[Written 15 July 1949. Published in: Machine translation of languages: fourteen essays, ed.

by William
N. Locke and A. Donald Booth (Technology Press of the Massachusetts Institute of Technology,
Cambridge, Mass., and John Wiley & Sons, Inc., New York, 1955), p.15-23.]

Translation

WARREN WEAVER

There is no need to do more than mention the obvious fact that a multiplicity of languages
impedes cultural interchange between the peoples of the earth, and is a serious deterrent to
international understanding. The present memorandum, assuming the validity and importance
of this fact, contains some comments and suggestions bearing on the possibility of
contributing at least something to the solution of the world-wide translation problem through
the use of electronic computers of great capacity, flexibility, and speed.
The suggestions of this memorandum will surely be incomplete and naïve, and may
well be patently silly to an expert in the field—for the author is certainly not such.

A War Anecdote—Language Invariants

During the war a distinguished mathematician whom we will call P, an ex-German

who had spent some time at the University of Istanbul and had learned Turkish there, told W.
W. the following story.
A mathematical colleague, knowing that P had an amateur interest in cryptography,
came to P one morning, stated that he had worked out a deciphering technique, and asked P to
cook up some coded message on which he might try his scheme. P wrote out in Turkish a
message containing about 100 words; simplified it by replacing the Turkish letters ç, ğ, ĭ, ö, ş,
and ü by c, g, i, o, s, and u respectively; and then, using something more complicated than a
simple substitution cipher, reduced the message to a column of five-digit numbers. The next
day (and the time required is significant) the colleague brought his result back, and remarked
that they had apparently not met with success. But the sequence of letters he reported, when
properly broken up into words, and when mildly corrected (not enough correction being
required really to bother anyone who knew the language well), turned out to be the original
message in Turkish.
The most important point, at least for present purposes, is that the decoding was done
by someone who did not know Turkish, and did not know that the message was in Turkish.
One remembers, by contrast, the well-known instance in World War I when it took our
cryptographic forces weeks or months to determine that a captured message was coded from
Japanese; and then took them a relatively short time to decipher it, once they knew what the
language was.
During the war, when the whole field of cryptography was so secret, it did not seem
discreet to inquire concerning details of this story; but one could hardly avoid guessing that
this process made use of frequencies of letters, letter combinations, intervals between letters
and letter combinations, letter patterns, etc., which are to some significant degree independent
of the language used. This at once leads one to suppose that, in the manifold instances in
which man has invented and developed languages, there are certain invariant properties which
are, again not precisely but to some statistically useful degree, common to all languages.
This may be, for all I know, a famous theorem of philology. Indeed the well-known
bow-wow, woof-woof, etc. theories of Müller and others, for the origin of languages, would of
course lead one to expect common features in all languages, due to their essentially similar
mechanism of development. And, in any event, there are obvious reasons which make the
supposition a likely one. All languages—at least all the ones under consideration here—were
invented and developed by men; and all men, whether Bantu or Greek, Islandic or Peruvian,
have essentially the same equipment to bring to bear on this problem. They have vocal organs
capable of producing about the same set of sounds (with minor exceptions, such as the glottal
click of the African native). Their brains are of the same general order of potential
complexity. The elementary demands for language must have emerged in closely similar
ways in different places and perhaps at different times. One would expect wide superficial
differences; but it seems very reasonable to expect that certain basic, and probably very
nonobvious, aspects be common to all the developments. It is just a little like observing that
trees differ very widely in many characteristics, and yet there are basic common
characteristics—certain essential qualities of "tree-ness,"—that all trees share, whether they
grow in Poland, or Ceylon, or Colombia. Furthermore (and this is the important point), a
South American has, in general, no difficulty in recognizing that a Norwegian tree is a tree.
The idea of basic common elements in all languages later received support from a
remark which the mathematician and logician Reichenbach made to W. W. Reichenbach also
spent some time in Istanbul, and, like many of the German scholars who went there, he was
perplexed and irritated by the Turkish language. The grammar of that language seemed to him
so grotesque that eventually he was stimulated to study its logical structure. This, in turn, led
him to become interested in the logical structure of the grammar of several other languages;
and, quite unaware of W. W.'s interest in the subject, Reichenbach remarked, "I was amazed
to discover that, for (apparently) widely varying languages, the basic logical structures have
important common features." Reichenbach said he was publishing this, and would send the
material to W. W.; but nothing has ever appeared.
One suspects that there is a great deal of evidence for this general viewpoint—at least
bits of evidence appear spontaneously even to one who does not see the relevant literature.
For example, a note in Science, about the research in comparative semantics of Erwin Reifler
of the University of Washington, states that "the Chinese words for 'to shoot' and 'to dismiss'
show a remarkable phonological and graphic agreement." This all seems very strange until
one thinks of the two meanings of "to fire" in English. Is this only happenstance? How
widespread are such correlations?

Translation and Computers

Having had considerable exposure to computer design problems during the war, and
being aware of the speed, capacity, and logical flexibility possible in modern electronic
computers, it was very natural for W. W. to think, several years ago, of the possibility that
such computers be used for translation. On March 4, 1947, after having turned this idea over
for a couple of years, W. W. wrote to Professor Norbert Wiener of Massachusetts Institute of
Technology as follows:

One thing I wanted to ask you about is this. A most serious problem, for UNESCO and for the
constructive and peaceful future of the planet, is the problem of translation, as it unavoidably affects the
communication between peoples. Huxley has recently told me that they are appalled by the magnitude and the
importance of the translation job.
Recognizing fully, even though necessarily vaguely, the semantic difficulties because of multiple
meanings, etc., I have wondered if it were unthinkable to design a computer which would translate. Even if it
would translate only scientific material (where the semantic difficulties are very notably less), and even if it did
produce an inelegant (but intelligible) result, it would seem to me worth while.
Also knowing nothing official about, but having guessed and inferred considerable about, powerful new
mechanized methods in cryptography—methods which I believe succeed even when one does not know what
language has been coded—one naturally wonders if the problem of translation could conceivably be treated as a
problem in cryptography. When I look at an article in Russian, I say: "This is really written in English, but it has
been coded in some strange symbols. I will now proceed to decode."
Have you ever thought about this? As a linguist and expert on computers, do you think it is worth
thinking about?

