W3C Web Ontology Working Group IRC Chat Logs for 2003-03-04

15:51:19 Users on #webont: logger pfps @mdean

15:51:32 <mdean> pfps: this dives into the trees, and may be ignoring any nearby forest

15:51:42 <mdean> pfps: different philosophies between Peter and Jeremy

15:51:59 <mdean> Thanks to Dave Beckett for starting the logger

15:52:03 <DaveB> hi there

15:52:06 <DaveB> logger, ptr?

15:52:07 <logger> I'm logging. I found 2 answers for 'ptr'

15:52:08 <logger> 0) 2003-03-04 15:52:06 <DaveB> logger, ptr?

15:52:09 <logger> 1) 2002-04-09 08:18:21 <JosD> i.e. constrain interptretaions / mentions like D + O syntax

15:52:14 <DaveB> ha ha

15:52:20 <DaveB> logger, pointer?

15:52:20 <DaveB> See http://ilrt.org/discovery/chatlogs/webont/2003-03-04#T15-52-20

15:52:29 <mdean> wonderful!

15:52:29 <DaveB> ok, have fun

15:53:15 <mdean> Jeremy: concrete syntax needs to be tractable (reasonably easy for a developer or power user to be able to read and write and understand constraints)

15:53:40 <mdean> pfps: more important to consider what you're trying to say than the triples that you're using to say it

15:54:17 <mdean> Ian: fundamental difference is to get abstract syntax right and then make RDF mapping as nice as it can be - Jeremy more focused on triples

15:54:58 <mdean> Jeremy: suggests making the mapping rules "ugly" to keep the other things pretty

15:55:05 <mdean> Guus: also prefers changes to mapping rules

15:56:10 <mdean> Ian: intermediate consideration: implementors perspective (e.g. turning OilEd into OwlEd) - effort primarily involves mapping rules

15:56:48 <mdean> Jeremy: not all abstract syntax trees map to concrete syntax, e.g. using same URI for class and property

15:57:11 <mdean> Jeremy: Restriction with no constraints also not mapped

15:57:16 <mdean> pfps: latter is a bug

15:57:49 <mdean> Jeremy: could decide that abstract syntax is clean way to think about things, and not map difficult constructs

15:58:10 <mdean> Jeremy: RDF graph is in OWL DL if it corresponds to abstract syntax under the mapping rules

15:58:29 <mdean> PatH: this means abstract syntax is no longer the syntax of OWL DL

16:00:16 <mdean> Attendees: Mike Smith, Jeremy Carroll, Jonathan Dale, Harold Boly, Peter Patel-Schneider, Pat Hayes, Raphael Volz, Guus Schreiber, Ian Horrocks, Mike Dean

16:00:42 <mdean> pfps: if you turn the crank now, you get OWL Full - motivates allowing of duplicate names

16:01:16 <mdean> Ian: no implementation is going to check everything, just do a heuristic mapping

16:01:29 <mdean> Jeremy: sounds like recipe for lack of interoperability

16:02:30 <mdean> Jeremy: if we make it sufficiently difficult, then heuristics are required - we don't want to encourage their use

16:03:36 <mdean> Ian: only well-formed if they could be the result of the mapping from the abstract syntax

16:04:30 <mdean> pfps: alternatively, pass long list of constraints (Jeremy or Peter's "5 ugly pages")

16:04:58 <mdean> pfps: believes that 2 pass linear algorithm will implement Peter's 5 ugly pages

16:05:18 <mdean> Jeremy: believes peter's list misses too many cases, e.g. n-way ...

16:05:48 <mdean> Guus: still wants to go through list, keeping in mind goals

16:05:56 <mdean> Guus: prefer not to make changes to abstract syntax

16:06:26 <mdean> Jeremy reordered items in his response

16:08:13 <mdean> discussion regarding unionOf ...

16:08:47 <mdean> Guus: changes to mapping rules don't affect abstract syntax?

