as a synonym goes; it may come up as a distinct concept. owl:type goes. owl:value goes. owl:subClassOf goes. owl:domain: deferred. maybe. owl:range maybe. label: goes. comment: goes. seeAlso: goes. isDefinedBy: goes.

(we came back to this proposal later.)

RDF Core Datatypes issue (break-out)

discussion notes inIRC log, starting at 13:08:09Z

ACTION Brian McBride: provide slides, summary for the record. (done 8 Oct)

Guide break-out

Changes to Guide identified in breakout session
W3C WebOnt Face to Face 4, Bristol, UK, 
Michael Smith
Oct 8, 2002

1. Identify OWL Lite vs Full OWL constructs.  
Add paragraph to explain.

Use some notation (either color or mark) to identify full owl
constructs.  Thus, un-noted constructs are both lite and full.

*FUTURE* reorganize the document so that owl lite constructs come
 first followed by full owl constructs.

2. Namespaces - fix per Jeremey Carroll suggestion.

3. Fix part whole - 

Change class / subclass for region to 
Create class Region and property located-in
Remove Region subclass hierarchy.
Replace with region instances (Europe, New Zealand, ...) thing locatedIn regions
Permits Winery at leaf of region part whole taxonomy.

 Document will not refer to this as a part-whole taxonomy.  We
 will just mention that we are encoding a notion of geographic
 containment through the use of the transitive property locatedIn

4. Make sure we have complete coverage of features.  For now, at 
least insure we have a placeholder.

5. Highlight any departures (for pedagogic reasons) in the examples from wine.owl.

6. Change identifiers to CamelCase.  Classes begin with caps.
Properties, lower case.

7. Use vin:prefix always

8. Change GRAPE to Varietal, MADE-FROM-GRAPE to madeFromVarietal

9. COLOR, TASTE, etc properties - change structure.
TASTE to hasWineDescriptor, plus merge color into this group.

 Thus hasWineDescriptor is the property with the following 4
 subproperties: hasColor, hasFlavor, hasSugar, hasColor

 hasColor has domain wine and range Color

 range WineDescriptor - a class with subclasses color, taste, flavor, sugar 
 2 - see union type below 

 Use Union type example as alternative description of WineDescriptor range.  

10. Get rid of GREEN WINE-COLOR.

11. Modify all properties to form hasProperty

12. In enumerated classes document alternative approach using unionOf.

13. complementOf: change example to consumable vs nonConsumable 

14. Eliminate ALL non-wine examples (Pending: symmetric, inverseFunctional)

15. Eliminate VINTAGE.  Add year as option feature to support datatype example.

16.  Change EveningStar / MorningStar 
     to     MikesFavoriteWine / St-Genevieve-Texas-White

17. Check syntax on whiteBurgundy.  Does it have the necessary and
sufficient conditions as we expect even though the syntax has a
conjunction of a necessary condition and an intersection?

18. Pending resolution of issue:

Indicate tradeoffs between imports vs. sameAs
combining two predefined ontologies vs. partioning.@

survey/discussion of other outstanding issues

discussion notesin IRC log, starting at 14:03:17Z

Connolly led a walk-thru of the issues list, clarifying that "POSTPONED" means that this WG does not intend to discuss the issue further, nor to address it in substance in our specs.

Chairs noted that all issues marked RAISED are now OPEN.

RESOLVED: close 2.5 closed sets by using rdf:parseType="Collection"

RESOLVED: close issue 2.6 ordered property values by noting the availability of rdf:List

Peter suggests we send our sugestion to use this approach to DAML-S

RESOLVED: to POSTPONE 4.3 structured datatypes.
by consensus; abstaining: Mike Dean, Mike Smith, Evan Wallace, Stephen Buswell

4.4 extralogical feature sets: OPEN and needs discussion

4.6 EquivalentTo - suggested that Peter and Pat look at it, issue to remain OPEN

5.1 Uniform treatment - Closed earlier

5.2 Closed Earlier

5.3 semantic layering - Closed earlier

RESOLVED: to close 5.4 Owl:quote by POSTPONING. by consensus

ACTION Jeremy Carroll: will move this issue (5.5 List syntax or semantics) forward

5.6 imports - still OPEN

ACTION Chairs: see if Ziv wishes to push this issue (5.7 Range restrictions should not be separate URIs) or Postpone

5.8 Datatypes - category error - not POSTPONED, is just waiting on RDF Core

5.9 Malformed D+O restrictions: Peter and Pat will consider as part of the semantics documents

RESOLVED to close 5.10 DAML Semantics are too weak per Consensus on semantic layering

RESOLVED: to postpone 5.12 entailing inconsistencies

ACTION Connolly: to move issue 5.13 Internet Media Type for OWL forward

noted: 5.14 versioning: Jeff Heflin is issue owner

5.15 Feature - closed earlier

ACTION Peter Patel-Schneider: to move issue 5.17 XML presentation syntax forward

ACTION Hendler: to move this one [5.18 support for unique names assumption] forward

ACTION Chairs: will ask Raphael if we want to bring this (5.19 Classes as Instances) forward

ACTION Frank: to address issue of class as instance (5.19) in OWL Lite

ACTION Jeremy: to do test case for 5.19

RESOLVED: to close 5.20 Owl synonyms by removing all the sameXXX statements. (see also: earlier discussion of 5.20)

ACTION: Mike to remove the sameXXX statements from owl reference

ACTION: Peter to generate test case to show disjointUnionOf can be rephrased w/existing constructs [and propose to close 5.21 by removing disjointUnionOf]

RESOLVED: to close 5.22 OWL:Class still needed per Consensus on semantic layering; yes, it's needed.
by consensus; Carroll abstaining.