Professor Wiener, in a letter dated April 30, 1947, said in reply:

Second—as to the problem of mechanical translation, I frankly am afraid the boundaries of words in
different languages are too vague and the emotional and international connotations are too extensive to make any
quasimechanical translation scheme very hopeful. I will admit that basic English seems to indicate that we can go
further than we have generally done in the mechanization of speech, but you must remember that in certain
respects basic English is the reverse of mechanical and throws upon such words as get a burden which is much
greater than most words carry in conventional English. At the present tune, the mechanization of language, beyond
such a stage as the design of photoelectric reading opportunities for the blind, seems very premature. . . .

To this, W. W. replied on May 9, 1947:

I am disappointed but not surprised by your comments on the translation problem. The difficulty you
mention concerning Basic seems to me to have a rather easy answer. It is, of course, true that Basic puts multiple
use on an action verb such as get. But, even so, the two-word combinations such as get up, get over, get back, etc.,
are, in Basic, not really very numerous. Suppose we take a vocabulary of 2,000 words, and admit for good measure
all the two-word combinations as if they were single words. The vocabulary is still only four million: and that is
not so formidable a number to a modern computer, is it?

Thus this attempt to interest Wiener, who seemed so ideally equipped to consider the
problem, failed to produce any real result. This must in fact be accepted as exceedingly
discouraging, for, if there are any real possibilities, one would expect Wiener to be just the
person to develop them.
The idea has, however, been seriously considered elsewhere. The first instance known
to W. W., subsequent to his own notion about it, was described in a memorandum dated
February 12, 1948, written by Dr. Andrew D. Booth who, in Professor J. D. Bernal's
department in Birkbeck College, University of London, had been active in computer design
and construction. Dr. Booth said:

A concluding example, of possible application of the electronic computer, is that of translating from one
language into another. We have considered this problem in some detail, and it transpires that a machine of the type
envisaged could perform this function without any modification in its design.

On May 25, 1948, W. W. visited Dr. Booth in his computer laboratory at Welwyn,
London, and learned that Dr. Richens, Assistant Director of the Bureau of Plant Breeding and
Genetics, and much concerned with the abstracting problem, had been interested with Dr.
Booth in the translation problem. They had, at least at that time, not been concerned with the
problem of multiple meaning, word order, idiom, etc., but only with the problem of
mechanizing a dictionary. Their proposal then was that one first "sense" the letters of a word,
and have the machine see whether or not its memory contains precisely the word in question.
If so, the machine simply produces the translation (which is the rub; of course "the"
translation doesn't exist) of this word. If this exact word is not contained in the memory, then
the machine discards the last letter of the word, and tries over. If this fails, it discards another
letter, and tries again. After it has found the largest initial combination of letters which is in
the dictionary, it "looks up" the whole discarded portion in a special "grammatical annex" of
the dictionary. Thus confronted by running, it might find run and then find out what the
ending (n)ing does to run.
Thus their interest was, at least at that time, confined to the problem of the
mechanization of a dictionary which in a reasonably efficient way would handle all forms of
all words. W. W. has no more recent news of this affair.
Very recently the newspapers have carried stories of the use of one of the California
computers as a translator. The published reports do not indicate much more than a word-into-
word sort of translation, and there has been no indication, at least that W. W. has seen, of the
proposed manner of handling the problems of multiple meaning, context, word order, etc.
This last-named attempt, or planned attempt, has already drawn forth inevitable scorn,
Mr. Max Zeldner, in a letter to the Herald Tribune on June 13, 1949, stating that the most you
could expect of a machine translation of the fifty-five Hebrew words which form the 23d
Psalm would start out Lord my shepherd no I will lack, and would close But good and
kindness he will chase me all days of my life; and I shall rest in the house of Lord to
length days. Mr. Zeldner points out that a great Hebrew poet once said that translation "is
like kissing your sweetheart through a veil."
It is, in fact, amply clear that a translation procedure that does little more than handle
a one-to-one correspondence of words cannot hope to be useful for problems of literary
translation, in which style is important, and in which the problems of idiom, multiple
meanings, etc., are frequent.
Even this very restricted type of translation may, however, very well have important
use. Large volumes of technical material might, for example, be usefully, even if not at all
elegantly, handled this way. Technical writing is unfortunately not always straightforward and
simple in style; but at least the problem of multiple meaning is enormously simpler. In
mathematics, to take what is probably the easiest example, one can very nearly say that each
word, within the general context of a mathematical article, has one and only one meaning.

The Future of Computer Translation

The foregoing remarks about computer translation schemes which have been reported
do not, however, seem to W. W. to give an appropriately hopeful indication of what the future
possibilities may be. Those possibilities should doubtless be indicated by persons who have
special knowledge of languages and of their comparative anatomy. But again, at the risk of
being foolishly naïve, it seems interesting to indicate four types of attack, on levels of
increasing sophistication.

Meaning and Context

First, let us think of a way in which the problem of multiple meaning can, in principle
at least, be solved. If one examines the words in a book, one at a time as through an opaque
mask with a hole in it one word wide, then it is obviously impossible to determine, one at a
time, the meaning of the words. "Fast" may mean "rapid"; or it may mean "motionless"; and
there is no way of telling which.
But, if one lengthens the slit in the opaque mask, until one can see not only the central
word in question but also say N words on either side, then, if N is large enough one can
unambiguously decide the meaning of the central word. The formal truth of this statement
becomes clear when one mentions that the middle word of a whole article or a whole book is
unambiguous if one has read the whole article or book, providing of course that the article or
book is sufficiently well written to communicate at all.
The practical question is: "What minimum value of N will, at least in a tolerable
fraction of cases, lead to the correct choice of meaning for the central word?
This is a question concerning the statistical semantic character of language which
could certainly be answered, at least in some interesting and perhaps in a useful way. Clearly
N varies with the type of writing in question. It may be zero for an article known to be about a
specific mathematical subject. It may be very low for chemistry, physics, engineering, etc. If
N were equal to 5, and the article or book in question were on some sociological subject,
would there be a probability of 0.95 that the choice of meaning would be correct 98% of the
time? Doubtless not: but a statement of this sort could be made, and values of N could be
determined that would meet given demands.
Ambiguity, moreover, attaches primarily to nouns, verbs, and adjectives; and actually
(at least so I suppose) to relatively few nouns, verbs, and adjectives. Here again is a good
subject for study concerning the statistical semantic character of languages. But one can
imagine using a value of N that varies from word to word, is zero for he, the, etc., and needs
to be large only rather occasionally. Or would it determine unique meaning in a satisfactory
fraction of cases, to examine not the 2N adjacent words, but perhaps the 2N adjacent nouns?
What choice of adjacent words maximizes the probability of correct choice of meaning, and at
the same time leads to a small value of N?
Thus one is led to the concept of a translation process in which, in determining
meaning for a word, account is taken of the immediate (2N word) context. It would hardly be
practical to do this by means of a generalized dictionary which contains all possible phases
2N + 1 words long: for the number of such phases is horrifying, even to a modern electronic
computer. But it does seem likely that some reasonable way could be found of using the
micro context to settle the difficult cases of ambiguity.