16:09:14 <mdean> Jeremy: interaction related to annotation declarations

16:09:48 <mdean> Guus: try to maximize expectations for concrete and abstract syntax

16:10:12 <mdean> Jeremy: main concern is that rules are unclear

16:10:38 <mdean> Jeremy: still worried about entailment - example on board

16:12:10 <mdean> Ontology(Annotation(prop "string")) can map to O prop "string"

16:12:25 <mdean> but also O type Ontology

16:12:28 <mdean> O prop string

16:12:35 <mdean> prop type AnotationProperty

16:12:51 <mdean> means O prop "string" entails latter

16:13:40 <mdean> pfps: rule has got to be that it's necessary or forbidden, or cheat by making everything an AnnotationProperty

16:13:50 <mdean> pfps: ... and Ontology

16:13:53 <mdean> Jeremy: very ugly

16:14:16 <mdean> Ian: why is second interpretation bad?

16:14:32 <mdean> Peter: bad because of RDF manifesto that all triples have meaning

16:14:46 <mdean> Peter: semantic import in this syntactic encoding

16:14:53 <mdean> Peter: more of this on Thursday ...

16:15:04 <mdean> Jeremy: not taking RDF position, just want to make it work

16:15:12 <mdean> Peter: optionally typing a syntactic category is a bad idea

16:15:23 <mdean> Ian: why is mandatory typing a problem?

16:15:42 <mdean> Jeremy: having to do it 100% of the time is better than in 80% of the cases

16:15:55 <mdean> pfps: need declarations to type them without using them

16:16:32 <mdean> jjc: subissue involving URI instead of string - does this also need to be typed

16:16:44 <mdean> Jeremy = jjc, easier to type

16:17:50 <mdean> jjc: could allow any typed URIref for property value

16:18:54 <mdean> Ian: are we still waiting for RDF to decide ...

16:19:28 <mdean> pfps: rdfs:comment is an unusual case because of social meaning - bad example

16:19:37 <mdean> not waiting for RDF Core on anything

16:19:49 <mdean> PatH: give an example of a property that isn't an annotation property

16:20:18 <mdean> pfps: want to tag classes with dc:creator, etc. in OWL DL

16:20:36 <mdean> pfps: not a problem in OWL Full

16:21:25 <mdean> Ian: don't we need to require that an explicit AnnotationProperty?

16:22:04 <mdean> pfps: technically possible to tag only unknown properties

16:22:49 <mdean> pfps: AnnotationProperty is not used for anything else

16:22:56 <mdean> Ian: how to tell from OWL Full?

16:23:26 <mdean> jjc: would prefer no ... constraints

16:24:02 <mdean> pfps: not typing AnnotationProperties is dangerous in some sense

16:24:54 <mdean> pfps: AnnotationProperties are an ugly hack

16:25:21 <mdean> pfps: ... but really convinced that they work

16:25:37 <mdean> ... within DL view of world

16:25:49 <mdean> ... but OWL Full lets you talk about things like how many annotation properties you have

16:26:19 <mdean> pfps: any OWL Full conclusions that fit in OWL DL subset match up

16:27:06 <mdean> jjc: ... correspondence theorem ...

16:27:25 <mdean> jjc: do we allow blank nodes as object of annotation properties

16:27:33 <mdean> pfps: currently no

16:27:37 <mdean> jjc: awful ...

16:28:04 <mdean> jjc: e.g. anonymous individual for dc:creator

16:28:39 <mdean> pfps: can't see any reason not to - add to list of actions for this meeting

16:32:55 <mdean> ... junk ...

16:33:34 <mdean> ... add directive to Abstract Syntax

16:34:05 <mdean> pfps: all properties need an axiom - we should be consistent

16:34:30 <mdean> jjc: last issuette: do we allow annotations on annotations

16:34:42 <mdean> Mike Smith: annotations all the way down ...

16:35:40 <mdean> Ian: no implementation is going to worry about AnnotationProperties that are never used

16:36:07 <mdean> summary:

16:36:37 <mdean> 1) types required on all non-builtin urirefs

16:36:52 <mdean> 2) optional for builtin urirefs

16:37:42 <mdean> [builtin means something defined in our documents]

16:41:52 <mdean> 3) [blank nodes -- to appear]

16:42:24 <mdean> 4) object of annotation property can be any uriref (see 1) or literal or blank node

16:42:43 <mdean> [xsd datatypes are builtins]

16:43:09 <mdean> [pfps: user-defined data types are another issue]

16:43:47 <mdean> 3) Types requires on blank nodes [includes Lists]

16:44:14 <mdean> 5) Top level directive for annotation properties

16:44:24 <mdean> 6) permit annotations on annotations

16:49:35 <mdean_> reconnected (argh)

