3.562 supercomputing the humanities, cont. (142)

Willard McCarty (MCCARTY@vm.epas.utoronto.ca)
Tue, 10 Oct 89 20:56:54 EDT

Humanist Discussion Group, Vol. 3, No. 562. Tuesday, 10 Oct 1989.

(1) Date: Mon, 9 Oct 89 17:26:44 EDT (7 lines)
From: Daniel Boyarin <BOYARIN@TAUNIVM>
Subject: Re: 3.550 supercomputing the humanities, cont. (47)

(2) Date: Mon, 9 Oct 89 22:24:53 EDT (50 lines)
From: "A. Ralph Papakhian" <PAPAKHI@IUBVM>
Subject: Re: 3.554 supercomputing the humanities, cont. (92)

(3) Date: Tue, 10 Oct 89 08:22:25 EDT (61 lines)
From: amsler@flash.bellcore.com (Robert A Amsler)
Subject: supercomputing jousting

(1) --------------------------------------------------------------------
Date: Mon, 9 Oct 89 17:26:44 EDT
From: Daniel Boyarin <BOYARIN@TAUNIVM>
Subject: Re: 3.550 supercomputing the humanities, cont. (47)

How about automatic collating of variations in fifty manuscripts of
a text? with many passes to make sure that mistakes weren't made because
of different ordering of materials or large skips etc.
(2) --------------------------------------------------------------58----
Date: Mon, 9 Oct 89 22:24:53 EDT
From: "A. Ralph Papakhian" <PAPAKHI@IUBVM>
Subject: Re: 3.554 supercomputing the humanities, cont. (92)

On Mon, 9 Oct 89 21:44:37 EDT you said:
>Humanist Discussion Group, Vol. 3, No. 554. Monday, 9 Oct 1989.
>Date: Mon, 9 Oct 89 12:20:00 EDT
>From: "Vicky A. Walsh" <IMD7VAW@OAC.UCLA.EDU>
>Subject: Re: 3.543 supercomputing the humanities, cont. (64)
>I can't believe how nervous some of you are about supercomputing. It is only
>a bigger faster box. U of Minn has a concordance program that runs on the
>Cray, e-texts can be easily transported there as needed, would you really
>rather wait 45 minutes for TLG output as Bob Kraft recently mentioned or
>would you prefer truly interactive access? There are data base programs
>now available on the Cray and probably soon a lot more. I know if you just
>ask what applications are available from the center consultants you will not
>likely get a satisfactory answer; they rarely hire people who have any
>knowledge of text based applications; so you shouldn't give up. Ask them
>to contact Cray or other universities to see what's being done. I know
>there are not really easy to use solutions there yet, but there won't be
>if we don't keep asking.
>Vicky Walsh, UCLA

Vicky Walsh's comments are extremely significant. Somehow it's
tremendously important for physicists to get computing results quickly
but humanists must be patient. A follow up is the still inadequate
support for multi-lingual text databases on mainframes (including
library databases and catalogs). One factor is the market place (too
few humanists are INSISTING on multi-lingual capabilities and an
insufficient market exists for paying for development costs). Another
factor is the simple historical fact that computers (especially
mainframes) are attuned to number crunching instead of text crunching.
This is NOT and CANNOT be a technical problem (many similar technical
problems have been solved). We continue to face a political/economic
problem. If HUMANISTS' needs, in terms of text processing, were
recognized as being as significant as say vector processing for
physicists, we would be seeing products providing these necessary
services. Trouble is that powers at be do not see thorns and
haceks and breves, etc. and word and text processing as needy
causes (not to mention music or art images). Are they worthy
causes? Or should we count on a SUN with six users adequately
processing one database?

Cordially, *****
**** *** **** MUSIC
** *** **
A. Ralph Papakhian, Music Library ** ******* ** LIBRARY
Indiana University, Bloomington, IN 47405 ***
(812) 855-2970 *****
(3) --------------------------------------------------------------75----
Date: Tue, 10 Oct 89 08:22:25 EDT
From: amsler@flash.bellcore.com (Robert A Amsler)
Subject: supercomputing jousting

There is an argument I recall that makes me question
whether anyone should be allowed to say,

``These machines are, in my opinion, best used for very
complex calculations representing physical processes.''

The argument is that `best used' is a subjective judgement based on
prior bias. Just because physical scientists have numeric problems
that need supercomputing capabilities, they have decided that people
without such a requirement ought not to use their toys.

This is equivalent to saying that the poor wouldn't know what to
do with haute cuisine, penthouse suites and yachts and that they
shouldn't be concerned about such things anyway.

I remember the Sociology Dept. at the University of Texas issuing a
memo which chided students who were using the computers to type in
their dissertations for misusing valuable scientific computing
equipment for non-scientific work. It began with the sentence
``Somehow some individuals have discovered how to use the computer to
type in text.'' Well... that was before someone invented word
processing and a whole new market opened up.

I also remember a story of the Cray at Bell Labs that remained idle
because nobody could afford to pay for the computing time on it. A
10-megabuck machine sitting there with nothing to do because someone
set up a charging algorithm to price its valuable computing cycles so
high that in turn NOBODY had calculations that were THAT much in need
of supercomputing cycles to perform.

There are two reasons people create supercomputers. The first is
that they know of classes of problems in the world which they believe
they know how to compute (i.e. software exists) and for which the
supercomputer (a new hardware design) would do the tasks far faster
than ordinary computers. The second is that they know of techniqies
for building hardware that will do something really really fast and
leave to the world the task of discovering whether this hardware will
spur software writers to come up with algorithms to make use of its
new capabilities.

This interaction between hardware and software is a vital force in
computer science. It leads to discoveries of deficiencies in existing
hardware that can then be corrected by the engineers.

In my experience humanists tend to think up tasks for computers which
cover the whole spectrum between trivial and impossible. Scientists,
being more practical, tend to think up extrapolations of their
existing techniques and only ask designers to make faster
versions of their older computers.
(Incidentally, whether you consider co-occurrences an I/O problem of
not depends on how you program the machine. On a Connection Machine
you'd put one word in each processor and do everything in parallel.
The trick is maximal use of whatever parallelism one has available
on whatever type of supercomputer.)