Language and Logic

A more general basis for hoping that a computer could be designed which would cope
with a useful part of the problem of translation is to be found in a theorem which was proved
in 1943 by McCulloch and Pitts.1 This theorem states that a robot (or a computer) constructed
with regenerative loops of a certain formal character is capable of deducing any legitimate
conclusion from a finite set of premises.
Now there are surely alogical elements in language (intuitive sense of style,
emotional content, etc.) so that again one must be pessimistic about the problem of literary
translation. But, insofar as written language is an expression of logical character, this theorem
assures one that the problem is at least formally solvable.

Translation and Cryptography

Claude Shannon, of the Bell Telephone Laboratories, has recently published some
remarkable work in the mathematical theory of communication.2 This work all roots back to
the statistical characteristics of the communication process. And it is at so basic a level of
generality that it is not surprising that his theory includes the whole field of cryptography.
During the war Shannon wrote a most important analysis of the whole cryptographic problem,
and this work is, W. W. believes, also to appear soon, it having been declassified.
Probably only Shannon himself, at this stage, can be a good judge of the possibilities
in this direction; but, as was expressed in W. W.'s original letter to Wiener, it is very tempting
to say that a book written in Chinese is simply a book written in English which was coded
into the "Chinese code." If we have useful methods for solving almost any cryptographic
problem, may it not be that with proper interpretation we already have useful methods for
translation?
This approach brings into the foreground an aspect of the matter that probably is
absolutely basic—namely, the statistical character of the problem. "Perfect" translation is
almost surely unattainable. Processes, which at stated confidence levels will produce a
translation which contains only X per cent "error," are almost surely attainable.
And it is one of the chief purposes of this memorandum to emphasize that statistical
semantic studies should be undertaken, as a necessary preliminary step.
The cryptographic-translation idea leads very naturally to, and is in fact a special case
of, the fourth and most general suggestion: namely, that translation make deep use of
language invariants.

Language and Invariants

Indeed, what seems to W. W. to be the most promising approach of all is one based
on the ideas expressed on pages 16-17—that is to say, an approach that goes so deeply into
the structure of languages as to come down to the level where they exhibit common traits.
Think, by analogy, of individuals living in a series of tall closed towers, all erected
over a common foundation. When they try to communicate with one another, they shout back
and forth, each from his own closed tower. It is difficult to make the sound penetrate even the
nearest towers, and communication proceeds very poorly indeed. But, when an individual
goes down his tower, he finds himself in a great open basement, common to all the towers.
Here he establishes easy and useful communication with the persons who have also
descended from their towers.
Thus may it be true that the way to translate from Chinese to Arabic, or from Russian
to Portuguese, is not to attempt the direct route, shouting from tower to tower. Perhaps the
way is to descend, from each language, down to the common base of human
communication—the real but as yet undiscovered universal language—and then re-emerge by
whatever particular route is convenient.
 Such a program involves a presumably tremendous amount of work in the logical
structure of languages before one would be ready for any mechanization. This must be very
closely related to what Ogden and Richards have already done for English—and perhaps for
French and Chinese. But it is along such general lines that it seems likely that the problem of
translation can be attacked successfully. Such a program has the advantage that, whether or
not it lead to a useful mechanization of the translation problem, it could not fail to shed much
useful light on the general problem of communication.

REFERENCES
1
Warren S. McCulloch and Walter Pitts, Bull. math. Biophys., no 5, pp. 115-133, 1943
2
For a very simplified version, see "The Mathematics of Communication," by Warren Weaver, Sci. Amer., vol.
181, no. 1, pp. 11-15, July 1949. Shannon's original papers, as published in the Bell Syst. tech. J., and a longer
and more detailed interpretation by W. W. are about to appear as a memoir on communication, published by the
University of Illinois Press. A book by Shannon on this subject is also to appear soon. [A joint book, The
Mathematical Theory of Communication, by Shannon and Weaver, was published by the University of Illinois
Press in 1949—Editors' Note'
 Anexo 2: The Development and use of machine translation
systems and computer-based translation tools, de John hutchins
 International Symposium on Machine Translation and Computer Language Information
Processing, 26-28 June 1999, Beijing, China

The development and use of machine translation systems

and computer-based translation tools

John Hutchins
(University of East Anglia, Norwich NR4 7TJ, England)
(E-mail: [email protected])

Abstract: This survey of the present demand and use of computer-based translation
software concentrates on systems designed for the production of translations of
publishable quality, including developments in controlled language systems, translator
workstations, and localisation; but it covers also the developments of software for
non-translators, in particular for use with Web pages and other Internet applications,
and it looks at future needs and systems under development. The final section
compares the types of translations that can be met most appropriately by human and
by machine (and computer-aided) translation respectively.

Keywords: machine translation, computer-aided translation, translator workstations,

multilingual systems

Types of translation demand

When giving any general overview of the development and use of machine
translation (MT) systems and translation tools, it is important to distinguish four basic
types of translation demand. The first, and traditional one, is the demand for
translations of a quality normally expected from human translators, i.e. translations of
publishable quality – whether actually printed and sold, or whether distributed
internally within a company or organisation. The second basic demand is for
translations at a somewhat lower level of quality (and particularly in style), which are
intended for users who want to find out the essential content of a particular document
– and generally, as quickly as possible. The third type of demand is that for translation
between participants in one-to-one communication (telephone or written
correspondence) or of an unscripted presentation (e.g. diplomatic exchanges.) The
fourth area of application is for translation within multilingual systems of information
retrieval, information extraction, database access, etc.
The first type of demand illustrates the use of MT for dissemination. It has
been satisfied, to some extent, by machine translation systems ever since they were
first developed in the 1960s. However, MT systems produce output which must
invariably be revised or ‘post-edited’ by human translators if it is to reach the quality
required. Sometimes such revision may be substantial, so that in effect the MT system
is producing a ‘draft’ translation. As an alternative, the input text may be regularised
(or ‘controlled’ in vocabulary and sentence structure) so that the MT system produces
few errors which have to be corrected. Some MT systems have, however, been
developed to deal with a very narrow range of text content and language style, and
these may require little or no preparation or revision of texts.
In recent years, the use of MT systems for dissemination purposes has been
augmented by developments in translation tools (e.g. terminology databases and
translation memories), integrated in authoring and publishing processes. These