16:49:52 <mdean_> jjc: what about rdf:first and rdf:rest in annotations? - issue to be considered

16:49:57 <mdean_> mdean_ is now known as mdean

16:51:05 <mdean> new topic: disjointness

16:52:20 <mdean> DisjointClasses(d1, d2, d3, d4)

16:52:34 <mdean> if all are blank nodes

16:52:52 <mdean> allows blank node to be the subject of more than 1 triple, requiring use of RDF nodeIDs

16:53:08 <mdean> jjc: suggest redefining mapping to avoid problem

16:55:08 <mdean> pfps: what about mapping backwards?

16:57:26 <mdean> Ian: want to maximize round-tripping

16:58:17 <mdean> jjc: current mapping rules even preserve order

17:02:06 <mdean> pfps: jeremy's mapping rule duplicates descriptions

17:05:04 <mdean> pfps: ... liberalization ...

17:05:20 <mdean> jjc: fewer syntactic categories

17:08:04 <mdean> allow reuse of syntactically equivalent constructs using nodeID

17:08:09 <pfps> proposal: liberalize mapping rules for disjoint classes to only require sufficient disjoint triples

17:10:50 <mdean> pfps: corresponce proof would have to be carefully examined to ensure that this works

17:11:35 <mdean> pfps: ... descriptions are "half-Platonic", as Platonic as we could make them and still be in RDF

17:13:20 <mdean> jjc: Peter's proposal

17:13:29 <mdean> 1) bnodes at descriptions form DAGs

17:13:37 <mdean> 2) the mapping rules permit

17:14:12 <mdean> T(description1) =T(description2)

17:14:37 <mdean> globally, whenever description1=description2

17:14:58 <mdean> [pfps: 2) doesn't work for Individuals]

17:15:52 <mdean> 3) bnodes of lists, all different, individuals, dataranges unchanged

17:16:01 <mdean> Unknown

17:16:14 <mdean> 4) bnodes of individuals form arbitrary graph?

17:18:06 <mdean> pfps: table DisjointClasses until we can investigate this?

17:20:29 <mdean> Known

17:21:12 <mdean> 5) No acylicity constraint in owl:disjointWith

17:23:26 <mdean> tentative agreement that jjc and pfps need to review - proof is in the proof

17:24:13 <mdean> new issue: syntax of property restrictions

17:27:49 <mdean> Ian: why does abstract syntax allow this

17:28:34 <mdean> ACTION (pfps): remove restrictions with n=0

17:31:50 <mdean> jjc: example

17:32:05 <mdean> restriction(p someValuesFrom(c1) allValuesFrom(c2))

17:32:06 <mdean> =

17:32:12 <mdean> intersectionOf

17:33:14 <mdean> restriction(p someValuesfrom(c1))

17:33:24 <mdean> restriction(p allValuesFrom(c2))

17:33:27 <mdean> )

17:34:37 <mdean> concerned about OWL DL vs. OWL Lite

17:35:11 <mdean> proposal: OWL Lite excludes multipart restrictions

17:38:25 <mdean> Mike uncomfortable with this - since cardinality and allValuesFrom restrictions are very common - duplication is common complaint with DAML+OIL

17:38:36 <mdean> Mike = mdean

17:38:52 <mdean> resolved

20:31:05 <ChrisW> anything going on?

20:46:30 <horrocks> Minutes of WebOnt editors meeting

20:46:36 <horrocks> Session 13:45 - 15:30

20:46:36 <horrocks> Jeremy proposes to allow anonymous descriptions at the top level in

20:46:36 <horrocks> the abstract syntax so that unused descriptions are legal in the RDF

20:46:36 <horrocks> syntax for OWL DL.

20:46:36 <horrocks> One possible solution is to introduce new declarations for them;

20:46:37 <horrocks> another solution, requiring less change to the AS is to allow

20:46:39 <horrocks> sameClassAs axioms with only 1 class.

20:46:41 <horrocks> Everyone seems to agree that we would like graphs containing

20:46:46 <horrocks> "meaningless junk", but which are otherwise DL, to be treated as

20:46:48 <horrocks> DL. The question is, what answer would a validator give when asked "is

20:46:49 <horrocks> this valid DL (yes/no)"?

