The Forthcoder Diaries -- 2011 May 27

11 views
Skip to first unread message

Mentifex

unread,
May 27, 2011, 4:46:24 PM5/27/11
to
Am 27. Mai 2011, Freitag Morgen im SB/RB.
Yesterday I gave up a splendid chance to
go down to the U District Eigerwand and
relive my undergraduate youth. As the day
wore on, I kept having second thoughts --
until I achieved some major MindForth
programming.

[etwas später, in der RBCC]
I switched from coding the JavaScript AI
http://www.scn.org/~mentifex/AiMind.html
back into MindForth because I need to
see the intermittent activation levels
quite clearly. But when I tried entering
some sample input yesterday, I saw
that there were lots of festering problems
that I had to work on before I could
concentrate on neural inhibition in
response to KB queries. I typed in
the sentence, "Birds eat bugs," and
the Forth AI could not even assign
grammatical number properly.

After I found out that I could not solve
the problem in the AudRecog module,
I quickly intuited that I needed to couch
my troubleshooting in the rubric of
expecting the input of nouns. After all,
if we want properly to assign the num(ber)
of nouns, we had better be looking for
nouns in the first place. So I decided to
insert some special code into the area
of the mind handling auditory input, but
I had to coordinate the process with the
EnParser module.

I started by using some test code in the
AudInput module to detect the letter "S"
passing by. I made some sort of "IF"
blunder that cost me a lot of time and
distress -- the program kept crashing
when I even pressed a key at the "ok"
prompt.

Then I had to put in more and more
diagnostics so as to see the visible
and invisible input stream. I let the
letter "S" set the "num(ber)" variable
to two anywhere in the input word,
with the presumption that only a
terminal "S" had to govern the
terminal setting of the "num" variable.
But I also had to block the influence of
the ASCII 32 "SPACE" character and
the ASCII 13 "CR" (carriage-return)
character, because the noun is not
recognized and stored in auditory
memory until it is followed by either
a space or a carriage-return. And
it worked! Suddenly the AI was
assigning plural number to "birds"
and "bugs" as input words. And when
I typed in "Osprey eat bugs," the AI
corrected the verb form and responded,
"OSPREY EATS BUGS".

I need such basic I/O functionality working
before I can write complex code for neural
inhibition.

So I finished writing the MFPJ AI Lab Notes
and I uploaded the 26may11A.F Mindforth
http://www.scn.org/~mentifex/mindforth.txt
to the Web. This morning I checked the
JavaScript AI and it is not yet capable
of assigning plural number to input nouns
ending in "S", so I had better update the
JavaScript code before I keep going in
MindForth. Meanwhile, today on the
powerful KCLS computers I need to
publish the latest Cyborg weblog
MFPJ entry and I need to post the text
to comp.lang.forth on Usenet.

Maybe a professional programmer
would find it ridiculously easy to do
what I did yesterday in MindForth, but
I with my meager programming skills
had a deep sense of accomplishment
and satisfaction, when I saw the AI
distinguishing between the input of
singular and plural English nouns.

Mentifex (Arthur)
--
with a tip of the hat to Che Guevara and
"The Motorcycle Diaries"

Albert van der Horst

unread,
May 28, 2011, 11:11:26 AM5/28/11
to
In article <b4086e79-1a73-4ca3...@35g2000prp.googlegroups.com>,
Mentifex <ment...@myuw.net> wrote:
<SNIP>

>After I found out that I could not solve
>the problem in the AudRecog module,
>I quickly intuited that I needed to couch
>my troubleshooting in the rubric of
>expecting the input of nouns. After all,
>if we want properly to assign the num(ber)
>of nouns, we had better be looking for
>nouns in the first place. So I decided to
>insert some special code into the area
>of the mind handling auditory input, but
>I had to coordinate the process with the
>EnParser module.

<SNIP>

If a chimpansee understands the distinction
between a verb and a noun, I'm impressed.

The way you try to teach your AI the difference, suffers
from the Kelly Bundy syndrom:
As soon as you make it understand something, something else will
have to give way.

>
>Maybe a professional programmer
>would find it ridiculously easy to do
>what I did yesterday in MindForth, but
>I with my meager programming skills
>had a deep sense of accomplishment
>and satisfaction, when I saw the AI
>distinguishing between the input of
>singular and plural English nouns.

Yes, a professional programmer probably would.

Yes, a professional designer probably would force you
to specify, then you realise you're getting nowhere and
the programming wouldn't even start.
That would be bad, probably, because you may be the frog
that escapes from the bucket of milk.
(And apparently you've nothing better to do, anyway.)

>
>Mentifex (Arthur)

Groetjes Albert

--
--
Albert van der Horst, UTRECHT,THE NETHERLANDS
Economic growth -- being exponential -- ultimately falters.
albert@spe&ar&c.xs4all.nl &=n http://home.hccnet.nl/a.w.m.van.der.horst

Reply all
Reply to author
Forward
0 new messages