home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Cream of the Crop 26
/
Cream of the Crop 26.iso
/
educate
/
trutran2.zip
/
ARTICLE4
< prev
next >
Wrap
Text File
|
1996-12-14
|
46KB
|
844 lines
*** PRESS ANY KEY TO SEE THE NEXT SCREEN ***
If you wish to print this article or view it in its
entirety, please load it into your word processor
as ARTICLE4.
*********************************
* For an overview of these *
* articles, please first read *
* the file ARTICLE0.SEE *
*********************************
[The following paper was presented at the 1995 ATA
Conference in Nashville and published in the Pro-
ceedings. It was accompanied by a number of illustra-
tions, most of which cannot be provided in this elec-
tronic version, and even the published version
contained only four of them. Or rather, although
they could be provided, the work of scanning and
reediting fine line-drawings would take too long.
Wherever necessary in the text, sections in brackets
attempt to describe the missing illustrations. This paper
has been uploaded with permission from Information
Today, publishers of the ATA Proceedings. Anyone who
has a desperate urge to see the published drawings can
order the Proceedings from them. Their address is
Information Today, Inc., Medford, New Jersey, 08055.]
******************************************
Perfect MT: Logical Certainty
Or Recurrent Self-Delusion?
(Ten Fragments and Three Contentions
Connected by a Single Theme)
*******************************************
By Alex Gross
Cross-Cultural Research Projects, ATA
Keywords: Machine Translation, Human Translation,
History of Computers, Knowledge Interfaces, Data Bases,
Limits of Science
ABSTRACT: The speaker will discuss Machine Translation
as one of a number of language- and knowledge-
organizing devices that have developed over time and in
various cultures. His chief interest lies in the
assertions of MT's pioneers--though still echoed by
some specialists today--that perfect or near-perfect
translation by computers can in fact be accomplished.
Using overhead slides, he will move from early Chinese
to classical Arabic to medieval Christian beliefs about
the nature of knowledge and language along with various
theories attempting to explain or control them. He
will not neglect Swift's satirical Academy of Lagado as
he moves closer to the computer age and attempts to
show at least a few parallels between modern science
and past systems of knowledge. Problems with various
types of knowledge interfaces will be considered,
including those for computers, and the speaker will
conclude with some specific remarks about where MT is
currently headed and how translators can best
accommodate themselves to the kinds of work it
does best.
I want to begin this paper with the all-important two-
part proviso that has to accompany any treatment of MT
at a translators' Conference. 1) No one opposes MT
where it works, and 2) MT works quite well for those
tasks where it is suitable. Similarly, there are two
extremes which we must avoid at all costs:
a) MT is useless--it will never truly work; and
b) MT is inevitable, it will soon take over all
of translation, and we will all be out of jobs.
Now that I've made this abundantly clear, I'd like to
discuss the question that I find most truly
interesting. How is it possible that for nearly fifty
years many perfectly intelligent and even ingenious
engineers, linguists, and researchers could have ever
supposed that something resembling perfect MT could
ever come into being? And how is it possible, I might
add, that quite a few of these authorities have still
not totally climbed down from this position and
continue to believe that it will sooner or later become
possible to put most of the people in this room out of
work? And, finally, how does it come about that a
number of academic fields and even professional
organizations are still aimed at precisely this goal?
How has it become possible for any of this to take
place and for any of these people to go on believing
this?
To answer this multi-part question I want to take you
on a private tour through a number of artifacts,
images, or ideas from the past and/or from other
cultures. I also want to describe a few encounters and
episodes I've witnessed in our own age and culture.
I'm calling all of these collectively "fragments," and
I hope that by the end of this tour you will begin to
share my view of the connections between them and see
how they relate to MT today. What most of these
fragments have in common is that they have something to
do with a means of organizing either language or human
knowledge or reality itself--or all three together. In
the midst of our tour I will also introduce three
outrageous contentions, which I mean to defend quite
seriously.
The first such artifact [overhead slide 1] is a Chinese
geomantic compass, called a fengshui luopan or,
literally, a wind and water compass. As you can see,
it's a set of concentric circles and purports to
demonstrate how individual human beings fit into the
larger pattern of the universe. There is no way I can
really explain fengshui without all of you assuming
that it must be some form of astrology, even though I
know for a certainty that it covers far more than what
is properly considered as astrology. This wheel or
compass provides a metaphysical, medical, and even a
methodological guide to the classical Chinese universe.