20:46:51 <horrocks> Currently, such graphs are in full. One suggestion is to examine

20:46:53 <horrocks> semantics of document and allow arbitrary junk provided it was

20:46:56 <horrocks> semantically meaningless. This seemed highly undesirable as it would

20:46:57 <horrocks> force semantic analysis in order to determine syntactic class of an

20:46:59 <horrocks> ontology.

20:47:02 <horrocks> PROPOSAL: allow equivalentClasses with a single description, which

20:47:04 <horrocks> maps to the description. Result is that ontologies with "orphan"

20:47:05 <horrocks> descriptions are in DL. They are NOT in Lite.

20:47:07 <horrocks> Jeremy & Peter engage in incomprehensible techno-babble.

20:47:09 <horrocks> Discussion of meta-properties (e.g., priorVersion). Intention seems to

20:47:11 <horrocks> be that range and domain of priorVersion is ontologies (?), but what

20:47:13 <horrocks> if it is used between classes? Should we conclude that they are

20:47:17 <horrocks> ontologies? Is this useful and should it be allowed? Should be add

20:47:19 <horrocks> explicit domain and range constraints so it can't be used with

20:47:21 <horrocks> anything but ontologies?

20:47:23 <horrocks> PROPOSAL: that owl:priorVersion, owl:backwardsCompatibleWith and

20:47:26 <horrocks> owl:incompatibleWith are of syntactic category "ontologyPropertyID" in

20:47:27 <horrocks> the abstract syntax, and to add the following production:

20:47:29 <horrocks> directive ::= Annotation(ontologyPropertyID ontologyID)

20:47:31 <horrocks> The domain and range of these properties are owl:Ontology in both

20:47:34 <horrocks> owl.owl and owl full.

20:47:35 <horrocks> PROPOSAL: that owl:versionInfo is an annotationPropertyID that may be

20:47:37 <horrocks> used like any other annotationPropertyID.

20:47:39 <horrocks> ACTION: Guus to discuss the possibility of removing subProperty

20:47:41 <horrocks> relationship between owl:priorVersion and owl:backwardsCompatibleWith

20:47:42 <horrocks> and owl:in

20:47:48 <horrocks> owl.owl.

20:47:49 <horrocks> ACTION: Jeremy to generate domain and range constraints test cases for

20:47:51 <horrocks> owl:imports

21:01:37 <Zakim> SW_WebOnt(tp)8:00AM matches both webont and SW_WebOnt

21:09:27 <pfps> pfps start of phone call with Deborah McGuinness

21:09:49 <pfps> Guus: changes that require changes to overview - none?

21:10:05 <pfps> Guus: overview should have header information stuff - but versionInfo doesn't belong

21:10:31 <pfps> Guus: nothing needs to be done to overview

21:11:21 <pfps> Guus: move to guide

21:12:10 <pfps> Mike: wording changes, stuff about versionInfo - will be done soon

21:14:25 <pfps> Mike: ontology stuff rdf:about="", this should be acceptable (!), xml:base

21:15:36 <pfps> Guus: Mike Dean's comment on restrictions inside property axioms

21:15:55 <pfps> Mike: will be fixed, also [n]onNegativeInteger

21:17:16 <pfps> Jeremy: need type for data oneOf ?

21:17:55 <pfps> Guus: is this needed - yes

21:21:27 <pfps> various: what is the category?

21:22:01 <pfps> Pat: superclass of rdfs:Datatype

21:23:06 <pfps> Guus: the category is owl:DataRange

21:24:53 <pfps> Proposal: add owl:DataRange as syntactic category for data oneOf

21:25:24 <pfps> Deborah: just add a pointer whatever Reference says.

21:25:44 <pfps> Guus: need change to Reference to add owl:DataRange

21:26:14 <pfps> Mike: also add pointer

21:28:11 <pfps> Deb: overview only needs versionInfo, nothing for guide yet

21:34:28 <pfps> Guus: adding names for data ranges

21:34:45 <pfps> Ian: major change (i.e., its too late)

21:42:54 <pfps> Guus: treat this as a temporary fix

21:43:02 <pfps> Pat: temporary is dangerous

21:43:15 <pfps> Jeremy: not this one, as it is, and will be, in OWL Full

21:44:46 <pfps> Guus: leave it as it is, explanation to pacify Jim