Hi extropians,
Offlist I've discussed with a few extropian people a freeware NLU entity
that I'm undertaking to develop.
Some people had asked for an outline. Unfortunately, I've gone and deleted
much of my mail files; as I tend to get hundreds of fat posts per day and my
HD was getting choked.
I include a gross description here for anyone who might like to participate
in such a project; either with me or on your own... just try to keep it GPL
please; I'll attempt to answer any questions you may ask.
********************************************************
I'm doing a coarse-grained reverse-engineer of a cognitive linguistic
entity. What I've been doing is poring over the ordinary psychological
literature on (macroscopic) cognition; perception & language.
That being done; I set up a simple enough reverse engineered coarse grained
paradigm for me to write cognitive software:
Functionally, I've divided the macroscopic mind into "domains": SC; D; J; K;
L. Intradomain software process interaction is generally more pronounced
than interdomain interaction.
The SC domain houses models of the World; ontologies; objects; actions & so
forth. The higher constructs of the SC domain constitute "deep structure".
The D domain houses motivational (goal etc) & emotional weightings...it
allows context generation & "negotiation" among processes without hogging
bandwidth.
The L domain houses language... basically a compressional mechanism that
interacts seamlessy with the SC domain to facilitate extended world
modelling or machine "thought".
The J domain is basically a set of inhibitory modules that allow top-down
control of the mental processes of the machine via access from the K domain.
The K domain houses a Global Workspace; it is the highest of a hierarchy of
workspaces...; it includes an attentional system; a working memory; etc. It
is a
functional consciousness.
************************
The processes run independently and should talk to each other via shared
memory. Such a machine should run OK on Linux SMP on off-the-shelf or
"commodity" machines of the near future.
************************
This is just a hobby, and everything I do will be GPL. What I'm doing just
now is some component linguistic systems of the L domain... (i) simple
canned NLG & template stuff... mostly ALICEbot like stuff (ii) Sentence
Generators (iii) heuristic; algorithmic and pattern matching parsers (iv)
simple deep structure models for the SC domain.
Sundays only... I'm not yet retired.
*************************
Since there'd be adjudication for the dominance of any one process - some
sort of winner take all setup - and the processes are reading from
blackboards and such - you can just drop a process in. The system
theoretically scales well.
Each process has a lot of autonomy... but isn't mobile (...save for the
SMP - ha ha).
I can handle the computational linguistics but I'm almost blind
(toxoplasmosis at 2; and leptospirosis at 39 - fun huh?) so the coding is
slow.
***************************
Again, this is doable..I'm writing it in C and for the expected linux
server-quality multiprocessor machines of the next 3 years or so... I'd
appreciate input & actual coding help from anyone who would be crazy enough
to want to do this sort of thing.
Thanks (I hope this falls within the scope of this mailing list...),
alain
This archive was generated by hypermail 2b29 : Mon Oct 02 2000 - 17:35:05 MDT