1. Pseudocode revisited
- Posted by Michael Sabal <mjs at OSA.ATT.NE.JP>
Mar 28, 1998
-
Last edited Mar 29, 1998
After dodging Mike's flames and looking at David Cuny's prototype for a =
preprocessor (great macros/shortcuts, David), I thought a little =
"definition" might be warranted. I should probably start with my theory =
of Natural Language.
A computer cannot understand any human language naturally, because it =
is not human. I doesn't have the concepts associated with human =
experience. Its language consists of ports, numbers, logic, graphics, =
and a few other things. To create a natural language system means =
teaching the human language to the computer with similar expectations =
you would have of a learner of any foreign language. (Unless I told you =
or you've lived in Japan, you probably have no idea what a kotatsu is, =
or whether you put onigiri on your body, in your mouth, or in your car. =
NB: a kotatsu is a low table with a heating element used for keeping =
warm in winter -- Japan doesn't have central heating in houses; onigiri =
is a kind of rice ball, only triangular).
I tried making a natural language system using English to English =
interaction, but it didn't work very well for the very reasons cited by =
other members of the list. But now I'm using an idea originally =
proposed by the UN -- using an intermediate language that logically =
defines grammar and vocabulary sorted topically. A front-end program =
would get input from a typist (or later a microphone), or a file (the =
pseudocode file, for instance), send it to a translator program that =
converts it into the computer's language and grammar, which sends the =
result to a processor. This processor could either create Euphoria code =
from scratch (thus, it would have to be aware of Euphoria's syntax and =
command structure), or interactively communicate with the user (a true =
Turing machine). The reason I say it could never be completed, is =
because language and knowledge are constantly changing, optimizations =
are always possible, and even if the programs becomes intelligent enough =
to create new processes for new vocabulary, pruning of antiquated =
methods would have to be done manually. Or by then, a completely new =
architecture will be available to allow more optimizations, etc, etc.
Please, firestorms are completely welcome, and as soon as I come out =
from under my pile of books to finish a skeleton version, I'll zip it =
along.
Wavering on the edge of reality (but rooted in truth),
Mike
mjs at osa.att.ne.jp
http://home.att.ne.jp/gold/mjs/
