LOOM and PowerLoom

Hi

 

I have been searching for a language to represent ontologies. I have some questions (they are based on “Evaluating Knowledge Representation and Reasoning

Capabilities of Ontology Specification Languages” by Oscar Corcho and Asunción Gómez-Pérez)

 

1. Are there any features that LOOM supported and PowerLoom does not?
2. The other way round?
3. Is the inference engine of PowerLoom complete? (The mentioned paper says that for LOOM, the engine is sound but not complete)
4. With what purpose were LOOM and then PowerLoom created?
5. Does STELLA have an independent compiler or is there a translator from STELLA to JAVA, C++ and LISP and then the compilation is done with the compiler of the specific language?

 

Thanks

Sid

                                                                  Honeywell

Email           :   rajesh.kishore@...

                         swiftpanther@...

Mobile         :   +91 - 9845844025

 

_______________________________________________
powerloom-forum mailing list
powerloom-forum@...
http://mailman.isi.edu/mailman/listinfo/powerloom-forum

On Jun 18, 2008, at 7:37 AM, Goyal, Siddharth(IE10) wrote:

> Hi
>
> I have been searching for a language to represent ontologies. I have  
> some questions (they are based on “Evaluating Knowledge  
> Representation and Reasoning
> Capabilities of Ontology Specification Languages” by Oscar Corcho  
> and Asunción Gómez-Pérez)
>
> • Are there any features that LOOM supported and PowerLoom does not?

Yes.

*  Loom has a production rule system.
*  Loom has an action and method dispatch system.
*  Loom has had more work done on its classifier than PowerLoom
    - certain subsumption relationships based on cardinality  
constraints work in Loom but have not been implemented yet in PowerLoom.
    - qualified restrictions work in Loom but have not been  
implemented as frame predicates in PowerLoom (although you can always  
write the first order logic equivalent instead)

> • The other way round?

*  PowerLoom has a much more expressive representation language.  It  
uses 1st (and a bit of higher) order logic.  Loom uses a description
logic that is more expressive than OWL, but still less than first  
order logic.
*  PowerLoom reasons with a full logic rules system.
*  PowerLoom has facilities for mapping relations to database tables.  
(read-only at the moment)
*  PowerLoom is available in multiple target languages (C++, Java,  
Common Lisp).  Loom is available only in Common Lisp.
*  PowerLoom can do more reasoning with complex n-ary relations.

> • Is the inference engine of PowerLoom complete? (The mentioned  
> paper says that for LOOM, the engine is sound but not complete)

No.  It cannot be, since it implements first order logic.

> • With what purpose were LOOM and then PowerLoom created?

Both were created to provide expressive and practical knowledge  
representation for creating applications.

> • Does STELLA have an independent compiler or is there a translator  
> from STELLA to JAVA, C++ and LISP and then the compilation is done  
> with the compiler of the specific language?

The latter.  Stella is translated into the target language and then  
compiled by a standard compiler or that target language.


_______________________________________________
powerloom-forum mailing list
powerloom-forum@...
http://mailman.isi.edu/mailman/listinfo/powerloom-forum

> > [Siddharth Goyal]
> > paper says that for LOOM, the engine is sound but not complete)

> [Thomas Russ]
> No.  It cannot be, since it implements first order logic.

Just to be pedantic, it depends on what you mean by "complete."  There
are complete theorem provers for first-order logic, meaning that if a
theorem is true in all models of a set of axioms, the prover will
eventually find a proof of that fact.  The Powerloom theorem prover
sacrifices completeness for speed, or at least that's my guess.

The sense in which, say, Peano arithmetic is incomplete is quite
different, even though Peano arithmetic is a first-order theory.  It
is still the case that a resolution theorem prover can prove any
statement that is _true in all models_ of Peano arithmetic; but the
_theory itself_ is incomplete because there are statements that we
have compelling reasons to believe the truth of but which are not true
in all models.  That is, there are nonstandard models in which these
statements are false.  The compelling reasons revolve around the fact
that if the statements are false then Peano arithmetic is inconsistent
and hence worthless (and it has no models at all).

I'm sure this makes these distinctions perfectly clear :)

                                             -- Drew McDermott


_______________________________________________
powerloom-forum mailing list
powerloom-forum@...
http://mailman.isi.edu/mailman/listinfo/powerloom-forum

Just adding my two cents here:

>>>>> Drew McDermott <drew.mcdermott@...> writes:

>> > [Siddharth Goyal]
>> > paper says that for LOOM, the engine is sound but not complete)

>> [Thomas Russ]
>> No.  It cannot be, since it implements first order logic.

As Drew clarified :-), standard first order logic is only
semi-decidable, that is, if a statement is a theorem, an algorithm
such as a resolution theorem prover will find a proof for it, even
though for some theorems it might take a very very long time; however,
when applied to some non-theorems, the algorithm will never terminate
(hence, only "semi"-decidable).  However2, given the worst-case
exponential complexity for the decidable portion, even a complete
algorithm as used by a resolution prover will be practically
incomplete, since, for some theorems it will not terminate in your
life time.

As Drew guessed correctly, PowerLoom's inference engine is not
complete in this sense.  PowerLoom uses a natural deduction inference
engine, and, for example, one of the inference rules we do not
currently implement is reasoning by cases (A->C, B->C, AvB |- C).

What we should do is to either characterize these areas of
incompleteness more carefully, or, extend PowerLoom's inference engine
to be complete (in the sense a resolution prover is complete) for
knowledge bases that do not go beyond standard first order logic.  The
main reason why this hasn't happened yet, is that it is of relatively
little practical value for the overall system, and would primarily
make it easier for us to write certain papers about PowerLoom.