But what does this have to do with computers, you may
ask. If we look at the next slide [slide 2], the
answer should become clear. Once again we see a wheel
or a set of wheels, but instead of Chinese characters
we find English words, actually a translation from the
Latin. This set of wheels, unlike the Chinese one, is
credited by those in the field with providing the first
crucial step towards the computer. The reason for this
is that its various wheels move independently of one
another, while the fengshui luopan was a single rigid
piece. This is what patent lawyers call "the inventive
step" and possibly defines that point where the East
ends and the West begins, even though it may yet turn
out that its inventor based it on an Arab original.
This inventor, whose life bridged the thirteenth and
fourteenth centuries, was a truly great linguist and
even the founder of language schools. He managed to be
both a scholar and a popularizer, both a scientist in
terms of his own age and a fanatical Christian
apologist. And because he wrote in both Catalan and
Arabic, for four centuries his works were known in both
Christian and Muslim lands. His name in his native
language is El Beat Ramon Llull, or the Blessed Raymond
Llull, "Blessed" being the title one step below "Saint"
in the Catholic hierarchy. He was also deeply
interested in medical studies.
What do these Llullian wheels do? They supposedly
illustrate the attributes of God according to various
human and divine categories. To some extent they
overlap on the territory of the Chinese Fengshui
compass. But because the wheels move independently,
what we are also looking at here is an early example of
a relational data base, or at least of the hardware or
software shell for such a data base.
This invention, as vapid and metaphysical as it may
seem, pointed the way to the scientific age, which I
think we've all heard about, and so I'll skip to the
year 1726, when Jonathan Swift was busy sending up this
entire movement in Gulliver's Travels. Here we see
[slide 3] the Frame or Engine for Improving Speculative
Knowledge from the Academy of Lagado. Swift describes
it as follows:
[This illustration shows a remarkably nonsensical-
looking gridded square with 20 lines criss-crossing
horizontally and vertically to form 400 little squares.
Within each square is a silly looking "foreign"
character, perhaps a cross between Arabic and Siamese.
Around all four sides are representations of little
"handles" at each level of the grid.]
The first Professor I saw was in a very large
Room, with Forty Pupils about him. After
Salutation, observing me to look earnestly upon a
Frame, which took up the greatest Part of both the
Length and Breadth of the Room; he said, perhaps I
might wonder to see him employed in a Project for
improving speculative Knowledge by practical and
mechanical Operations. But the World would soon
be sensible of its Usefulness; and he flattered
himself, that a more noble exalted Thought never
sprang in any other Man's Head. Everyone knew how
laborious the usual Method is of attaining to Arts
and Sciences; whereas by his Contrivance, the most
ignorant Person at a reasonable Charge, and with a
little bodily Labour, may write Books in
Philosophy, Poetry, Politicks, Law, Mathematicks
and Theology, without the least Assistance from
Genius or Study. He then led me to the Frame,
about the Sides whereof all his Pupils stood in
Ranks. It was Twenty Foot Square, placed in the
Middle of the Room. The Superficies was composed
of several Bits of Wood, about the Bigness of a
Dye, but some larger than others. They were all
linked together by slender Wires. These Bits of
Wood were covered on every Square with Papers
pasted on them; and on these Papers were written
all the Words of their Language in their several
Moods, Tenses. and Declensions, but without any
Order. The Professor then desired me to observe,
for he was going to set his Engine at work. The
Pupils at his Command took each of them hold of an
Iron Handle, whereof there were Forty fixed round
the Edges of the Frame; and giving them a sudden
Turn, the whole Disposition of the Words was
entirely changed. He then commanded Six and
Thirty of the Lads to read the several Lines
softly as they appeared upon the Frame; and where
they found three or four Words together that might
make Part of a Sentence, they dictated to the four
remaining Boys, who were Scribes. This Work was
repeated three or four Times, and at every Turn
the Engine was so contrived, that the Words
shifted into new Places, as the square Bits of
Wood moved upside down.