Test document

From: Peter F. Patel-Schneider <pfps@research.bell-labs.com> To: Guus Schreiber <schreiber@swi.psy.uva.nl>, Jim Hendler <hendler@cs.umd.edu> Cc: Dan Connolly <connolly@w3.org> Subject: minutes for test session Date: 08 Oct 2002 12:12:19 -0400
                Scribe Log of Test Session

Jeremy Carroll presented his test document.

Discussion on a particular (contentious) test case E.2.3 001, namely one
of entailing ranges of properties.  This entailment currently does not hold
in RDFS, the question is whether it should hold in OWL.

Pat Hayes: compatability with RDFS is important
Ian Horrocks: range should have an extensional meaning
Pat Hayes: range should only *imply* the extensional meaning, 
           the extensional meaning should not be the definition of range
Dan Connolly: what is the situation in Fast OWL
Ian Horrocks: all are extensional
Pat Hayes: domain and range are stronger than FunctionalProperty
Peter Patel-Schneider: the opposite can be stated
Pat Hayes: no, there is something different with domain and range
           namely, that users have different intuitions
Jeremy Carroll: all these should be treated the same
Dan Connolly: they all look similar, so we should do the right thing
Jeremy Carroll: Fast OWL currently has them all extensional
Peter Patel-Schneider: but they can be easily changed
Pat Hayes: Why is there linking?
Dan Connolly: Why is there not linking?
Pat Hayes: domain and range may have technical reasons for being intensional
Ian Horrocks: it seems reasonable to make them all the same

Straw Poll:  Should test case E.2.3 001 hold?
Result: none in favour, some opposed, majority abstain, 
        calls for more information

Discussion on the form of RDF in test cases.

Jeremy Carroll: Presentation of the test document:
- There is a test document, and it has had some comments.
- The test document contains a number of approved and proposed tests.
- The document discusses its normative status.
  - The tests are dependant on the formal semantics.
- The document contains a number of stylistic preferences
  - striped syntax
  - some features of RDF/XML are not used
  - absolute URI references

Question: strike restrictions on rdf:ID and relative URI references
Straw Poll: several in favour, one against

Jeremy Carroll: one objection is that description of the test cases include
        descriptions that may be taken as normative
Dan Connolly: the prefered end result is to point to a normative
        description, or none

Jeremy Carroll: how to handle tests after rec
Dan Connolly: too premature to discuss this
ACTION: Jeremy Carroll to add an editorial comment to the document on
        errata process.

Jeremy Carroll: what is the conformance status of this document
                currently there is non-normative wording on this
Dan Connolly: there will be pressure to do something here
Jeremy Carroll: should things like OWL Lite tests be flagged

ACTION: Jeremy Carroll to add editorial comments like:
        OWL Lite tests may be flagged.
        Imports tests may be flagged.
        Conformance is not yet addressed.

Ian Horrocks:  What does the paragraph indicating that OWL reasoners should
               prove many of these tests?
Pat Hayes: Change to complete OWL reasoners should prove all of them.
Jeremy Carroll:  Will do.

ACTION: Dan Connolly will fix 403 errors arising from the document.

PROPOSAL: To publish the test document with outstanding actions performed.
          Expected date for publication is end of next week.

Imports Breakout

continued from discussion previous day; the group attempted to reconstruct the proposal from the previous day, to which Mike Dean had been the only dissent in a straw poll. During the breakout, Mike agreed to change his vote to an abstention for the sake of moving forward. However, after the breakout session, Welty et. al. presented their notes to the group and found that it did not reflect their views at all.

notes from Christopher Welty:

Imports breakout

Present:  Mike {Dean, Smith}, Jim H, Steve S, Evan W., Chris W., Jeff H. (phone)

Jim stated the basic position :


Socrates a URL2:Greek:Man.



Man subclass Mortal


<import URL2>

Socrates a Greek:Man

The semantics (operational or otherwise) does notrequire, but also doesnotprevent, that document URL1 entails:

Socrates a Greek:Mortal.

However, the semantics do requirethat the import closure of document URL3 entails this.

This represents a big compromise from two perspectives:

Mike Dean feels this compromise requires content providers to do too much work, that is, explicitly typing in the imports statements for the ontologies they want.  However, he also presented a new view, that there be some standard policy that specifies how someontologies should be automatically included when their symbols are referenced.

Mike's second choice would be to require that loading URL1 produce no entailment.

Mike submitted eventually and agreed to abstain.


If imports is present then we use what Jeff had in proposal 2 (syntax as in DAML+OIL, semantics are such that if graph A imports B, then  A|=C if and only if A+B|=C), without imports a reference to an external symbol has no required entailments.

recap of issue resolutions - planning working drafts / last call

discussion notes inIRC log, starting at 16:17:52Z

Note well: W3C publishing moratorum 18 - 29 November 2002. (announcement)

recap of Guide break-out

RESOLVED: to publish the guide, subject to editorial discretion and review by Welty; Jeremy to review the RDF/XML fiddly bits.

recap of test break-out...

RESOLVED: To publish the test document with outstanding actions performed.
Expected date for publication is end of next week.
by consensus; Hayes concurring.

... discussion of 5.6 imports was inconclusive.

RESOLVED: to publish, in 2nd half of Oct, the feature synopsis and reference.

ACTION MikeD catch ref up with issues, publish reference.

ACTION Deb M: publish synopsis

... discussion of Jan ftf, possibility of using it as an interop...

DanC: note that in order to do a ftf in January, I'll have probably have to propose to the Advisory to Continue; be prepared to get in touch with your AC rep to endorse continuing this work.