🚀 go-pugleaf

Patrick Juola <juola@mathcs.duq.edu> wrote in article
<7jlotk$c0f$1@quine.mathcs.duq.edu>...
> 
> The knowledge acquisition bottleneck becomes the madness.  We've just
> identified one syntactic construction (identical verb conjunction) to
> which you postulate a new KR primitive that must be stored as
> meta-knowledge (presumably on most base knowledge units in the database).
> I will then identify *another* syntactic construction, to which you
> will then postulate another KR primitive, and so forth.  Trust me on
this;
> I'm a linguist and I can pull constructions indefinitely -- and when I
> cease being able to pull them out of thin air, I can always pull them
> out of the literature. 8-)
> 
> So in the long run, I will present you with some thousand or so
> constructions, all of which require you to come up with an ad-hoc
> new primitive -- or let's be generous and assume that only half of them
> require new primitives.  That's still a lot of data that *isn't*
> represented yet.
> 

I think I understand your point. But the existence of the human knowledge
proofs that there must be a way out of this dilemma.
Currently I am constructing a piece of software to be used for knowledge
representation and I had the same problem you mention in the start-up
process.
The problem appears more and more as the abstraction process goes on and 
metaknowledge is introduced. 
Sometimes I needed a knowledge unit that could not be introduced, because
it depended on knowledge unit B and vice versa. So there were several hen -
egg problems.

I think there is a way to solve problems like that (I did it practically
for the software):

Metaknowledge cannot be introduced in an independent way. When a system has
metaknowledge then the information is present twice: First as the fact
itself and second as the knowledge on a higher level that the fact exists.
It is possible to introduce the fact itself (on a deeper unconscious level)
without being forced to introduce the metaknowledge at the same time. Later
we are able to add the metaknowledge as a retrievable information on a
higher level.

 
> >OK, I think it is no principal difference which syntactical
representation
> >to use. 
> >But there is one practical point: for the symbol "person" and for the
> >symbol "today"
> >we can invent many different relation structures and contexts where
these
> >symbols can be used.
> >But for the symbol "is_tired" we have a narrow set of relation parts:
> >1) the one who is tired, the subject
> >2) optionally a time
> >3) optionally a place
> >4) maybe some else, but not many.
> >
> >For the symbol "person" or "today" exist maybe some hundred relation
> >patterns or more where they can be used.
> 
> Pick up a good textbook on verb aspects, sir.  I think you're wrong
> about the apparent simplicity of verb relation patterns; "The Evolution
> of Grammar", by Bybee, Perkins, and Pagliuca lists about a hundred
> individual and mostly orthogonal verb "relation parts."
> 

I will do that. But I am not shure if you interpreted the issue the way I
did. To be shure I try to make it more clear: You speak about a hundred
individual and mostly orthogonal verb "relation parts."
I did not mean there are few "relation parts" generally for verbs. I agree,
there are many.
I just said there are few "relation parts" for the (semantic, not
syntactic) verb "is_tired". Maybe we can find more relation parts even for
that verb, but the main message can be expressed by not more than three or
four parts plus the verb "is_tired" itself.

Stefan M�hlbacher


> 	-kitten
>