Re: Elegant solution to let OWL/RDF cover Closed World Assumptions (CWA), incl unique Name Assumption (UNA)

Dear Pascal and Jie,

Thank you very much for your response. To my great relief I found out that it is a hot topic and that it indeed has W3C's full attention.

As to Pascals last remark "...it needs spelling out explicitly...", I'd like to give a small example.
Assume a class Person with 10 pre-defined individuals, 4 have the property+value "isMale true", 4 have "isMale false" and 2 don't have the property "isMale" at all. If one takes the Open World Assumption (OWA) VERY literally (i.e. into the extreme), then during instantiation of a class expression "Person and (isMale value true)", the reasoner (Fact++, Pellet, etc.) might reason as follows: I indeed see 4 individuals that have the property+value "isMale true" visible(/explicitly) and the rest that have either another visible value for this property or don't have the property visible at all. But applying OWA (into the extreme) ,now who doesn't tell me that the other individuals don't have some hidden "isMale true" that I cannot see? Ergo: I can't give a result at all! But it DOES give a result, namely the 4 individuals with "isMale true"! I.m.h.o: isn't this a result that stems from Closed World Assumption (CWA) or do I overlook something essential ?!

Many thanks in advance,
Best Regards,
DJ




From: Jie Bao 
Sent: Friday, November 19, 2010 3:16 PM
To: Pascal Hitzler 
Cc: DJA222 ; public-owl-comments@w3.org 
Subject: Re: Elegant solution to let OWL/RDF cover Closed World Assumptions (CWA), incl unique Name Assumption (UNA)


In similar spirit, axioms with CWA and UNA can be seen as special cases of integrity constraints. A semantics is proposed in [1]. DJ's proposal gave it a syntax. 


[1] http://www.cs.rpi.edu/~taoj2/publications/IC-AAAI-2010.pdf


Regards
Jie



On Thu, Nov 18, 2010 at 14:01, Pascal Hitzler <pascal.hitzler@wright.edu> wrote:

  It has indeed crossed my mind (and that of some of the people I've been talking with) that one may want to have a simple "closure" - i.e., something much simpler than what most non-monotonic formalisms provide. It seems, though, that some of the things you describe below can be achieved by the autoepistemic K operator [1,2], by DL-safe variables [3,4], or the approach proposed in [5]. In any case, it needs spelling out explicitly...

  Best Regards,

  Pascal.


  [1] http://portal.acm.org/citation.cfm?doid=1754399.1754403

  [2] http://knoesis.wright.edu/faculty/pascal/resources/publications/mknftheo.pdf

  [3] http://knoesis.wright.edu/faculty/pascal/resources/publications/ELP_iswc08.pdf

  [4] http://korrekt.org/page/Description_Logic_Rules_%28monograph%29

  [5] http://knoesis.wright.edu/faculty/pascal/resources/publications/ELP2.pdf 




  On 11/16/2010 12:52 PM, DJA222 wrote:

    Dear OWL Staff,

    Hereby I would like to suggest an elegant solution to let OWL/RDF
    cover Closed World Assumptions (CWA), incl Unique Name Assumption
    (UNA).

    Due to its Open World Assumption (OWA), OWL/RDF  can hardly be used
    for e.g.: 1. finding (:instantiating) pre-defined individuals with a
    certain number of properties or without these properties at all (e.g.
    cardinality 0). 2. validation. OWL/RDF's OWA asserts that everything
    is possible (->infinite) until asserted otherwise by constraints
    (->finite). But doesn't infinity envelopes finity? Isn't processing
    things in a finite world (CWA) just a valid part of the infinite
    world (OWA)? Finite means that things are or (immidiately) can be
    made explicit. Then why not simply introduce a term like
    e.g."Explicit" that can be added to every constraint and applies to
    things/values that are visible at the very moment of instantiation?

    Example class expression: Person and (hasChild exactly 0 Explicit) :
    at the very moment that this class is being instantiated it "scans"
    for (pre-definied) individuals in class Person for which the property
    "hasChild" is explicitly absent (:exactly 0 Explicit). Although
    OWL/RDF itself leaves open the possibility that the individuals still
    might have hidden "hasChild" properties, the reasoner just looks for
    explicitly absent properties because the class expression tells it to
    do so.

    This way it can also be used for validation/integrity-check  with a
    class expression with "Explicit" included: if something is asserted,
    that can't be derived from explicitly present assertions at the very
    moment of instantiation, than this will be reported: NOT as being a
    OWL/RDF error/conflict/inconsistency, but just as a note to the user
    who fabricated this class expression. Again, without OWL/RDF itself
    denying that there might still be assertions that are just not
    visible at the moment of instantiation.

    Same story for UNA: by adding a term like e.g. "Unique" in a class
    expression it might notify the user (who wrote the expression) upon
    instantiation, that it has found individuals who are asserted
    (directly or implied) to be identical but have different names or
    that it has found more individuals than expected. Again, without
    OWL/RDF itself denying that there might by assertions that are just
    not visible at the moment of instantiation.

    All above OWL/RDF examples would still comply with OWA and Non-UNA
    demands, by assuming the possible outcome (with the terms "Explicit"
    and "Unique" used in expressions) not as OWL/RDF conlicts, but just
    as (user) notifications.

    Above is extremely important in research where INDIVIDUALS and
    relations between them are at the focus, instead of the more generic
    class approach. In certain researches,  thousands and thousands of
    data snippets (:Individuals) come in from different places and you
    want to look for certain properties/relations that these pieces
    share/have with/to one another. This can not easily be automated with
    present OWL/RDF. Yes, one might use SPARQL in some ways, but the aim
    is to let simple class instantiation do its work.

    In my conviction, with a slight addition, OWL/RDF semantics can proof
    to be a more complete basis for ANY semantic real world application
    and solution, and not just for a limited part!

    Hopefully you might reconsider this idea, or find find some similar
    solution, that really is in the need of many (potential) OWL/RDF
    practitioners.

    Sincerely yours,

    DJ Alexander



  -- 
  Prof. Dr. Pascal Hitzler
  Dept. of Computer Science, Wright State University, Dayton, OH
  pascal@pascal-hitzler.de   http://www.knoesis.org/pascal/
  Semantic Web Textbook: http://www.semantic-web-book.org
  Semantic Web Journal: http://www.semantic-web-journal.net