F2F1 Minutes Session 5

From OWL
Jump to: navigation, search

This is part of F2F1 Minutes

OWL Working Group Meeting Minutes, 07 December 2007

DRAFT. Currently Under Review


See also: IRC log

Jeff Pan

(Scribe changed to Jeff Pan)

Alan Ruttenberg: Major issue seems to be whether to use xsd datatype semantics


Slides for this session: Media:sattler-f2f1.pdf

Joanne Luciano: Alanr, let's meet when you get back

Uli is presenting

Joanne Luciano: are we starting with Issue 25 as in the agenda?

OWL DL does not support user defined datatypes

Uli Sattler: users want to represent intervals

... and comparisons

Ian Horrocks: slides available at http://www.cs.man.ac.uk/~horrocks/owl1.1-datatypes.pdf

... in OWL DL no inverse functional datatype properties

... not to mention composite keys (not even OWL Full supports this)

Boris Motik: we might want to keep the unit mapping out of TBox

Jeremy Carroll: second

Sebastian Brandt: there are real world examples

... that we need datatype mapping in the TBox

Bijan Parsia: both needed

Uli Sattler: we have examples of seeing class subsumption checking based on datatype constraints

Carsten Lutz: it is difficult to choose one standard set, e.g. covering integers, rational, +, *, ...

Uli Sattler: as many as possible

Jeremy Carroll: each simple example is easy

... but have concern on having all of them, which makes it hard

Bijan Parsia: Sebatian's use case: http://www.w3.org/2007/OWL/wiki/N-ary_Data_predicate_use_case

Sebastian Brandt: combining DL and data values are important and useful, there are many tasks that you could not solve if you treat datatypes externally

Alan Ruttenberg: do we want to detect the problematic cases?

Joanne Luciano: what does alan mean by "detect"?

Jeff Pan: there are existing works on datatype groups, a mechanism is already there, see http://www.websemanticsjournal.org/papers/20050811/document.pdf

Thomas Schneider: Uli has just added "[Alan] add support to check whether this mechanism (second item of '3.') has been used 'safely'". Does this help?

... even in OWL DL, freely combinations are not possible, e.g. transitive properties are not allowed to used in number restrictions

Alan Ruttenberg: I mean during species validation, for example. Or via declarations of what features are used and flagging incompatible combinations

Jeremy Carroll: what happen if data in the user databases having both integers, rationals + and * ...

Michael Smith: link again http://www.w3.org/2007/OWL/wiki/N-ary_Data_predicate_use_case

Boris Motik: we need some datatype profile

Carsten Lutz: second boris point

Jeff Pan: two points: 1) profiling is a good idea, there have been work there such as datatype groups

... and we could provide a list of feasible datatype groups

2) if users have integers, rationals + and *, we could simply have type promotion, promoting integers into rational, and it is still decidable, see http://www.w3.org/TR/swbp-xsch-datatypes/ and http://www.w3.org/TR/xpath20/#promotion

Alan Ruttenberg: maybe we could have a stroll poll on this

Bijan Parsia: we all agree that some sort of datatypes are needed, no matter in OWL or RIF

... many of our cases cannot be addressed by RIF

Jeremy Carroll: transitive issue is different

Bijan Parsia: linear polynomial (in-)equations over the reals or cardinals with order relations,
Bijan Parsia: nonlinear multivariate polynomial (in-)equations over complex numbers,
Bijan Parsia: (from the racer manual: http://www.sts.tu-harburg.de/~r.f.moeller/racer/racer-manual-1-7-19.pdf )
Bijan Parsia: page 11
Bijan Parsia: See page 47 and 48
Bijan Parsia: e.g., (* real AN ) (AN of type real or complex)

Jeff Pan: besides Racer, an extension of FaCT (FaCT-DG) also supports n-ary and datatype groups

Uli Sattler: we could have some more general proposal, rather than specific ones

Bijan Parsia: we don't have to require all our implementors to implement everything, so we should be flexible somehow

Uli Sattler: the 4th point: easy keys

Markus Krötzsch: in foaf people use b-nodes rather than individuals, so the easy key might not solve the foat problem

Bijan Parsia: What I meant: It's a bad idea to, in committee, to significantly and somewhat arbitrarily increase the implementation burden. But without adding a hook, implementors *can't* (compatibly) experiment
Bijan Parsia: So, let's add the hook and be cautious about how we fill in the hook

stall poll 1: many +1, no -1, four 0


straw poll 2: (all) +1

straw poll 3 (about 2-b): many +1, two (conditional) -1, six 0

straw poll 4 (n-ary datatype): twelve +1, six -1, five 0

straw poll 5(easy key): 22 +1, one -1

Boris Motik: one profile proposal: a set of default profiles and allowing users to have arbitrary profiles

Boris Motik: or people would be able to define their own profiles

Another go, boris' profile: a fixed set of profile and also allowing people to define their owl profiles

alanr's proposal: a fixed set of profile

Bijan Parsia: Note the current support for unary datatypes is already fragmenty: http://spreadsheets.google.com/pub?key=pTmcCXR-dV6RpTEPxB0O-DQ

straw poll on profiling on datatype: eighteen +1, four 0

Thomas Schneider: dmitry tsarkov introduced himself ...
Thomas Schneider: ... email: tsarkov@cs.man.ac.uk

scribers should clean up yesterday's minutes by next telecon

(by IanH and no objections)