Six Hours a-Day the young Students were employed
in this Labour; and the Professor shewed me
several Volumes in large Folio already collected,
of broken Sentences, which he intended to piece
together; and out of those rich Materials to give
the World a compleat Body of all Arts and
Sciences; which however might be still improved
and much expedited, if the Publick would raise a
Fund for making and employing five Hundred such
Frames in Lagado, and oblige the Managers to
contribute in common their several Collections.
He assured me, that this Invention had employed
all his Thoughts from his Youth; that he had
emptyed the whole Vocabulary into his Frame, and
made the strictest Computation of the general
Proportion there is in Books between the Numbers
of Particles, Nouns, and Verbs, and other Parts of
Speech." (1)
In his description of the Academy of Lagado, Swift was
of course exercising his satirist's right to
exaggerate. Many of the experiments he singles out are
totally ridiculous, though perhaps no more so than some
experiments undertaken today. He was in fact
satirizing all of Science, and this may be the reason
why this episode has been cut from some abridged
versions of Gulliver's Travels.
How mistaken Swift could be is proven by my next
exhibit, a perfectly successful example of language-,
knowledge-, and reality-management, dating from the
year 1852. Once again there is a medical aspect,
because its inventor was not only a doctor but served
as Secretary and Vice-President of London's Medico-
Chirurgical Society and later as a member and Secretary
of the Royal Society. He also invented a slide rule,
wrote a regular column on chess problems, and even
created the first inexpensive chessboard. Perhaps most
intriguingly--he worked long and hard during the 1840s
on the invention of a calculating machine. But work on
his most important contribution to humanity began only
when he was in his seventy-first year. This invention
was so successful that we still use it in one version
or another today, 147 years after its creation. I used
it in writing this paper.
It is of course [slide 4, showing the frontispiece of
an early edition and a portrait of the author] the
"Thesaurus of English Words and Phrases" by Dr. Peter
Mark Roget. In the preface Roget advanced both
philosophical and practical claims for his work. Here
once again we encounter the notion that an invention
capable of organizing language can also affect human
knowledge, perhaps everyday reality itself. Roget
argued that anyone who used his Thesaurus would become
more persuasive in argument and hence better able to
influence events.
There's one important feature about his invention that
I would ask you to note, as it has some relevance to
MT. With all his interest in chess, slide-rules, and
automatic calculators, Roget never suggested that his
Thesaurus itself was automatic or that it could be
consulted mechanically by a non-human user. The whole
point of the Thesaurus is that would-be users must
examine the various lists of words and make their own
choices, based on a specific task and context. But
even today not everyone gets the "point" of a
thesaurus: there are those--mostly non-writers or
others lacking verbal sophistication--who imagine it is
a kind of failed dictionary. They want only the word
that they imagine they want and may actually blame the
Thesaurus for making them choose. I can't help
wondering if some of their descendants have not now
wandered into MT.
With Peter Mark Roget we are clearly only a few steps
from his contemporary Charles Babbage, who was already
at work--with aid from Lord Byron's daughter Duchess
Ada--building the "Analytical Engine," which had it
been completed would have qualified as a true prototype
of the computer.
At this point, I think my next fragment can be none
other than Alan Turing's famous statement in his paper
Intelligent Machinery (2). Here he foresaw the use of
"television cameras, microphones, loudspeakers, wheels
and handling servo-mechanisms" as well as some sort of
"electric brain." It would be capable of [slide 5,
which shows the above quotation, plus the indented
section below]:
"(i) Various games...
(ii) The learning of languages
(iii) Translation of languages (author's emphasis)
(iv) Cryptography
(v) Mathematics"
Now we are getting directly into computers, and my next
example is going to be rather contentious or will
appear as such to some people. It is in fact a set of
three connected arguments I made two years ago at a New
York Circle panel on MT. It goes as follows:
1) There will never, ever be a perfect computer
interface that works satisfactorily for all purposes
and for all people.
2) There will never, ever be a perfect hypertext
system that permits ideal information retrieval for all
people.
3) There will never, ever be a truly advanced system
of machine translation that allows all texts to be
adequately--not perfectly but no more than adequately--
translated for all purposes.