1
‘translation workstations’ are more attractive to human translators. Whereas, with MT
systems translators see themselves as subordinate to the machine, in so far as they
edit, correct or re-translate the output from a computer, with translation workstations
(or workbenches) the translators are in control of computer-based facilities, which
they can accept or reject as they wish.
The second type of demand – the use of MT for assimilation – has been met
in the past as, in effect, a by-product of systems designed originally for the
dissemination application. Since MT systems did not (and still cannot) produce high
quality translations, some users have found that they can extract what they needed to
know from the unedited output. They would rather have some translation, however
poor, than no translation at all. With the coming of cheaper PC-based systems on the
market, this type of use has grown rapidly and substantially.
With the third type – MT for interchange – the situation is changing quickly.
The demand for translations of electronic texts on the Internet, such as Web pages,
electronic mail and even electronic ‘chat’ lists, is developing rapidly. In this context,
the possibility of human translation is out of the question. The need is for immediate
translation in order to convey the basic content of messages, however poor the input.
MT systems are finding a ‘natural’ role, since they can operate virtually or in fact in
real-time and on-line and there has been little objection to the inevitable poor quality.
Another context for MT in personal interchange is the focus of much research. This is
the development of systems for spoken language translation, e.g. in telephone
conversations and in business negotiations. The problems of integrating speech
recognition and automatic translation are obviously formidable, but progress is
nevertheless being made. In the future – still distant, perhaps – we may expect on-line
MT systems for the translation of speech in highly restricted domains.
The fourth type of MT application – as components of information access
systems – is the integration of translation software into: (i) systems for the search and
retrieval of full texts of documents from databases (generally electronic versions of
journal articles in science, medicine and technology), or for the retrieval of
bibliographic information; (ii) systems for extracting information (e.g. product
details) from texts, in particular from newspaper reports; (iii) systems for
summarising texts; and (iv) systems for interrogating non-textual databases. This field
is the focus of a number of projects in Europe at the present time, which have the aim
of widening access for all members of the European Union to sources of data and
information whatever the source language.

Historical background
Systems for automatic translation have been under development for 50 years –
in fact, ever since the electronic computer was invented in the 1940s there has been
research on their application for translating languages (Hutchins 1986). For many
years, the systems were based primarily on direct translations via bilingual
dictionaries, with relatively little detailed analysis of syntactic structures. By the
1980s, however, advances in computational linguistics allowed much more
sophisticated approaches, and a number of systems adopted an indirect approach to
the task of translation. In these systems, texts of the source language are analysed into
abstract representations of ‘meaning’, involving successive programs for identifying
word structure (morphology) and sentence structure (syntax) and for resolving
problems of ambiguity (semantics). Included in the latter are component programs to
distinguish between homonyms (e.g. English words such as light, which can be a
noun, and adjective or verb, and solution, which can be a mathematical or a chemical

2
term) and to recognise the correct semantic relationships (e.g. in The driver of the bus
with a yellow coat). The abstract representations are intended to be unambiguous and
to provide the basis for the generation of texts into one or more target languages.
There have in fact been two basic ‘indirect’ approaches. In one the abstract
representation is designed to be a kind of language-independent ‘interlingua’, which
can potentially serve as an intermediary between a large number of natural languages.
Translation is therefore in two basic stages: from the source language into the
interlingua, and from the interlingua into the target language. In the other indirect
approach (in fact, more common approach) the representation is converted first into
an equivalent representation for the target language. Thus there are three basic stages:
analysis of the input text into an abstract source representation, transfer to an abstract
target representation, and generation into the output language.
Until the late 1980s, systems of all these kinds were developed, and it is true
to say that all current commercially available systems are also classifiable into these
three basic system types: direct, interlingual and ‘transfer’. The best known of the MT
systems for mainframe computers are in fact essentially of the ‘direct translation’
type, e.g. the Systran, Logos and Fujitsu (Atlas) systems. They are however improved
versions of the type; unlike their predecessors, they are highly modular in
construction and easily modifiable and extendable. In particular, the Systran system,
originally designed for translation only from Russian into English, is now available
for a very large number of language pairs: English from and into most European
languages (French, German, Italian, Spanish, Portuguese), Japanese, Korean, etc.
Logos, originally marketed for German to English, is also now available for other
languages: English into French, German, Italian and Spanish, and German into French
and Italian. The Fujitsu ATLAS system, on the other hand, is still confined to
translation between English and Japanese (in both directions).
Among the most important of the mainframe ‘transfer’ systems was METAL,
supported for most of the 1980s by Siemens in Germany. However, it was only at the
end of the decade that METAL came onto the market, and sales were poor. During the
1990s, rights to METAL have been transferred to two organisations (GMS and
LANT) in a complex arrangement. But the best known systems adopting the ‘transfer’
approach were research projects: Ariane at GETA in Grenoble, an MT project going
back to the 1960s, and Eurotra funded by the Commission of the European
Communities. There were hopes that Ariane would become the French national
system, and there were plans to incorporate it in a translator’s workstation for
Eurolang (see below), but in the end nothing came of them. As for Eurotra, it was
undoubtedly one of the most sophisticated systems, but after involving some hundred
of researchers in most countries of Western Europe for almost a decade, it failed to
produce the working system that the sponsors wanted. It had been hoped that Eurotra
would eventually replace the Systran systems that the Commission had acquired and
was developing internally. In the late 1980s, Japanese governmental agencies began
to sponsor an interlingua system for Asian languages, involving co-operation with
researchers in China, Thailand, Malaysia and Indonesia. However, this project too has
so far not produced a system after a decade of work. (For surveys of MT research and
development in 1980s and early 1990s see Hutchins 1993, 1994.)

