Re: comment on N-ary relations draft

From: Peter F. Patel-Schneider (
Date: 12/08/04

  • Next message: Benjamin Grosof: "regrets for today"
    I would strongly suggest that the note be changed to stress that the
    driving distinction between the two representation patterns is
    modelling-related and not logical.  I also suggest that that the note
    mentions that there is no real logical difference between the two patterns,
    and that OWL makes no distinction between the two patterns at all, at least
    as they are presented in the note.
    As well, I think that you should not use "representation pattern", as it
    gives the wrong impression.  If the driving difference is some modelling
    concern, and not logical structure, then it would be better to call the two
    patterns "modelling patterns" instead.
    The note also sabotages the point that Guss makes below about a naming
    difference between the two patterns, as it names the relationship class and
    the relationship instances in both patterns.  If this is the driving
    difference, or even just a substantive difference, then it is exceedingly
    strange that the note not only is very silent on the difference, mentioning
    the issue at all only obliquely and only near its end, but actually
    supresses it, using meaningful names for both patterns.  This supression is
    so unusual so as to make me wonder whether the distinction is actually a
    valid modelling distinction at all.  I, personally, see effectively no
    modelling difference between the two patterns.  In particular, I see no
    difference between the two patterns with regards to either the necessity or
    suitability of naming any of the constructs they require.
    As far as so-called "wizards" for promoting from simple properties go, I
    don't see that there is any reason why either pattern should be preferred.
    Similarly, if there is a wholesale need to promote simple properties then I
    don't see any advantage to moving to pattern 1 over pattern 2.  In fact, if
    uniformity is what is wanted, then why not just stick with only one
    pattern?  Surely that would prevent confusion over the difference!  Because
    neither RDF/RDFS nor OWL really distinguish between the two patterns, why
    should modellers be presented with the choice?
    Finally, if the notions from Guarino and Welty's work are important here,
    then surely the notions should be referenced, their relevance argued, and
    their differential applicability to the patterns demonstrated.
    Peter F. Patel-Schneider
    Bell Labs Research
    [The contents of the previous emails have been rearranged to better show
    the progression of the thread.]
    From: Alan Rector <>
    Subject: Re: comment on N-ary relations draft
    Date: Wed, 08 Dec 2004 08:45:08 +0000
    > All
    > Guus has said it more clearly than I could of.  The issue is modelling rather
    > than logical structure.  
    > Another point is that it is common in the evolution of an ontology to start by
    > using a simple property plus binary relation and then move to pattern 1 -
    > re-representing the property as a class - as it becomes clear that an n-ary
    > relation is needed in at least some cases.  This is a common enough operation
    > that we are implementing a "wizard" to make it easier, either for a single
    > property or a whole group of properties.  
    > Often it happens that there is a large set of properties/cases that one wishes
    > to model consistently.  In this case, if any of the cases needs to be modelled
    > with pattern 1, then they all do.  Otherwise, modellers become hopeless confused
    > as to which conceptual relationship is to be modelled in which way.  
    > Finally, version 1 matches to ontological notion of a "quality" and "quality
    > space" as in Guarino and Welty's work, whereas this consideration is not
    > relevant to pattern 2.
    > Regards
    > Alan
    Guus Schreiber wrote:
    > Peter, thanks very much for the comment.
    > Natasha and Alan are probably the best people to answer this, but I will
    > give you my reading of the difference.
    > The distinction is of a modelling nature. In Pattern 1 you create a
    > helper relation (represented as a class) which has (in most cases) no
    > name in the domain of interest. In Pattern 2 the relation class has some
    > name in the domain, typically a noun representing some activity (e.g.
    > purchase, enrolment, transaction, subscription). I think this is an
    > important difference of which developers should be aware. In particular,
    > in the case of pattern 1 a an engineer might find it weird to construct
    > a name from the blue, and it may help her/him to know it's actually good
    > practice. The resulting representation is very similar, I agree. Maybe
    > we should make this point more clear in the text.
    > Hope this helps,
    > Guus
    > Peter F. Patel-Schneider wrote:
    > > I just read the N-ary relations draft and I am somewhat confused as to why
    > > it has the two representation patterns.  I don't see that the two patterns
    > > are different in any substantial way as the only difference between them is
    > > the direction of one arrow.  This difference may matter in some formalisms
    > > but doesn't in RDF/RDFS (as they are too weak to notice much difference) or
    > > OWL (as it has the inverse construct).
    > >
    > > So, my question is why maintain the two different representation patterns?

    This archive was generated by hypermail 2.1.4 : 12/08/04 EST