These are the three contentions I mentioned in my sub-
title. They are closely related, and most of my
remaining fragments will be devoted to proving they are
true. I will also have a few words for anyone who may
be shocked by my use not only of the adverb never but
of its colloquial cousin "never, ever," but I will save
these for the Conference session itself.
[ASIDE, NOT IN PUBLISHED PAPER: At the "session
itself," the author took pains to inspire the audience,
by a show of hands, to express disagreement with his
position. He fully recounted many of the arguments
that could be used against his contentions. After all,
he pointed out, no one can predict what wonderful
progress science may make in the next 100 years: look
at the wonders of electricity, atomic power, airplanes,
space flight, all of them either inconceivable or in
their infancy 100 years ago. With these as bait, he
prompted audience members to raise their hands if they
believed all three of the obstacles he named in his
contentions would be readily overcome in the next 100
years. About one-fifth of those attending did so. He
then--amidst considerable laughter--challenged them to
consider the current condition of an extremely familiar
technological device (and its interface), which all of
them had used at least once before the conference
session and would use again after leaving it--and
which, moreover, has been commonly in use and under
constant improvement for the past ***120 years*** :
namely, ***the elevator***.
The basic functioning of an elevator could not possibly
be simpler, nor could its interface be more elementary.
There are two basic functions and two basic switches:
"UP" and "DOWN," in a sense a pure binary system. In
between there are a small or large (though
theoretically infinite) number of floors or levels, but
these are mere details, as are the "CLOSE" and "OPEN
DOOR" switches, the "STOP BUTTON," and the "ALARM
BELL." Yet with all this simplicity of design and
purpose, virtually no two elevators we enter are ever
the same or possess the same interface. As anyone
knows who has tried to figure out the controls of a
moving and/or misbehaving elevator, even after 120
years of development the interface is far from perfect
or consistent. Add to this that there are many urgent
reasons why this interface ought to have been
rationalized and perfected, potential loss of clothing,
arms, legs, and heads being perhaps foremost. Even if
we ignore the chaos of computer interfaces that now
surrounds us, are we really supposed to believe that the
development of considerably more complex computer,
hypertext, and MT interfaces will follow any smoother
course than that logged by the elevator over the
years to come? END OF ASIDE]
My next fragment is an episode and has to do with
CATNYP, the New York Public Library's computerized
catalog. I was using this system recently when a young
man sat down at the next computer. After spending about
half an hour fiddling through the help screens, he was
clearly close to despair. Finally, he turned to me and
whispered "Look, how do you use this thing anyway?" I
asked him what he was searching for, and he told me he
needed descriptions about clothing worn by the middle
classes in early nineteenth century England. "I've
tried `clothing,'" he lamented, "I've tried `middle
classes' and I've tried `England,' but none of it
works."
For the next half hour we all did our best--soon two
near-by researchers were also whispering solutions--to
refine (or should we say pre-edit?) his question so
that the computer could handle it. We ran through
"garb," "apparel," "attire," and other generic hedges
for "clothing," and finally our efforts were rewarded.
CATNYP produced a screen listing an illustrated book
that seemed to meet our friend's needs. We were elated
by our collective success, but then he asked another
question: "Hey, that's great, what a terrific system--
now how do I bring the book up on the screen?" With
some embarrassment, as though we ourselves were
responsible for the system's shortcomings, we explained
to him that he would have to fill out a slip, hand it
in at the desk, wait ten minutes or so, and finally be
handed a heavy, old-fashioned, page-ridden analogue
book. He was clearly annoyed by this, and to some
extent so were we.
But wait, I hear the cry ringing out, before you know
it, the great computer miracle will soon have solved
this. By tomorrow at the latest every single page of
every single book ever written will soon be accessible,
graphics, fold-outs, tables, and all--with just a few
keystrokes. Those who suppose this will truly become
possible, as with perfect machine translation, have not
even begun to focus on the scope of the problem. In
writing this paper, I made a few inquiries of
librarians, and they confirmed my worst suspicions many
times over.