Governmental and non-commercial use

The earliest installations of MT systems were in national and international
governmental and military translation services – primarily because they could afford
the costs of the computer hardware required. The US Air Force introduced Systran in

3
1970 for translating Russian military scientific and technical documentation into
English. Although some documents were edited, much of the output was passed to
recipients without revision; over 90% accuracy for technical reports is claimed. The
National Air Intelligence Center, which took over the service from the USAF, now
produces translations (many unedited) for a wide range of US government
organisations (Pedtke 1997). As well as Russian-English it has available systems from
Systran for translating Japanese, Chinese and Korean into English, and under
development with Systran is a system for SerboCroat into English.
In Europe, the largest translation service is that of the European Commission,
and was one of the first organisations to install MT. It began in 1976 with the Systran
system for translating from English into French. In subsequent years, versions were
developed for many other language pairs, covering the needs for translation among
the European Union languages. While the translation of many legal texts continues to
be done by human translators, the Systran systems are used increasingly not only for
the translation of internal documents (with or without post-editing) but also as rough
versions for the assistance of administrators when composing texts in non-native
languages (Senez 1996).

Production of technical documentation

Until the 1990s the normal assumption was that MT systems were intended to
be used for the production of documentation of publishable quality, primarily but not
exclusively of a scientific and technical nature. The assumption was, in other words,
that MT systems were to be used in conditions where otherwise human translators
would be employed with expertise in the subjects concerned. Evidently, the actual
quality of MT output was inadequate for direct use. It had to be extensively revised
before it could be published, and translators were therefore employed as ‘post-
editors’. In these circumstances, the use of MT became a matter of economics. It was
viable only if overall quality and speed could be achieved at lower cost than the
employment of human translators.
Although today there are other uses for MT, as we have already indicated, this
application remains the most important, particularly for the vendors and developers of
the larger ‘mainframe’-type systems (Systran and Logos). The main customers and
users are the multinational companies exporting equipment in the global market
(Vasconcellos 1993; Brace et al. 1995). The need here is for translation of
promotional and technical documentation. In the latter case, technical documents are
often required in very large volumes: a set of operational manuals for a single piece of
equipment may amount to several thousands of pages Furthermore, there can be
frequent revisions with the appearance of new models. In addition, there must be
consistency in translation: the same component must be referred to and translated the
same way each time. This scale of technical translation is well beyond human
capacity. Nevertheless, in order to be most cost-efficient, a MT system should be well
integrated within the overall technical documentation processes of the company: from
initial writing to final publishing and distribution. Systems developed for the support
of technical writers – not just assistance with terminology, but also on-line style
manuals and grammar aids – are now being linked seamlessly into translation and
publishing processes.
There are numerous examples of the successful and long-term use of MT
systems by multinationals for technical documentation. One of the best known is the
application of the Logos systems at the Lexi-Tech company in New Brunswick,
Canada; initially for the translation into French of manuals for the maintenance of

4
naval frigates, the company has built up a service undertaking many other large
translation projects. Also using Logos are Ericsson, Osram, Océ Technologies, SAP
and Corel. Systran has many large clients: Ford, General Motors, Aérospatiale,
Berlitz, Xerox, etc. The METAL German-English system has been successfully used
at a number of European companies: Boehringer Ingelheim, SAP, Philips, and the
Union Bank of Switzerland.
A pre-requisite for successful MT installation in large companies is that the
user expects a large volume of translation within a definable domain (subjects,
products, etc.) The financial commitment to a terminology database and to dictionary
maintenance must be justifiable. Whether produced automatically or not, it is
desirable for company documentation to be consistent in the use of terminology.
Many companies in fact insist upon their own use of terms, and will not accept the
usage of others. To maintain such consistency is almost impossible outside an
automated system. However, it does mean that before an MT system can be installed,
the user must have already available a well-founded terminological database, with
authorised translation equivalents in the languages involved, or – at least – must make
a commitment to develop the required term bank.
For similar reasons, it is often desirable if the MT system is to produce output
in more that one target language. Most large-scale MT systems have to be customised,
to a greater or lesser extent, for the kind of language found in the types of documents
produced in a specific company. This can be the addition of specific grammatical
rules to deal with frequent sentence and clause constructions, as well as the inclusion
of specific rules for dealing with lexical items, and not just those terms unique to the
company. The amount of work involved in such customisation may not be justifiable
unless output is in a number of different languages.

Controlled language and domain-specific systems

In these circumstances, however, it has often been found feasible to introduce
a greater degree of control. One of the earliest and best known examples is the
application of the Systran system by the Xerox Corporation. At Xerox technical
authors are obliged to compose documents in what is called Multinational Customized
English, where not only the use of specific terms is laid down but also the
construction of sentences (Elliston 1979). The advantages of this approach are: the
avoidance of ambiguities in the input which the MT system cannot deal with
adequately, the consequential better quality output, the faster production of technical
documents simultaneously in a number of different languages, and (not least) the
production of more easily comprehensible English documents. These advantages have
been recognised by other multinational companies, and the use of ‘controlled
languages’ is increasing: for example, the Caterpillar Corporation has devised its own
form of English to facilitate translation in a knowledge-based MT system being
developed for it at the Carnegie-Mellon University (Mitamura and Nyberg 1995).
There are some companies offering to build ‘controlled’ language MT systems for
specific clients. The oldest established – and the pioneer in this approach – is the
Smart Corporation, New York. Systems have been developed by Smart for a number
of major clients: Citicorp, Chase, Ford, General Electric, etc. Each incorporates a
system for ‘normalising’ English documents. This system component is considered so
crucial to success that the actual translation process is regarded as virtually a ‘by-
product’ (Lee 1994). There are Smart systems translating into French, German,
Greek, Italian, Japanese, and Spanish. The largest Smart installation, perhaps, is the
system designed for the Canadian Ministry of Employment, where it has been used

