Brian T. Rice
Fri, 07 Jan 2000 07:47:57 -0800
As always, I welcome flames.
At 02:02 AM 1/8/00 +1100, David Vennik wrote:
>Well, people, I have been doing a lot of thinking about language in the
>last few weeks.
>If I may quote something which has become the inspiration for me to keep
>trying to figure all this out:
>"Paradoxically, an exact and detailed specification of any function
>will, in principle, allow a computer to perform it." Pete Carrol
What does this mean? Have you looked at the lambda-calculus? It doesn't
seem so paradoxical. Yet there are programming systems that can't be
entirely specified by lambdas themselves. Just recently a Tunes member
talked about GEB without asking anyone on irc for feedback or reading
anything, simply because he was intuitively inspired and fealt like he was
worth listening to. To make an analogy, just because you see towering
mountains near the horizon doesn't mean you can get there without looking
at the ground beneath your feet. You have to take definite steps in that
direction and learn from other systems. I just don't see how GEB or
Psybermajick helps you or us.
>I have thought from the very beginning of my recent new foray into the
>field of OS designing, that some sort of communication system would have
>to be at the heart of any OS in order to make it easily able to handle
>many of the desired features that are enumerated on the TUNES web site.
"Some sort of"? What kind of discussion do you expect to result from this?
I can't disagree because it's much too vague. What do you mean by
communication system? How do you distinguish it from the message-passing
ideas of microkernels? Have you even looked at a "kewrnel-less" design,
such as the Secure Language (no-)Kernel project? What would it communicate?
What entities would do the communicating? Are you familiar with the
smalltalk / self paradigm, where all computation is done in terms of
objects sending messages to each other? These are all important ideas to
Do you expect to be able to code this in c, or do you understand why the
tunes project seeks such a thing with high-level languages?
>What we need is a program which processes 'natural' language really.I
>have now decided that the only thing preventing a computer from
>understanding human language is the lack of a sufficient model (an exact
>and detailed specification) of how it works. We can all describe how we
>process information in general terms, but in all the gaps between the
>words we describe lurks details which we may not reveal until we try to
>make a model, and discover we can't find the right words to describe it.
>I have started on this, and I have only sketched out the language
>database structure and functions relating to matching words with words
>in the meme database (as I am calling it) Further off in the distance I
>can see other things which I presently only have vague descriptions of,
>like the procedure required to recognise a 'legal' sentence, and error
>correction (typos and grammar errors).
Are you even familiar with Godel's incompleteness and undecidability
theorems? What about computational complexity? I've seen a lot of people
brainstorm there ideas on this list without any kind of idea about what's
out there and what's been done. There are lots of projects out there that
try to capture one or several aspects of natural language, but Tunes
members as a whole don't deal with those.
I've looked really extensively into this and related issues, and there is a
developing consensus in those fields that no single coherent model can
completely describe natural language and it's dynamic nature, even in
theory. If, however, you want a tiny fragment of english grammar, there are
plenty of vastly incomplete formalisms that people have been working on for
decades. How can you expect to bypass all of their work without at least
studying some of it? There is also the aspect that statement parsing can be
undecidable or intractable in general.
>I have only really just started on this, but I feel very strongly that
>this can be done, if we can only describe how it happens well enough. If
>anyone wants to see what I am up to with it I can send them the latest
>version of what I am doing.
It's great that you're working on your ideas, but please be open to ideas
that others have had so you can build from their mistakes and successes.
Otherwise, you will (I guarantee) stumble into problems that others have
addressed and studied without any sort of vision.
>My brain is boiling over with ideas relating to this, and I have trouble
>starting talking about it, because I have trouble stopping my
>elaborations. And I think they would be better placed in the description
>I am developing at the moment, in logical order, rather than the lateral
>jumping order that I tend to write in.
I'm asking you to please do some reading and studying so your ideas don't
go to waste, which I would terribly hate to see happen, like it happens
with so many others. There's so much out there that the usual person like
yourself hasn't been exposed to that would benefit anyone in the field.
Join some tunes members on IRC at irc.openprojects.net, channel #tunes, and
we can share information in real-time about various projects, with url's to
all sorts of explanations and tutorials on the variosu ideas in computing
that concern tunes and general os developers with ideas like you.
>Nice to see you all back after the crazy end of year stuff.