Ravi and Pat C., (01)
RS> Are these semantic distances similar to Mahalanobis
> distances of Multivariate statistics? (02)
That depends on what you mean by "similar". The specific
algorithms I was thinking of are very different. But many
measures have been used to define some sort of "distance",
infinitely many have not yet been explored, and there is no
assurance that any single method could be considered "best"
for all purposes. (03)
RS> We talked about "picture theory of theory meaning" and one
> of the threads in our conversation was that understanding is
> often through visualization rather through language. (04)
I certainly agree with that point. I believe that everybody
uses visualization (consciously or unconsciously) to some extent
and that some people depend on it much more than others. I
would call visualization a kind of perceptual method, but
I also believe that there is a very close relationship between
percepts and concepts. Very likely, concepts are a kind of
stylized percepts that have evolved from a common source. (05)
For a person who uses visualization to an extreme, I suggest
the autistic woman Temple Grandin, who managed to earn a PhD
in animal husbandry despite her difficulty with abstract thought.
You can find much more about her, her modes of thought, and
citations to her books (which I recommend) on the web. (06)
RS> I have often wondered as to how - for few limited phenomena
> that I can explain in multiple languages - limited and varied
> as the translation is - it draws from conceptual rather than
> textual understanding of the phenomena or subjects under
> consideration. Is this not cognitive reasoning or concept
> (understanding)? (07)
I would say that human methods are primarily conceptual and
that most current computer methods are primarily textual.
But since people generate texts by conceptual or perceptual
means, the texts that appear on the WWW contain a great deal
of information about the unobservable underlying patterns. (08)
PC> What I meant was that accessing Web information without
> any accounting for the likely reliability of specific pages
> is a very risky enterprise, and it requires human-level
> language interpretation to determine what is reliable. (09)
I completely agree with the first clause of that sentence,
but I would add that there are many computational methods
for the second. The Google page-rank algorithm, for example,
was very good before web designers found ways of gaming the
system. But Google still has good methods for selecting
news stories, and much, much more can be done. (010)
PC> But the issue that I was addressing was whether any form
> of automatic translation can be interpreted as "understanding"
> (a word you used) at anything close to the human level... (011)
I used the word "understanding" in an informal sense, and I
admit that it's too vague to be helpful as a technical term. (012)
PC> Some day, statistical translation may come close, if it
> is possible to tag many texts with precise semantic categories
> and have those texts mapped to semantic interpretations. I
> just expect rules-based language understanding to get there first. (013)
The advantage of case-based methods is that their capabilities
automatically increase with every new case that is added. And if
you derive those cases from the WWW, you get many more each day. (014)
I regard rules (including all deductive systems) as fossilized
abstractions from cases. They're valuable for applications
that change very slowly over time and have a limited number of
of easily characterizable exceptions. (015)
My recommendation is to adopt a case-based method as *primary*
and to use rules to handle those cases that fall into categories
that have been thoroughly analyzed. But there is a lot more
to say about this point. (016)
Following is an article that compares computer and human methods
in cognitive science. I wrote it in 2005, but it is still close
to what I believe today: (017)
http://www.jfsowa.com/pubs/cogcat.htm
Categorization in Cognitive Computer Science (018)
If I were rewriting it today, I wouldn't delete anything from it,
but I would put much more emphasis on learning. (019)
John (020)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (021)
|