Copy Link
Add to Bookmark
Report
AIList Digest Volume 1 Issue 002
AIList Digest Sunday, 1 May 1983 Volume 1 : Issue 2
Today's Topics:
New BBoards
The T Programming Language
Parallel Nonnumeric Algorithms
Pattern Recognition
Standardized Correspondence
Alternate Distribution of AIlist
Facetia
----------------------------------------------------------------------
Date: Sat 30 Apr 83 17:17:00-PDT
From: AIList <AIList-Request@SRI-AI.ARPA>
Subject: New BBoards
The following new BBoards and remailing lists have been created:
AIList-BBOARD@RUTGERS
NYU-AIList@NYU
"XeroxAIList^.PA"@PARC-MAXC
UCI-AIList.UCI@Rand-Relay
I am told that the PARC list has 94 members. As yet there is no
BBoard at CMU (36 members); someone might want to establish one. I
will publish an updated list of hosts after the membership settles
down.
-- Ken Laws
------------------------------
Date: Tue, 26 Apr 83 18:26:42 EDT
From: John O'Donnell <Odonnell@YALE.ARPA>
Subject: The T Programming Language
I am pleased to announce the availability of our implementation of the
T programming language for the VAX under the Unix (4.xBSD) and VMS
(3.x) operating systems and for the Apollo Domain workstation.
T is a new dialect of Lisp comparable in power to other recent
dialects such as Lisp Machine Lisp and Common Lisp, but fundamentally
more similar in spirit to Scheme than to traditional Lisps.
The current system, version 2, is in production use at Yale and
elsewhere, in AI and systems research and in education. A number of
large programs have been built in T, and the implementation is
acceptably stable and robust. Yale and Harvard successfully taught
undergraduate courses this semester in T (Harvard used Sussman and
Abelson's 6.001 course). Much work remains to be done; we are
currently expanding the programming environment and improving
performance. Our next release is planned for sometime this fall.
Please contact me directly if you're interested in getting the
distribution.
John O'Donnell
Department of Computer Science
Box 2158 Yale Station
New Haven CT 06520
(203) 432-4666
ODonnell@Yale
...decvax!yale-comix!odonnell
------------------------------
Date: Thu 28 Apr 83 14:40:26-PDT
From: David Rogers <DRogers@SUMEX-AIM.ARPA>
Subject: parallel non-numeric algorithms
Part of my Ph.D. work has been in parallel processing
algorithms in graph theory (unfortunately, no hardware is currently
available for an implementation, but that only makes the excursion a
little less satisfying). Specifically, I have been simulating the
performance of an algorithm for the utilization of parallel processing
in speeding up the common subgraph search problem. Commonly, this
problem involves finding all sufficiently large subgraphs common to
two given graphs. No efficient algorithm exists for doing this
search.
I know that several AI groups are working on parallel
processing in AI, but have not found any discussion involving graph
searching techniques. The bias in parallel processing has been toward
numerical algorithms and the use of array processors; I figured that
there MUST be some AI group working at parallel processing in a
non-numerical field such as graph searching. I would like to hear
from anyone who knows of such or similar work.
By the way, I had heard that workers had had 'problems' with
the parallel LISP machines, but have not been able to pin anyone down
exactly as to the nature or extent of these problems. Anyone know
exactly what was discovered in those researches?
Thanks--
David Rogers DROGERS@SUMEX-AIM.ARPA
------------------------------
Date: Fri 29 Apr 83 08:35:25-PDT
From: Ken Laws <Laws@SRI-AI.ARPA>
Subject: Pattern Recognition
PR People should take note of "Candide's Practical Principles of
Experimental Pattern Recognition", by George Nagy, in the March issue
of IEEE PAMI. I particularly like
... any feature may be presumed to be normally
distributed if its mean and variance can be
estimated from its empirically observed distribution.
and
... adapting the classifier to the test set is
superior to adaptation on the training set.
-- Ken Laws
------------------------------
Date: 30 April 1983 04:00 EDT
From: Steven A. Swernofsky <SASW @ MIT-MC>
Subject: Standardized Correspondence
Rather than distributing the same software to every site, it would
make more sense to develop a machine-to-machine language which would
express (dunning-letter-7 formality-level-3 car-payment-overdue
$127.38) in an easily parseable form. English is complex, redundant,
and vague. Is there any reason why we can't design a language which
is simple, efficient, and precise? It would be awful for
(intelligent) people, but great for (stupid) machines.
-- Steve
[If the parsing and synthesis functions were common, the software
might be compiled into hardware; if rare, it might be accessed
remotely over a network. I doubt that software storage requirements
will be a problem for long.
There have been attempts at developing simpler natural languages.
One idea is to structure the language so that any idea can only
be expressed in one canonical form (DuckSpeak, Basic English,
controlled-vocabulary English as taught in our grade schools).
The other idea is to allow any semantic term to fill any syntactic
slot (sign language, Loglan).
Langauges of the first type present difficulties because of the
overloading of words (e.g., "get" in English). This can be avoided
in limited domains such as repair manuals, but for general usage
something like Roger Shank's canonical forms would be needed.
I don't know what computational difficulties are presented by
languages of the second type. If the Whorfian hypothesis is correct,
more ideas can be "thought", which may lead to greater complexity.
On the other hand, the algorithm needn't keep track of awkward or
stereotyped methods of expressing basically simple concepts. ("I
greened my house", or what is the past tense of "beware"?)
I trust that computational difficulties can be overcome. The
greatest problem in achieving user acceptance of parsed transmissions
may be that resynthesis will generate a paraphrase, or corrected
version, of the original. Humans tend to be sentimental about their
own syntactic constructs, even down to where the lines are broken.
-- KIL ]
------------------------------
Date: Thu 28 Apr 83 00:52:52-PDT
From: Dan Dolata <DOLATA@SUMEX-AIM.ARPA>
Subject: Alternate distribution of AIlist
I am moving to Sweden soon, and while I will be able to touch back to
my home base via international-net occasionally, the long ditance
rates make it prohibitive to try to read any large number of lines
each day. I was wondering if it might be possible to set up some ort
of system where AIlist could be spooled onto small tapes or floppies
monthly, and mailed to people who are away from the net? Of course, I
would be happy to pay for mailing costs, and would be happy to buy the
person who did the grunt work a nice meal when I got back from Europe
(or in Karlsruhe during IJCAI).
Of course, if it became neccesary to charge $ because you had to hire
a person to mount the media, I would be happy to subscribe!
Thanks for your time
Dan [dolata@sumex]
[I'm afraid that I haven't the resources to oblige. I suggest that
printed copies be sent, providing that doesn't violate any technology
export laws. Dan would like to know if others are interested in
getting or providing machine-readable copies. -- KIL]
------------------------------
Date: Fri 29 Apr 83 09:02:22-PDT
From: AIList <AIList-Request@SRI-AI>
Subject: Facetia
I hope everyone kept V1 #1. Someday it may be as valuable as the
first edition of Superman comics.
-- Ken Laws
------------------------------
End of AIList Digest
********************