5
for many years to translate information about job advertisements and similar
documentation.
In Europe, the Cap Volmac company in the Netherlands and the LANT
company in Belgium offer similar services, building for various clients specialised
translation systems utilising their own software for controlled languages. Cap Volmac
Lingware Services is a Dutch subsidiary of the Cap Gemini Sogeti Group. Over the
years this software company has constructed controlled-language systems for textile
and insurance companies, mainly from Dutch to English (Van der Steen and
Dijenborgh 1992). However, possibly the best known success story for custom-built
MT is the PaTrans system developed for LingTech A/S to translate English patents
into Danish. The system is based on methods and experience gained from the Eurotra
project of the European Commission (Ørsnes et al. 1996)
These last examples of systems illustrate that a growing number of companies
and organisations are developing their own MT facilities, as opposed to purchasing
commercial systems. This has been a feature from early days. The successful Météo
system in Canada for translating weather forecasts from English into French (and later
from French into English) was effectively a customer-specific system – in this case
the Canadian Environment service. It may be noted that a variant of the Météo
software was successfully operated during the Olympic games in Atlanta (Chandioux
and Grimaila 1996). Météo is an example of a ‘sublanguage’ system, i.e. designed for
to deal with the particular language of meteorology.
Another example of a custom-built system is TITUS, a highly constrained
‘sublanguage’ system for translating abstracts of documents of the textile industry
from and into English, French, German, and Spanish, in regular use since 1970. Better
known are the two customer-specific systems for translating between English and
Spanish built at the Pan American Health Organization in Washington – designed and
developed by workers in the organisation itself. These highly successful systems (now
also available to users outside PAHO) are general-purpose systems, not constrained in
vocabulary or text type, although obviously the dictionaries are strongest in the
health-related social science fields (Leon and Aymerich 1997).
In the 1990s there have been a number of other examples. In Finland, the
Kielikone system was developed originally as a workstation for Nokia
Telecommunications. Subsequently, versions were installed at other Finnish
companies and the system is now being marketed more widely (Arnola 1996). A
similar story applies to GSI-Erli. This large language engineering company developed
an integrated in-house translation system combining a MT engine and various
translation aids and tools on a common platform AlethTrad. Recently it has been
making the system available in customised versions for outside clients (Humphreys
1996).
On a smaller scale, but equally successful, has been the system developed by
the translation service of a small British company Hook and Hatton. In this case, the
need was for translation of chemical texts from Dutch into English (Lewis 1997). The
designer began by simple pattern matching of phrases, and gradually built in more
syntactic analysis as and when results were justifiable and cost-effective.
Based on experience over many years in developing knowledge-based MT and
experimenting with speech translation and corpus-based methods, members of the
group at Carnegie-Mellon University have developed an architecture for the rapid
production of usable MT systems for specific clients in some less common languages,
such as SerboCroat and Haitian Creole (Frederking et al. 1997). There is no pretence
of high quality, merely ‘usefulness’ for languages otherwise inaccessible.

6
 Another example of custom-built MT in a specialised area is the program
developed for TCC Communications at the Simon Fraser University for translating
closed captions on television programs (Toole et al. 1998). Not only are there time
constraints – translation must be in real-time – but also there are the challenges of
colloquialisms, dialogue, robustness, and paucity of context indicators. The system, at
present running live for English into Spanish, demanded techniques otherwise found
mainly in Internet applications (see below.)
In Japan, there are further examples of custom-built systems. The Japan
Information Centre of Science and Technology translates abstracts of Japanese
scientific and technical articles into English. In the late 1980s it assumed
responsibility of the Mu Japanese-English MT system developed at the University of
Kyoto. From this, it now has one of the largest MT operations in Japan (O’Neill-
Brown 1996). Other custom-built systems of significance in Japan are the SHALT
system developed by IBM Japan for its own translation needs, the ARGO system
developed by CSK in Tokyo for translating Japanese stock market reports into
English, and the NHK system for translating English news articles into Japanese.

Translation workstations
In the 1990s, the possibilities for large-scale translation broadened with the
appearance on the market of translation workstations (or translator workbenches). The
original ideas for integrating various computer-based facilities for translators at one
place go back to the early 1980s, in particular with the systems from ALPS.
Translation workstations combine multilingual word processing, means of receiving
and sending electronic documents, OCR facilities, terminology management software,
facilities for concordancing, and in particular ‘translation memories’. The latter is a
facility that enables translators to store original texts and their translated versions side
by side, i.e. so that corresponding sentences of the source and target are aligned. The
translator can thus search for a phrase or even full sentence in one language in the
translation memory and have displayed corresponding phrases in the other language.
These may be either exact matches or approximations ranked according to closeness.
It is often the case in large companies that technical documents, manuals, etc.
undergo numerous revisions. Large parts may remain unchanged from one version to
the next. With a translation memory, the translator can locate and re-use already
translated sections. Even if there is not an exact match, the versions displayed may
usable with minor changes. There will also be access to terminology databases, in
particular company-specific terminology, for words or phrases not found in the
translation memory. In addition, many translator workstations are now offering full
automatic translations using MT systems such as Systran, Logos, and Transcend. The
translator can choose to use them either for the whole text or for selected sentences,
and can accept or reject the results as appropriate (Heyn 1997).
There are now four main vendors of workstations: Trados (probably the most
successful), STAR AG in Germany (Transit), IBM (the TranslationManager), and
LANT in Belgium (the Eurolang Optimizer, previously sold by SITE in France). The
translation workstation has revolutionised the use of computers by translators. They
have now a tool where they are in full control. They can use any of the facilities or
none of them as they choose. As always, the value of each resource depends on the
quality of the data. As in MT systems, the dictionaries and terminology databases
demand effort, time and resources. Translation memories rely on the availability of
suitable large corpora of authoritative translations – there is no point in using

7
translations which are unacceptable (for whatever reason) by the company or the
client.
Although widely used by administrators within the European Commission, the
full-scale MT system Systran is relatively little used by the Commission’s
professional translators. For them, the translation service is developing its own
workstation facility, EURAMIS, i.e. European Advanced Multilingual Information
System (Theologitis 1997). This combines access to the Commission’s own very large
multilingual database (Eurodicautom), the dictionary resources of Systran, facilities
for individual and group terminology database creation and maintenance (using
Trados’ MultiTerm software), translation memory (again for individuals and groups),
access to CELEX (the full-text database of European Union legislation and
directives), software for document comparison (to detect where changes have taken
place), and also, of course, access to the Systran MT systems themselves. The latter
are now available from English into Dutch, French, German, Greek, Italian,
Portuguese, and Spanish; from French into Dutch, English, German, Italian, and
Spanish; from Spanish into English and French; and from German into English and
French. The whole EURAMIS system is linked to other facilities such as authoring
tools (spelling, grammar and style checkers, and multilingual drafting aids), the
internal European Commission administrative network, and to outside resources on
the Internet.

