Notes from 19-20 July DAML Jumpstart Workshop

meeting home | DAML home

status: still drafty; meeting is adjourned, intend to fill in pointers to more of the presentations. $Date: 2000/08/13 16:13:34 $

Introductions & brief description of jumpstart projects

Mike Dean, BBN; PI for DAML Integration

Tom Martin, Schafer Corp; support to Jim Hendler DAML, TASK, HORUS

Brian Kettler, ISX; PI for HORUS Initiative

Jeff Heflin, University of Maryland; SHOE

Deborah McGuinness, Stanford Knowledge Lab

Ora Lassila, Nokia Research Center; Agent Technologies

Dan Brickley, University of Bristol, W3C Visiting Fellow

Massimo Marchiori, University of Venice & W3C

Tim Berners-Lee, MIT/W3C

Stefan Decker, Stanford

Lynn Stein, MIT/LCS

Dan Connolly, MIT/W3C

Ralph Swick, MIT/W3C

Longer presentations about jumpstart projects

Decker "Layers for Interoperability on the Web"

w/Sergey Melnik

wants n-ary relations, ordering of relations

Lynn: why is order so important as to be part of the base, and not say synonomy, and transitivity?

Stefan: transitivity requires computation, which we wanted to avoid at the lowest level. Ordering is motivated in part by UML.

Tim: How much effort to produce an RDF syntax canonicalizer using XSLT? Dan: not much, ~40 hours.

Ora: can do this easily with a set of SAX event handlers that output canonicalized RDF. I have an RDF parser in Common Lisp which I intend to contribute to the DAML community.discussion of oil:subClassOf

DanC writes:

     type(x,AandB)
        []  -- rdf:type --> oil:AND
                                      -- lhs --> A
                                      -- rhs --> B

Connolly

see Dan's notes Revision: 1.12 of Date: 2000/07/19 09:21:40 by Author: connolly (was then known as "hs78", but Dan expanded that into an explanation of his use of XSLT to extract RDF from XHTML pages.)

outline:

  1. the basic RDF statement model: subject, predicate, and object
  2. Schema/Ontology basics: Classes, properties
  3. propositional calculus: introducing implies, false
  4. predicate calculus: introducing forall (and in doing so, variables)
  5. Future work

Lassila: RDF Common Lisp tools

expect to release some tools to the DAML community:

XML parser
passes ~75% of the ~900 OASIS conformance rules
missing DTD support (not necessary for RDF)
API is "CL-SAX"; ~like SAX 1
has XML namespace support
XML formatter
produces XML text stream from a CL-SAX event stream
RDF parser
streaming; creates triples on the fly from the CL-SAX event stream
RDF "syntax normalizer"
SAX event stream filter; produces a stream using only
RDF non-abbreviated syntax two versions: Java 1.1.x and CL
RDF "data suite"
CLOS support for processing RDF triples
on-going work on an RDF graph visualizer (experimental)

Heflin: SHOE crawler demo

The Knowledge Annotator

Exposť

SHOE Search

TSE Ontology and risk path analyser

Decker: Expressing rules in GINF using RDF

see GINF page

  Knowledge Base
    includes other Knowledge Bases
    contain rule objects
     rule objects have head & tail objects
      head & tail have terms
       terms have subject, predicate, and objects
 

DanC: this model relies on closed world to know that an object is not a variable.

Tim: note that in an RDF model, you can remove any triple and the remainder will remain true.

DanC: want the database to be "sound" under deletion of arcs.

Tim: note that RDF implicitly has the existential quantifier; The subject of an RDF statement is implicitly said to exist.

Mike: how do you account for temporal data?

DanC: hard problem.

Tim: can associate expiration dates via HTTP.

McGuinness: pointers to particularly interesting ontologies

http://www-ksl-svc.stanford.edu:5915/ - main ontolingua server

http://www.ksl.stanford.edu/software/chimaera/ - chimaera - a merging and diagnostic ontology environment

Soundness under deletion of arcs

DanC:

must model this such that deleting one of the premise arcs doesn't lead you to incorrect conclusions. You will know less about some of the objects, but you will know that there is something about which you don't know anything.

Stefan: It is not completely correct to view RDF as a conjunction when modelling rules.

DanC: quoting is the technique to express the two levels together.

Massimo: adding a Cardinality property to Stefan's representation will resolve the monotonicity.

Jeff: when negation is added, does this soundness still hold?

Dan: consider 'it is not raining today'

{subject, x, today}, {predicate, x, weatherOf}, {object, x, raining}, {implies, x, false}

If you delete the 'implies' arc, all you have left is "there is a sentence that says it is raining today".

