Re: SUO: Program Semantics
On Fri, 29 Mar 2002, John F. Sowa wrote:
> 1. The problems of knowledge soup (Ch. 6 of my 2000 book or Ch. 7
> of my 1984 book) are fundamental to the way people think, and
> they must be accommodated by any cognitive agent or reasoning
> system that has any hope of attaining the flexibility of human
> reasoning (or even a rough approximation to it).
clip
I like that..."knowledge soup". My work, over the years includes a decade
as a psychologist in a mental institution. People can say the darndest
things. They can say almost anything under a given set of circumstances
(stimulus conditions). There's your "flexibility" in human language.
Put a word or symbol on the computer screen. Somebody, somewhere, will
follow it with any other word or symbol.
So what can we say about the search for an "algorithm" which will enable
us to write NLP? IMO it is to be found in the norms of language usage.
This is sort of an "outside-in" approach rather than an inside-out
approach which seems to dominate the field at present. The inside-out
approach works fine for all kinds of table games. Once you enter the game,
just follow the rules and all will be well. Don't know the game? Figure it
out. Find the internal algorithm. But the game of human language usage
doesn't have this kind of algorithm. It has maximum "flexiblity", so we
look at the outside conditions, and tables of norms become our algorithm.
In clinical phil-psych that is what we are constantly dealing
with...tables of norms. If the patient is saying something which is deemed
to be inappropriate or not normal, we look to tables of norms, whether
they are available from standardized tests or surmised.
According to "I,Robot" by Asimov, the world's first robo-psychologist
graduates in 2009 so we have to work fast on figuring out how to socialize
our robots, like Etcetera, <http://www.lightwaverobotics.com>. I would
suggest two approaches to NLP for Etcetera and many more of his kind.
(1) Start at around a two year level where we have a pretty good
understanding of the norms for verbal responses in humans and the
corresponding stimulus conditions/circumstances; we work our way up the
developmental scale. (2) Similarly we can start with the level of mental
handicap, say moderate retardation (IQ 40-55 and corresponding SQ) at
which we have a good understanding of verbal norms and work our way up the
IQ/SQ scales.
OTOH, if we want to program the robot to be a master logician, let's do
that explicitly. Write the programs so that it will be proficient in all
forms of arithmetic, logic and mathematics. But conversational ability is
indeed a "knowledge soup" with maximum and almost unlimited,"flexibility".
FWP/POC