The entire national library system long ago spent
millions on "up-to date technologies" in the form of
microfilm and microfiches. But even then, using a
relatively inexpensive technology, they came nowhere
near preserving all the world's books or periodicals
but only a small fraction of the most valuable of
these. Now both microfilm and microfiche are
considered a passe' technology, at least by computer
advocates and salesmen. But by the time these new
tools can possibly hope to record a comparable fraction
of printed materials, what further technology will lie
in wait with its own set of fabulous claims? Have we
willy-nilly been placed in the position of the Red
Queen in Alice in Wonderland, who had to run as fast as
she possibly could simply to remain in one place?
A few decades ago some of us collected 78 RPM records,
only to see them replaced by 33's and 45's, themselves
later eclipsed by audio cassettes and 8-track stereo,
all of them now replaced by CD's and/or their CD ROM
cousins. I am once again happily collecting these last
two products, but the word has been out for a few years
that even their days are numbered. And at no point did
any one of these remarkable media ever come close to
exhausting the simply unbelievable wealth of classical
music that exists in print or manuscript form. Is it
any wonder that librarians are skeptical of some of the
claims advanced by computer enthusiasts?
Now let's move in a bit closer to Machine Translation.
I hope everyone understands that in computer terms
there are distinct similarities in building a data base
of any kind, whether it is to catalog books, build a
thesaurus, set up an MT system, or create a model of
the universe. Basically, in all these cases, what we
are doing is constructing a data base, with just a few
exotic (or perhaps not-so-exotic) differences thrown
in. The computer doesn't care in the slightest which
one you are doing. In fact, the computer never knows
that it is word processing or accessing information or
plotting a map or even telecommunicating or printing
something on a page. In fact, the computer is so dumb
that it can't even tell when it is displaying an erotic
image.
When the time came to deepen the computer's
relationship with printers, it had to be fed something
called a page-description language--the most famous of
these is "Postscript." When it was decided to store
literary works in electronic form, a book-description
language had to be invented--here the best known is
"SGML." And with the advent of the World Wide Web, it
even became necessary to produce a screen-description
language--the now famous "HTML," which the experts are
already discussing how to change. But in order for
merely adequate machine translation to occur, it will
certainly be necessary to create something far more
ambitious, namely a "language-description language."
Such a construct would have to take into account most
phenomena that can occur in language, including
semantic and contextual elements.
Here is the crux question: how is such a language-
description language to be created? If we truly
attempt to include a large number of linguistic
aspects, then we will greatly increase the potential
for error and also end up with something quite
unwieldy. No matter how vast or fast our computers can
become, it may still be unwieldy in human terms, and
we're the ones who have to use such a program. If on
the other hand, we attempt to include only a limited
subset of language, then we will end up with something
like the Controlled Languages evolved by Caterpillar
and a few other firms and will have failed to reach our
original goal. This entire conundrum brings to mind
the problems encountered by cartographers in a one-
minute parable by Borges, which I will now read in its
entirety as my next fragment:
"Of Exactitude in Science
"...In that Empire, the craft of Cartography
attained such Perfection that the Map of a Single
province covered the space of an entire City, and
the Map of the Empire itself an entire Province.
In the course of Time, these Extensive maps were
found somehow wanting, and so the College of
Cartographers evolved a Map of the Empire that was
of the same Scale as the Empire and that coincided
with it point for point. Less attentive to the
Study of Cartography, succeeding Generations came
to judge a map of such Magnitude cumbersome, and
not without some irreverence, they abandoned it to
the Rigors of Sun and Rain. In the western
Deserts, tattered Fragments of the Map are still
to be found, Sheltering an occasional Beast or
beggar; in the whole Nation, no other relic is
left of the Discipline of Geography." (3)
Building a truly comprehensive language-description
language is altogether likely to involve the same order
of complexity and impracticality as this mission to
build a same-scale, point-to-point map of the Empire.
But now let's turn to how Machine Translation really
does work and see what lessons it holds for us.
Essentially MT will work best--perhaps not perfectly
even here--when you have what I call a Level Playing
Field Translation setup. [slide 6, which essentially
shows what is described below]
As you can see from the slide, what we have here is
literally a "level playing field" with something like a
tennis net in the middle. On one side, just to take
one possible example, (you could have any scientific
specialty at all, so let's be a trifle whimsical) you
have Professional Japanese Hydraulic Biochemical Micro-
Nuclear Space Scientists. On the other side, in the
English-speaking world, we find Professional American
Hydraulic Biochemical Micro-Nuclear Space Scientists.