Tim: note that you don't have to reify the {implies, x, false} statement.

Jeff expresses concern about the syntax for reification.

Tim: RDF is the form for exchanging on the Web, should we have a short-hand form for e-mail, irc, whiteboard?

DanC: there will be SVG tools that let us draw the same diagrams that we're drawing on the whiteboard. For exchange with colleagues, exploit one of the existing syntaxes; prolog, etc.; don't need more syntaxes.

Deborah: good to exploit existing KIF tools too.

Tim: need KIF with URIs. Consider adding namespaces to one of the existing KR languages?

DanC: Guha is doing a form of this with SQL.

Questions

  1. SHOE's motivation for binary relations to n-ary relations [Brickley]
  2. Modeling definitive lists / local closed-worlds
  3. Is 'order' primary or secondary?
  4. What are the primary constructors?
    [DanC: implies, NOT, forAll]

    inverse

    maximum cardinality 1 (unique/equivalent)

  5. Expectations about what is in DAML [0.x]
    Ora: RDF for basic underlying representation plus SHOE with only binary predicates
  6. Are lists an acceptable way to close parts of your model?
  7. [Jeff] How to deal with inconsistent statements you may find.
  8. [Jeff] How to know which rules are applicable? like import.
  9. [Tim] Drop anything from RDF 1.0, RDF Schema?

    Jeff: e.g. aboutEachPrefix

  10. Dan's list at http://www.w3.org/2000/07/hs78/
  11. [Tim] pointer from a document to a transform that will create RDF and to the result of this transform.

Proposition: DAML 0.x is RDF 1.0 + [parts of] RDF Schema + SHOE horn clauses w/o negation with just binary relations and without claim groups

Layers

layer 3: ?n-ary relations

layer 2: ?FOPC

layer 1: ?Horn clauses

layer 0: RDF 1.0 Model ?+Schema

DanC: don't want to require every application to have a first-order theorem prover.

Deborah: interoperability at the level of deductive closure; different logic systems will conclude the same things. Levels of compatibility: able to consume the facts without complaint, finding the explicit facts, making the same deductions.

Mike: anticipate adding digital signatures, temporal qualification, etc. later.

Deb: description logics discussion circa 1992 produced set of descriptive constructors; Knowledge Representation Specification System (KRSS) from Peter Patel-Schneider.

Deb: all new K-R papers must describe: syntax, denotational semantics, inference rules, computational complexity. Some things, like loom, are incomplete; there are true things they cannot infer.

Lynn: should not assume that the Description Logic community consensus represents a consensus of the full K-R community.

Deb: refer to description logics for a variety of notions of equality.

Tim: see also Identity in Web Design Issues.

ACTION: Dan write a rule meaning "equivalent for the purposes of xxx" [e.g. xxx == W3C Advisory Committee representative, a relation with cardinality 1.] (isomorphism)

Some chapter/section headings for an august paper:

  1. RDF Normal form
  2. Some useful properties and classes for the above
  3. Building FPOC on top of RDF-NF
  4. Solidifying DAML-0.5
  5. Practicalities of logic on the web: document management etc
  6. Things for future work
  7. Appendix: WWW architecture

ACTION Mike: submit some uses cases

DanBri: [re: naming of relations] see Philippe Martin, "Propositions of Conventions and Ontologies for the Resource Description Framework".

** Ralph: propose a different term than namespace or schema for.

**DanBri: Take to system team "please separate (in the master list) obsolete mailing lists from real ones and make semantic-web@ obsolete"

"preceded by itself within quotes is obsolete" preceded by itself within quotes is obsolete.

<RalphS> use www-rdf-logic@w3.org for open discussion; archived

<RalphS> Note irc.w3.org:6665 channel #sw is also open

Comments

Dan: please agree to publish Web addresses for all your facts furthermore, please agree to only make monotonic changes to the content at those Web addresses.

RESOLVED: obsoletes(x, y) = y makes x obsolete http://www.w3.org/2000/07/document-maintenance/#obsoletes

RESOLVED: sublanguage(x, y) = x is a sublanguage of y http://www.w3.org/2000/07/document-maintenance/#sublanguage (see timbl for examples/semantics)

PROPOSED: rdfExtractor(x, y) = y will extract RDF from x http://www.w3.org/2000/07/document-maintenance/#rdfExtractor

note well:
for inverseof, see oil:inverseRelationOf


Ralph R. Swick <swick@w3.org> 19 Jul 2000
$Revision: 1.78 $ of $Date: 2000/08/13 16:13:34 $ by $Author: swick $