Localisation of software
One of the fastest growing areas for the use of computers in translation is in
the industry of software localisation. Here the demand is for supporting
documentation to be available in many languages at the time of the launch of new
software. Translation has to be done quickly, but there is much repetition of
information from one version to another. MT and, more recently, translation
memories in translation workstations are the obvious solution (Schaeler 1996).
Among the first in this field was the large software company SAP AG in Germany.
They use two MT systems: METAL for German to English translation, and Logos for
English to French, and plan to introduce further systems for other language pairs.
Most localisation, however, is based on the translation memory and
workstation approach. Typical examples are Corel, Lotus, and Canon. It is interesting
to note that much of this localisation activity is based in Ireland – thanks to earlier
government and European Union support for the computer industry. However,
localisation is a multi-national and global industry, with its own organisation
(Localization Industry Standards Association, based in Geneva) holding regular
seminars and conferences in all continents (For details see LISA Forum Newsletter)
Localisation companies have been at the forefront of efforts in Europe to
define standardised lexical resource and text handling formats, and to develop
common network infrastructures. This is the OTELO project coordinated by Lotus in
Ireland, with other members such as SAP, Logos, and GMS. The need for a general
translation environment for a wide variety of translation memory, machine translation
and other productivity tools is seen as fundamental to the future success of companies
in the localisation industry.

Systems for personal computers

Software for personal computers began to appear in the early 1980s (with the
Weidner MicroCAT system becoming particularly successful). Nearly all the main
Japanese computer companies produced systems for translation to and from English,

8
e.g. the PIVOT system from NEC, the ASTRANSAC system from Toshiba, HICATS
from Hitachi, PENSEE from Oki and DUET from Sharp.
Outside Japan, systems for personal computers began to appear a little earlier,
but from relatively few companies. The first American systems came in the early
1980s from ALPS and from Weidner. The ALPS products were intended primarily as
aids for translation, providing tools for accessing and creating terminology resources
but they did include an interactive translation module. Although at first sold with
some success, the producers concluded by the end of the decade that the market was
not yet ready and the products were in effect withdrawn. Instead, ALPS turned itself
into a translation service (ALPNET), using its own tools internally. By contrast,
Weidner sold a full translation system in a growing number of language pairs
(English, French, German, Spanish), and the business flourished. Weidner produced
two versions of its systems: MicroCat for small personal computers, and MacroCat
for larger minicomputers or workstations. The company was then purchased by a
Japanese company Bravis, a Japanese version was marketed, but soon afterwards the
owner decided that the MT market for personal computers was still undeveloped and
the business was sold. MicroCat disappeared completely, but MacroCat was
purchased by Intergraph, who modified and developed it for its range of publishing
software and sold it later as Transcend – recently Transcend was acquired by
Transparent Language Inc. (For these developments see Hutchins 1993, 1994).
At the end of the 1980s, most of the commercial systems on the market today
appeared. First came the PC-Translator systems (from Linguistic Products, based in
Texas) for low-end personal computers. Over the years, many language pairs have
been produced and marketed, apparently successfully as far as sales are concerned.
Next came Globalink with systems for French, German and Spanish to and from
English. (There was also a Russian-English system deriving essentially from the
original owner’s experience on the 1960s Georgetown project.) Within a few years,
Globalink merged with MicroTac, a company which had been very successful in
selling its cheap Language Assistant series of PC software – essentially automatic
dictionaries, with minimal phrase translation facility. In the early 1990s, Globalink
produced its now well-known ‘Power Translator’ series for translation of English to
and from French, German and Spanish, and recently Globalink has marketed the more
advanced ‘Telegraph’ series of translation software products, and Globalink itself was
acquired by Lernout & Hauspie, a leading speech technology company.
Since the beginning of the 1990s, many other systems for personal computers
have appeared. For Japanese and English there are now also LogoVista from the
Language Engineering Corporation, and Tsunami and Typhoon from Neocor
Technologies (also now owned by Lernout & Hauspie). From the former Soviet
Union – where particularly in the 1960s and 1970s there was very active research on
MT – we have now Stylus (recently renamed ProMT) and PARS, both marketing
systems for Russian and English translation; Stylus also for French, and PARS also
for Ukrainian. Other PC-based systems from Europe include: Hypertrans for
translating between Italian and English; the Winger system for Danish-English,
French-English and English-Spanish, now also marketed in North America; and
TranSmart, the commercial version of the Kielikone system, for Finnish-English
translation.
Vendors of older mainframe systems (Systran, Fujitsu, Metal, Logos) are
being obliged to compete by downsizing their systems; many have done so with
success, managing to retain most features of their mainframe products in the PC-
based versions. Systran Pro and Systran Classic, for example, are Windows-based

9
versions of the successful system developed since the 1960s for clients worldwide in a
large range of languages; the large dictionary databases offered by Systran give these
systems clear advantages over most other PC products. Both Systran Classic (for
home use) and Systran Pro (for use by translators) are now sold for under a five
hundred dollars in many language pairs: English-French, English-German, English-
Spanish; and for English to Italian and Japanese to English. The publishing company
Langenscheidt acquired rights to sell a version of METAL, in collaboration with
GMS (Gesellschaft für Multilinguale Systeme, now owned by Lernout & Hauspie) –
the system is called ‘Langenscheidt T1’ and offers various versions for German and
English translation. Also from Germany is the Personal Translator, a joint product of
IBM and von Rheinbaben & Busch, based on the LMT (i.e. Logic-Programming
based Machine Translation) transfer-based system under development since 1985.
LMT itself is available as a MT component for the IBM TranslationManager. Both
Langenscheidt T1 and the Personal Translator are intended primarily for the non-
professional translator, competing therefore with Globalink and similar products. (For
these developments see proceedings of MT conferences: AMTA, EAMT, MT
Summit, and MT News International.)
Sales of commercial PC translation software have shown a dramatic rise.
There are now estimated to be some 1000 different MT packages on sale (when each
language pair is counted separately.) The products of one vendor (Globalink) are
present in at least 6000 stores in North America alone; and in Japan one system
(Korya Eiwa from Catena, for English-Japanese translation) is said to have sold over
100,000 copies in its first year on the market. Though it is difficult to establish how
much of the software purchased is regularly used (some cynics claim that only a very
small proportion is tried out more than once), there is no doubting the growing
volume of ‘occasional’ translation, i.e. by people from all backgrounds wanting
renderings of foreign texts in their own language, or wanting to communicate in
writing with others in other languages, and who are not deterred by poor quality. It is
this latent market for low-quality translation, untapped until very recently, which is
now being discovered and which is contributing to massive increases in sales of
translation software.