Here, except for the difference in language, you have
an almost perfect match-up of knowledge and experience,
so this example ought to be absolutely ideal for
setting up an MT system, building terminology
databases, constructing lists of new or unknown words,
and pouring them into the original program.
But even here, in this nearly ideal MT setting, you can
still encounter problems. What if lexical entries
don't match up perfectly in the two languages? What if
the scientists in the two countries aren't following
the same procedures--in science or technology a near
certainty? What if they don't even see themselves as
performing the same steps for the same reasons? (Or
what if--for some reason--they don't want the other
scientists to **know** what they are doing?)
Any or all of this could potentially happen even in our
ideal level playing field example. But what if--let's
just suppose--there were some factors present that
tilted the playing field a bit to either side, or
tilted it for some of our experts in one direction but
for others in the opposite way? [slide is tilted] Some
possible instances: supposing half of the people on the
Japanese side turn out to be not Professional Hydraulic
Biochemical etceteras but newly trained interns
instead? Or if they are in fact professionals but come
from related scientific fields with slightly different
terminologies? What if they are students, or merely
stockholders in the company, or investigative
reporters, or members of the general public who have
wandered in to find out what the company is up to?
Much the same questions can be asked on the American
side, and the answers to questions on one side can
raise further questions on the other. What happens to
the translation process in any combination of these
conditions, even assuming human beings are providing
the translation? But, most important, what will happen
to an MT system under such circumstances?
Here we come to a crucial point which I have made
elsewhere in other terms. Contrary to our facile
belief that there can be such a thing as a "good
translation" or a "correct translation" that will work
in every case, no such thing as "generic translation"
may exist at all. It may simply be a convenient
fiction we have employed to shield us from the true
complexity of the translation process and/or as a way
of reassuring ourselves or our customers that we are in
all cases capable of producing a "correct translation."
Let me say this another way: there is one other crucial
factor involved in a translation besides the two
languages involved and the nature of the subject
matter--it concerns the audience and/or the occasion
for such a translation. Wherever this audience or
occasion changes even slightly, there may have to be a
corresponding shift in the tone of the translation.
Where either of these factors changes more than
slightly, we enter the territory of rejected
translations, possibly even charges of incompetence.
But even the most conscientious translator or
translation company may not always be prepared to meet
every demand these circumstances are capable of hurling
at us.
What we have run into here--or perhaps it has run into
us with a big stick in its hand--is the true extent of
the complexity of language. It is hard enough for
humans to work under such circumstances--how can we
expect machines to handle them? The real explanation
here may well be that we all make some outrageously
false assumptions about language and are totally
unaware we are doing so. Once again, we assume that we
are all walking around on a level playing field, where
anyone can readily communicate with anyone else across
a short and easily bridged distance.
But the truth is that we do not inhabit a level playing
field at all where language is concerned. On the
contrary, if we were to visualize ourselves and
everyone around us as walking about on stilts of
completely different heights, textures, and stability,
so that even our very own two stilts are not
necessarily of the same height or composition, we would
have a better notion of how we actually move through
linguistic space and communicate with others. You can
easily persuade yourself that this is true by the way
you react to others the next time you are in a social
situation.
We each of us have our own store of linguistic tricks
and devices, and we look out almost instinctively for
those who have complementary tricks and devices.
Whenever we meet such a person, we become flushed with
enthusiasm, sometimes even love, and go on talking
forever. But we just as quickly abandon those who do
not respond to our conversational rhythms. True, we
also carry on everyday conversations with persons who
do not share our interests or language style, but we
usually do not speak at length or in detail or about
more than a few topics with them. What I am trying to
suggest is that there is a whole universe of language
habits we are simply unaware of. And if we are not
aware of them, how can we suppose that a computer can
gain such awareness?
