| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Another thought. Sentences or utterances in NLP
could be keyed and used as indexes to the OAV keys
so that one complete sentence would index all the
OAV assertions or inferences it produces.
But how can events be represented? Normally, an
event might be
<actionkey><activity><start/stop/hold><timestamp><OAVKey>
so that all OAV assertions or inferences related to
a single action can be grouped.
But the index structure that's starting to unfold
would be very complicated and internested with
the sentence structure indexes. Sentences about
events and events on sentences could cause infinite
recursion if not unravelled in some clear way.
Any further thoughts about effective representations?
Thanks,
Rich
--------------------------------------------------
This is an excellent presentation of the concepts
underlying human language. It also points out a
simple concept for storing language statements
is simply <key><object><attribute><value><author>.
So all language statements can be made in a five
column relation.
However, searching such a relation is an enormous
computing task. So clearly there is a need for
more comprehensive descriptions of the base relation.
These "index" relations can be nearly any form, just
so long as they refer to the <key> values of the
statements.
But this leads to the next question; how can we
organize the index relations? The answer to that
question tends to defeat the "simplification" of
putting all statements into the single base relation.
Any comments on this issue?
Thanks,
Rich
> Stephen,
>
> I managed to dig out an ancient version 4.0 of IE, and read
> your slides. I think they make some good points:
>
> http://www.downes.ca/files/one_standard_files/frame.htm
>
> Your concluding slide, which I very much agree with, presents
> a fundamental challenge to any attempt to develop a standard
> upper ontology:
>
> > Objects are best described using multiple vocabularies.
> >
> > There is no way to determine which vocabulary will be relevant
> > to either an author or a user of a given object.
> >
> > Trying to stipulate a canonical vocabulary a priori needlessly
> > reduces the effectiveness of a system of communication.
>
> This conclusion does not mean that it is impossible to have
> a sharable upper ontology, but it does imply that any such
> upper level must be capable of supporting an open-ended range
> of vocabulary for all possible uses and all possible purposes.
>
> As I said in earlier notes, the question of how much can be
> usefully shared is an empirical issue. It is probably more
> than just a single node at the top, but how much more is
> unknown.
>
> Whether a useful amount can be shared before we have to
> start admitting incompatible (i.e., mutually inconsistent)
> alternative modules is also unknown.
>
> I believe that it is probably possible to accommodate some
> sharing together with some system for managing mutually
> inconsistent alternatives. But again, this is an empirical
> question that cannot be answered a priori.
>
> What this means is that we cannot just pick an ontology and
> legislate it as an official standard. There is still a lot
> of research to be done before we can say exactly how much
> can be shared and how we can best accommodate inconsistent
> alternatives.
>
> John Sowa
>
> PS: I'll be doing some traveling for the next week or so,
> and this might be the last note I'll be able submit on this
> topic for a while.
>