While I am at it: in my opinion, a KR&R system such as PowerLoom
occupies a very different space compared to theorem provers such as,
for example, Otter (or Prover9).  What a KR system should do well is
to allow you to represent potentially large amounts of knowledge
easily and with minimal restrictions.  The job of the inference engine
is then to efficiently make the expected inferences that are more or
less "just around the corner".  That is, you should generally not be
surprised by the inferences your KR system makes.  If a particular
expected inference requires a long reasoning chain for the system to
find, you probably should add some rules or model your KB differently.
This is very different from the space that theorem provers address,
which generally try to prove deep theorems - say Fermat's Last Theorem
- from a small number of axioms or assumptions.  Turning that around,
completeness in the sense described above isn't really that important
for a KR&R system such as PowerLoom, what's much more important is
completeness in the sense of finding all the inferences one should
"reasonably expect" in "reasonable time".  The problematic part with
this last statement is that this "reasonable" space is difficult to
define or formalize.

Hans

--------------------------------------------------------------------------
Hans Chalupsky, PhD                     USC Information Sciences Institute
Project Leader, Loom KR&R Group         4676 Admiralty Way
<hans@...>                          Marina del Rey, CA 90292
(310) 448-8745
--------------------------------------------------------------------------

> Just to be pedantic, it depends on what you mean by "complete."  There
> are complete theorem provers for first-order logic, meaning that if a
> theorem is true in all models of a set of axioms, the prover will
> eventually find a proof of that fact.  The Powerloom theorem prover
> sacrifices completeness for speed, or at least that's my guess.

> I'm sure this makes these distinctions perfectly clear :)

>                                              -- Drew McDermott


> _______________________________________________
> powerloom-forum mailing list
> powerloom-forum@...
> http://mailman.isi.edu/mailman/listinfo/powerloom-forum
_______________________________________________
powerloom-forum mailing list
powerloom-forum@...
http://mailman.isi.edu/mailman/listinfo/powerloom-forum

Hi

Thanks to everyone for your response. Some more clarifications…

1.      So PowerLoom has no production rule system? It’s completely missing?

2.      Is there any plan to provide the missing features of LOOM in PowerLoom? If yes, by when?

Sid

-----Original Message-----
From: Thomas Russ [tar@...]
Sent: Wednesday, June 18, 2008 11:45 PM
To: Goyal, Siddharth(IE10); powerloom-forum@...
Subject: Re: [PowerLoom Forum] LOOM and PowerLoom

On Jun 18, 2008, at 7:37 AM, Goyal, Siddharth(IE10) wrote:

> Hi

>

> I have been searching for a language to represent ontologies. I have 

> some questions (they are based on “Evaluating Knowledge 

> Representation and Reasoning

> Capabilities of Ontology Specification Languages” by Oscar Corcho 

> and Asunción Gómez-Pérez)

>

>       • Are there any features that LOOM supported and PowerLoom does not?

Yes.

*  Loom has a production rule system.

*  Loom has an action and method dispatch system.

*  Loom has had more work done on its classifier than PowerLoom

    - certain subsumption relationships based on cardinality 

constraints work in Loom but have not been implemented yet in PowerLoom.

    - qualified restrictions work in Loom but have not been 

implemented as frame predicates in PowerLoom (although you can always 

write the first order logic equivalent instead)

>       • The other way round?

*  PowerLoom has a much more expressive representation language.  It 

uses 1st (and a bit of higher) order logic.  Loom uses a description

logic that is more expressive than OWL, but still less than first 

order logic.

*  PowerLoom reasons with a full logic rules system.

*  PowerLoom has facilities for mapping relations to database tables.  

(read-only at the moment)

*  PowerLoom is available in multiple target languages (C++, Java, 

Common Lisp).  Loom is available only in Common Lisp.

*  PowerLoom can do more reasoning with complex n-ary relations.

>       • Is the inference engine of PowerLoom complete? (The mentioned 

> paper says that for LOOM, the engine is sound but not complete)

No.  It cannot be, since it implements first order logic.

>       • With what purpose were LOOM and then PowerLoom created?

Both were created to provide expressive and practical knowledge 

representation for creating applications.

>       • Does STELLA have an independent compiler or is there a translator 

> from STELLA to JAVA, C++ and LISP and then the compilation is done 

> with the compiler of the specific language?

The latter.  Stella is translated into the target language and then 

compiled by a standard compiler or that target language.

>

> Thanks

> Sid

>                                                                   

> Honeywell

> Email           :   siddharth.goyal@...

>                          swiftpanther@...

> Mobile         :   +91 - 9845844025

>

> _______________________________________________

> powerloom-forum mailing list

> powerloom-forum@...

> http://mailman.isi.edu/mailman/listinfo/powerloom-forum

_______________________________________________
powerloom-forum mailing list
powerloom-forum@...
http://mailman.isi.edu/mailman/listinfo/powerloom-forum

On Jun 20, 2008, at 8:15 AM, Goyal, Siddharth(IE10) wrote:

> Hi
>
> Thanks to everyone for your response. Some more clarifications…
>
>
> 1.      So PowerLoom has no production rule system? It’s completely  
> missing?
>

Correct.  There is no production rule system.  Instead, PowerLoom  
implements a much more extensive logic rule reasoner than Loom.  Our  
experience with Loom seemed to indicate that the production rule  
system was not particularly widely used.

> 2.      Is there any plan to provide the missing features of LOOM in  
> PowerLoom? If yes, by when?
>
At the moment we do not have any plans to add production rules to  
PowerLoom, as none of our projects need that capability.


_______________________________________________
powerloom-forum mailing list
powerloom-forum@...
http://mailman.isi.edu/mailman/listinfo/powerloom-forum