Re: SUO: Questions-comments about SUMO116.pprj: Concepts and Types
Jos,
Thanks for making your first post so substantive. Comments below:
At 12:43 PM 11/2/2001 +0100, Jos Lehmann wrote:
>Dear SUOntists,
>
> as this is my first message to the list, I should
> probably introduce
>meself. At this purpose there is a short description of my research
>project at the end of this e-mail. For any further relevant information
>on my research, please see the paper on causality which was presented at
>the SUO workshop (Lehmann & Breuker 2001).
>
>Here I would like to go immediately to the point, by putting a
>(structured) question to whom in the list might be interested to provide
>an answer, or at least a replay. I need to solve some doubts of mine,
>which might have some relations to SUMO. In order to avoid unnecessary
>boredom to the reader, I invert the logical order of paragraphs (i.e.,
>first the questions (4) and then the premises to the question (1, 2, 3)).
>
>Looking forward to some feedback,
>Jos.
>
>
>4. Questions: I'm really positively impressed by the quantity of
>information in SUMO (-116.pprj). It seems to me there is a problem,
>though, in the organization of such information. I think such luck of
>organization is *also* due to an abuse of the subclass relation, which
>is used for representing both subconcepts and subtypes. (For instance: I
>would consider a Region to be a type of the class-concept object rather
>than a subconcept of it - i.e., a region is a subconcept of the
>class-concept object which *only* inherits *a part* of the slots
>comprised by the class-concept object, not all of them! Why would a
>region need, for instance, such a slot as "exploits"?). So, I wonder:
>
Do you have a formal definition in first order logic that you recommend for
type and subclass that shows how these differ?
What are the slots that you feel Region should not inherit?
> a. Is there any technical reason for not distinguishing between
>concepts and their types, and "squeezing them all" in the ambiguous
>subclass relation?
I wasn't aware there was an ambiguity in the subclass relation, but I'd be
happy to be made aware.
>
> b. If not, wouldn't it be possible to make such distinction by means
>of an incremental inheritance of slots by the types of an object?
>
> c. How would this incrementality be expressed in Protege, Ontolingua
>or KIF?
>
>1. Premise: I'm working at the implemantation in Protege 1.6.1 of a
>revised version of the ideas presented in (Lehmann & Breuker 2001).
>After a month of stuggeling with Ontoligua, I decided to pass to
>Protege. I thought that this choice would in the end make more difficult
>to produce a KIF version of my ontology and therefore complicate
>communication with anything produced by SUO. To my pleasent surprise,
>though, two days after I started working with Protege, Adam posted
>SUMO116.pprj to the Protege-users list.
Thanks go to Pat Cassidy for producing the SUMO in Protege version.
>2. Premise: A fundamental distinction I "cherish" in my definition of
>causal relations, is the distiction between Concept and Type.
>
> a. A concept is just a set of attributions (in Guarino's terminology).
>
> b. A type is a concept plus a priority, i.e. an order between the
>attributions, plus a fixed range of the attributes, i.e. the values of
>the attribution (in Guarino's terminology).
Could you explain further the semantics of the ordering you propose? What
does it mean to have "an order between the attributions"
>3. Premise: Just like in SUMO, the fundamental concepts of my ontology
>of causal relations are the notions of object and process. By following
>what explained in 2, I would proceed as follows in defining such two
>concepts and their types:
>
> a. Define a series of attribution (as slots in Protege). Example:
>:Location, :Mass. Ranging on attributes (classes in Protege) Location,
>Mass
SUMO already includes the relations 'located' and 'measure' which take the
arguments Physical and Object, and Object and ConstantQuantity
respectively. See
<http://ontology.teknowledge.com:8080/rsigma/SKB.jsp?req=SC&skb=WordNetMerge&id=87>
and
<http://ontology.teknowledge.com:8080/rsigma/SKB.jsp?req=SC&name=measure&skb=WordNetMerge>
> b. Define a class (Object) comprising such slots (:Location, :Mass).
>Example:
>Object:
>:Location, :Mass
>
> c. Define a relation (a facet in Protege) that fixes the ontological
> Priority
>among slots. Example: Priority(:Location) = 1, Priority(:Mass) = 2. (Done)
What does the priority signify? What inferences does it allow you to draw?
> d. Define Type of Object (sub-class of Object in Protege), according
>to the fixed Priority and to ranges of the instances of the classes
>underlying the (prioritized) slots. Example:
>ObjectType1 (Universal): :Location = Any;
>ObjectType1.1 (Concentrated): :Location = Any, :Mass > 0; ObjectType1.2
>(Concentrated): :Location = Any, :Mass = 0.
>
>The result would look as follows (A tree, I hope...;*):
>
> Object: (Concept)
> :Location,
> :Mass
> |
> ObjectType1 (Universal Object):
> :Location = Any.
> /\
>ObjectType1.1 (Concentrated): ObjectType1.2 (Distributed):
>:Location = Any, :Location = Any
>:Mass > 0 :Mass = 0
>
Is this a proposal for creating two new subclasses of 'Object', where one
has the axiom
(=>
(instance ?X DistributedObject)
(measure ?X (MeasureFn 0 Kilogram))
>======================================================
>
>Background of my reasearch
>
>I am currently working on the development of
>an ontology of concepts (i.e. classes) for reasoning about the causal
>relations that hold in the description of a legal case. The general aim
>of my project is to come up with a meta-model of the essential data that
>are necessary (and sufficient) for (automatically) determining chains of
>causation (and legal responsibility) in a case. I intend to use Protege
>2000 for storing my ontology, which in its natural language form
>consists of approx. 40 definitions.
That sounds very interesting. Have you already developed the 40 natural
language definitions? If so, could you post them? We'd be happy to help
formalize them.
Adam
>======================================================
>--
>Jos Lehmann
>Research assistant and Ph.D. candidate
>Department of Computer Science and Law
>Faculty of Law
>University of Amsterdam
>jos@lri.jur.uva.nl
Adam Pease
Teknowledge
(650) 424-0500 x571