Re: Understanding PowerLoom's upper ontology

On Aug 2, 2007, at 4:17 PM, Cameron Ross wrote:

> I need to understand PowerLoom's upper ontology so that I can  
> construct a representation of it using CLIF.  I'm not versed in  
> Stella and have primarily been analyzing the PL-KERNEL-KB module  
> through the Ontosaurus website (http://blackcat.isi.edu:8950/ploom/ 
> shuttle.html).  I have the following questions based on my  
> investigations to date:
> Ontosaurus reports that the PL-KERNEL-KB module contains 28  
> concepts, 121 relations and 42 instances.  However, I count 122  
> relations (see attached file).  Is this an "off by 1" counting error?

I'll have to investigate this.  I'll answer tomorrow.

The short answer is that the numbers were meant to be representative  
of the size and not necessarily definitive.  I will take a look and  
give a better answer tomorrow.

> Where can I find a description of the information displayed on the  
> Ontosaurus web pages?

I had hoped that most of the information and headings for the  
individual information would be generally self-explanatory, so there  
isn't any direct description of the content itself.  I'll answer any  
specific questions that you have, though.

> What exactly is the difference between an "instance" and "direct  
> instance"?

An instance is an instance of a particular class or any subclass.
A direct instance is an instance of a class with no intervening  
subclasses to which it belongs.

In other words, there is no subclass of the displayed class that has  
a direct instance as one of its instances.

> How come some relations make reference to instances (e.g. BINARY- 
> RELATION) while others refer to tuples (e.g. RELATION-CONSTRAINT)?

This is about how we display and label the extension of a relation.   
It depends on the arity.  Concepts (a particular type of unary  
relation) display instances.  Higher arity relations display tuples.

BINARY-RELATION is a concept, whose instances are binary relations.   
RELATION-CONSTRAINT is a binary relation, so it has tuples instead of  
instances in its extension.

> Where are the COMPUTATION-SPECIALIST and CONSTRAINT-SPECIALIST  
> instances, mentioned in the detail page for COMPUTED-PROCEDURE,  
> defined?

They are defined in internal structure.  There are certain items in  
the upper model of PowerLoom that have internal primitive definitions  
rather than being defined directly in the logic itself.  Literal  
types such as INTEGER, FLOAT and STRING are also defined directly as  
Stella objects.  This is a special-case mechanism to make the system  
able to reason effectively and efficiently with those constructs.

Specialists are defined in Stella itself, since they are really code  
that is executed and the logic doesn't have the procedural constructs  
for writing code.  In essence they escape from the logic.  So, those  
are specialists that know how to apply computations and constraint  
code to evaluate particular relations.  These are often used for  
numeric reasoning, and also at other times when it is more efficient  
to substitute special purpose programs for general deductive problem  
solving.

Ontosaurus was designed for examining the logic, so there are  
references to certain items that are not displayed.   There are a few  
other items that are not inspectable in Ontosaurus as well.   For  
example, we don't make skolem instances inspectable using Ontosaurus,  
although they will appear as instances of some concepts.  There are a  
number of reasons for that, one of which is that skolems don't have a  
unique printed representation, so we can't keep track of them over  
http connections where we need a unique reference.



> Thanks much,
> Cameron Ross.
> <pl_relations.txt>
> _______________________________________________
> powerloom-forum mailing list
> [email protected]
> http://mailman.isi.edu/mailman/listinfo/powerloom-forum