Re: assert vs instance-of and other questions

Hi, 

On 2006/01/13, at 8:36, Hans Chalupsky wrote:

> Here's a little trick to work around the restriction of `instance-of'
> in PowerLoom.  You can define your own relation, e.g., `isa', to do
> what you want.  For example:
>
> STELLA(16): (defconcept animal)
> |c|ANIMAL
> STELLA(17): (defconcept monkey (animal))
> |c|MONKEY
> STELLA(18): (defrelation isa (?i ?c))
> |r|ISA
> STELLA(19): (assert (=>> (and (isa ?i ?c)
>                               (concept ?c))
>                          (and (instance-of ?i ?c)  ;; not really necessary
>                               (?c ?i))))
> |P|(FORALL (?i ?c)
>    (=>> (AND (ISA ?i ?c) (CONCEPT ?c))
>        (AND (INSTANCE-OF ?i ?c) (HOLDS ?c ?i))))
> STELLA(20): (assert (isa george monkey))
> |P|(ISA GEORGE MONKEY)
> STELLA(21): (ask (animal george))
> TRUE

This is nice.

> STELLA(22): (all-facts-of george)
> (|P|(ISA GEORGE MONKEY) |P?|(MONKEY GEORGE) |P?|(INSTANCE-OF GEORGE MONKEY))
> STELLA(23): 
>
> The `all-facts-of' shows that `(MONKEY GEORGE)' was inferred via
> forward-inference.

You mean |P?| indicates a forward-inference proposition, right?
Now another mystery is resolved :-)

Thank you very much.
I also CC this mail to the ML. This info might be helpful for others.

Kotaro

>
> Hans
>
>>>>>> Hans Chalupsky <[email protected]> writes:
>
>> One follow-up to Tom's reply:
>>>>>> Kotaro FUNAKOSHI <[email protected]> writes:
>
>>> Dear members,
>>> I have three questions about PowerLoom.  I will highly appreciate it 
>>> if somebody gives me answers.
>
>>> (1) assert vs instance-of
>
>>> I found an asymmetricity between (A) asserting conceptual propositions using 
>>> command "assert" directly and (B) asserting conceptual propositions using 
>>> relation "instance-of".
>
>>> ...
>
>>> Why cannot PowerLoom answer that (monkey george) is TRUE when I use
>>> "instance-of" even though it answers correctly against (ask (instance-of george animal))?
>
>>> I just wanted to use "instance-of" because I prefer frame-style descriptions
>>> such as
>>> (definstance george 
>>> :instance-of monkey
>>> :likes banana
>>> :at forest1
>>> :age 4).
>
>>> I can simply avoid this problem by using always command "assert" like
>>> (definstance george
>>> :likes banana
>>> :at forest1
>>> :age 4)
>>> (assert (monkey george)).  ( yet not so elegant :-( )
>
>>> Even avoidable, such a behavior of PowerLoom makes me feel ill at ease...
>>> Is this an intended behavior or just a bug?
>
>> Well it's somewhere inbetween.  "instance-of" and "type-of" are really
>> just intended to be query relations that allow you to retrieve the types
>> an instance belongs to.  PowerLoom's inference machinery makes strong
>> assumptions about the nature of type assertions in many places and
>> looks for assertions of the type `(<concept> <i>)', but not for
>> `(instance-of <i> <concept>)'.  Currently, `instance-of' assertions
>> are not forbidden but also not normalized into the corresponding
>> `(<concept> <i>)' assertion.  One way to work around this would be
>> by adding a rule as you suggest below, however, there are two caveats:
>
>> (1) the rule should be a forward rule for efficiency, that way every
>> time you assert `instance-of' the corresponding type assertion is
>> generated instantly and no inference is needed during backchaining to
>> generate it.  You would do this as follows (note the double arrow):
>
>> (forall (?x ?c)  (=>> (and (concept ?c) (instance-of ?x ?c)) (?c ?x)))
>
>> This also avoids the rule indexing problem.
>
>> (2) unfortunately, this doesn't work since for performance reasons
>> `instance-of' is currently marked as a non-chainable relation, so the
>> rule will not do what you want it to.  Conceivably, this might be
>> relaxed at some point.
>
>>> (2)  Warning: PowerLoom can't index the rule
>
>>> I tried to avoid the problem described above by using a rule. But I got a
>>> warning message "Warning: PowerLoom can't index the rule"  and the rule
>>> seems not to work.
>
>>> The rule is here:
>>> (defrule "instance-of TO assert"
>>> (forall (?x ?c)  (=> (and (concept ?c) (instance-of ?x ?c)) (?c ?x))))
>
>>> What's the meaning of the message and what's wrong with my rule?
>
>> Just to elaborate on Tom's reply: the reason that this can't be
>> indexed is that the rule consequent is second-order since it has a
>> variable in the relation position.  This rule would be way too general
>> and could lead to serious performance problems, since it would match
>> any unary goal.  Turning the rule into a forward rule fixes this,
>> since in that case the relation variable ?c gets instantiated during
>> forward inference and produces a regular assertion (again with the
>> caveat that currently this won't work for "instance-of").
>
>> Hans
>> _______________________________________________
>> powerloom-forum mailing list
>> [email protected]
>> http://mailman.isi.edu/mailman/listinfo/powerloom-forum
>
>