Let me now penetrate to the core of practicality about
MT: its place in the office environment. In every
company over a certain size, there exists at least one
individual whose sole duty is to make your existence as
a translator extremely unpleasant. This person may be
an office manager, an accountant, or perhaps even the
boss's personal assistant. In all these cases, such a
person will constantly be looking for ways of saving
money. Almost invariably, their gaze will fall upon
the translation department, whose employees are clearly
being overpaid to do work that should be accomplished
in a fraction of the time. Your doom may well be
spelled if truly persuasive sales reps from an MT
company pay a call, and this accountant (or whoever)
falls under their spell. If you do get the sack or
find yourself being retrained as an MT post-editor, you
have only one consolation. As likely as not, three
years later a completely new office manager (or
accountant or boss's assistant) will be looking for
ways to save again, and this time they will decide that
the MT system costs too much and makes too many errors,
and the time has come to retrain MT post-editors as
humans. Your best bet in any such situation is to
become as knowledgeable about MT as you possibly can be
and learn how to play office politics, so as to
influence the decision-making process before it
happens.
Finally, I'd like to round out this paper in two ways.
I started out by noting that several professional
organizations and perhaps even entire academic fields
are still dedicated to the goal of perfecting MT or
still hold that such an option is viable. Here is a
partial list of these groups and some of the
conferences they have been sponsoring [slide 7, which
duplicates the list shown below]:
REACHING FOR MIND:
FOUNDATIONS OF COGNITIVE SCIENCE
Call for Papers for the
Fourth International Conference on
The Cognitive Science of Natural Language Processing
LAST CALL FOR PAPERS
FORMAL GRAMMAR
in conjunction with the
European Summer School in Logic, Language and
Information
GROUNDING REPRESENTATIONS:
Integration of sensory information in
Natural Language Processing,
Artificial Intelligence and Neural Networks
IEE COLLOQUIUM
IEE Computing and Control Division
The School of Applied Languages, Dublin City University
The Association for Computational Linguistics
Fourth International Workshop on Parsing Technologies
II International Conference on Mathematical Linguistics
Fifth International Workshop on Natural Language
Understanding and Logic Programming
Many of the members of these groups are highly
respected and highly paid academics, and it might seem
sacrilegious to some to suggest that they can possibly
fail in their goal. I nonetheless believe that most of
the people in this room--perhaps even most of the
people at this Conference--possess deeper and more
useful knowledge about language than most of the
members of these groups, who could in fact profit
greatly by listening to what many of you could tell
them about language.
I hope that these fragments have now begun to fall into
place to some extent. And now, to conclude, let's look
briefly at our first two slides again. [slides of
Fengshui Compass and Llullian Wheels are shown again.]
And in that context let's look at some of the diagrams
produced by some MT advocates. Though only one is
reproduced in the Proceedings, several will be shown at
the Session itself. [four further slides serve to
make the point that follows--the diagrams themselves
are either pretentiously complex or simple-mindedly
silly.] Someone from another planet might suppose that
all these diagrams--MT, geomantic, theological, and
satirical alike--shared certain characteristics. After
all, they are all based on circles and other geometric
figures, and they may also share in a certain circular
reasoning in that they purport to represent clear and
reproducible relationships between the abstract and
practical realms, as do the magical drawings of many
cultures and ages. But it is by no means certain that
any of these diagrams fully succeeds in delineating--
much less establishing--such a relationship. Each one
may succeed up to a certain point only because of our
self-serving desire that it should do so, even our need
to believe in our own self-fulfilling prophecies. Yet
each one also fails because it falls short of
representing the complexities of both language and
reality by several orders of magnitude and makes no
allowance for the altogether variable identity of the
human value at its center. Thus, it may still remain
to be determined whether and/or to what extent our
Doctors of Artificial Intelligence, Cognitive Science,
and Mathematical Linguistics truly differ from the
Professors at the Academy of Lagado.
NOTES:
(1) Swift, Jonathan. Gulliver's Travels. Part Three,
Chapter 5.
(2) Hodges, Andrew (1983) Alan Turing: The Enigma Simon &
Schuster, New York. p. 382.
(3) Ostensibly from Travels of Praiseworthy Men (1658)
by J. A. Suárez Miranda, actually a part of Jorge Luís
Borges' A Universal History of Infamy, translated by
Norman Thomas di Giovanni, London: Allen Lane, 1973.
This paper is Copyright 1995 by Information Today, Inc.,
Medford, NJ.