MT on the Internet
At the same time, many MT vendors have been providing network-based
translation services for on-demand translation, with human revision as optional extras.
In some cases these are client-server arrangements for regular users; in other cases,
the service is provided on a trial basis, enabling companies to discover whether MT is
worthwhile for their particular circumstances and in what form. Such services are
provided, for example, by Systran, Logos, Globalink, Fujitsu, JICST and NEC.
Some companies have now been set up primarily for this purpose: LANT in
Belgium is a major example, based on its rights to develop the METAL system and on
the Eurolang Optimizer, which it also markets (Caeyers 1997). Its speciality is the
customisation of controlled languages for use with its MT and translation memory
systems. In late 1997 it launched its multilingual service for the translation of
electronic mail, Web pages and attached files. And in Singapore, there is MTSU
(Machine Translation Service Unit of the Institute of Systems Science, National
University of Singapore), using its own locally-developed systems for translation
from English into Chinese, Malay, Japanese and Korean (with Chinese its main
strength) and with editing by professional translators. The service is providing large
scale translation over the Internet for many customers world wide (mainly

10
multinational organisations), and including much of the localisation needs for
software companies in the Chinese-language markets (LISA Forum Newsletter 4(3),
August 1995, p.12.)
A further sign of the influence of Internet is the growing number of MT
software products for translating Web pages. Japanese companies have led the way:
nearly all the companies mentioned above have a product on this lucrative market;
they have been followed quickly elsewhere (e.g. by Systran, Globalink, Transparent
Language, LogoVista). As well as PC software for translating Web pages, we are now
seeing Internet services adding translation facilities: the most recent example is the
availability on AltaVista of versions of Systran for translating French, German and
Spanish into and from English – with what success or user satisfaction it is too early
to say (Yang and Lange 1998).
Equally significant has been the use of MT for electronic mail and for ‘chat
rooms’. Two years ago CompuServe introduced a trial service based on the Transcend
system for users of the MacCIM Support Forum. Six months later, the World
Community Forum began to use MT for translating conversational e-mail. Usage has
rocketed (Flanagan 1996). Most recently, CompuServe introduced its own translation
service for longer documents either as unedited ‘raw’ MT or with optional human
editing. Soon CompuServe will offer MT as a standard for all its e-mail. As for
Internet chat, Globalink has joined with Uni-Verse to provide a multilingual service.
The use is not simple curiosity, although that is how it often begins.
CompuServe records a high percentage of repeat large-volume users for its service,
about 85% for unedited MT – a much higher percentage than might have been
expected. It seems that most is used for assimilation of information, where poorer
quality is acceptable. The crucial point is that customers are prepared to pay for the
product – and CompuServe is inundated with complaints if the MT service goes
down!
It is clear that the potential for MT on, via and for the Internet is now being
fully appreciated – no company can afford to be left behind, and all the major players
have ambitious plans, e.g. Lernout & Hauspie (McLaughlin and Schwall 1998), which
has now acquired MT systems from Globalink, Neocor and AppTek as well as the old
METAL system (from GMS).

Future needs and developments

Despite the recent growth of systems for personal computers and of Internet
services, it is still true to say that there is nothing yet really suitable for the
independent professional translator, i.e. for those not working for large companies or
in translation organizations. It is known that some translators have tried to apply
commercial PC-based software to their needs, but the amount of adaptation required
and the generally poor output has made them unsatisfactory and uneconomic. More
suitable for the independent translator would be a cost-effective translation
workstation. However, current workstations on the market are still too expensive for
the individual translator. Although there is promise of low-cost computer tools for this
potentially large market – e.g. terminology and concordancing software, and perhaps
alignment software – there is no doubt that this segment is not being covered as well
as many other areas.
Another area at present poorly served is the need for reliable but low-cost
translation of documents into unknown foreign languages where users do not want to
engage expert bilingual translators. There is no problem with translation into
recipients’ own languages – PC systems can give adequate ‘rough’ versions for users

11
to get some idea of the basic message – but for translation into an unknown language
there are still no solutions. There have been recently some cheap Japanese products
which serve this specific ‘foreign language authoring’ demand in the case of writing
business letters (based on standard phrases and document templates), but for other
areas and for longer documents, where there is less ‘stereotyping’, there is nothing as
yet. For translation into another language unknown (or poorly known) by the sender,
what is really required is software which can be relied upon to provide good quality
output (and most PC products are not good enough). A number of research groups are
investigating interactive systems, where the sender composes an MT-friendly version
of a letter or document in collaboration with the computer. With a sufficiently
‘normalised’ input text, the MT system can guarantee grammatically and stylistically
correct output. As yet, however, this work (e.g. at GETA in France) is still at the
laboratory stage (Boitet and Blanchon 1995).
The same is true for software combining MT with information access,
information extraction, and summarisation software. There are no commercial
systems yet on the market; developments are still at the research stages. The potential
and the demand has been recognised: for example, in recent years, most research
funds of the European Union have been focused not on MT or ‘pure’ natural language
processing (as it was during the 1980s), but on projects for multilingual tools with
direct applications in mind; many involve translation of some kind, usually within a
restricted subject field and often in controlled conditions (Hutchins 1996; Schütz
1996). As just one example, the AVENTINUS project is developing a system for
police forces in the area of drug control and law enforcement: information about
drugs, criminals and suspects will be available on databases accessible in any of the
European Union languages.
There is growing interest in such multilingual applications worldwide. The
application that has received most attention has been ‘cross-language information
retrieval’, i.e. software enabling users to search foreign language databases in their
own languages. As yet most work has focussed on the construction and operation of
appropriate translation dictionaries, for the matching of query words against words or
phrases in document databases (Bian and Chen 1998, Oard 1998) – although the
provision of software for fast translation of original texts into the enquirer’s own
language is naturally also envisaged (McCarley and Roukos 1998). Clearly it will not
be long before commercial software is available for this application.
The future application that is probably most desired by the general public is
the translation of spoken language. But, from a commercial (and even research)
perspective, the prospects for automatic speech translation are still distant (Krauwer et
al. 1997). It was only in the 1980s that developments in speech recognition and
synthesis made spoken language translation a feasible objective. In Japan a joint
government and industry company ATR was established in 1986 near Osaka, and it is
now one of the main centres for automatic speech translation. The aim is to develop a
speaker-independent real-time telephone translation system for Japanese to English
and vice versa, initially for hotel reservation and conference registration transactions.
Other speech translation projects have been set up subsequently. The JANUS system
is a research project at Carnegie-Mellon University and at Karlsruhe in Germany. The
researchers are collaborating with ATR in a consortium (C-STAR), each developing
speech recognition and synthesis modules for their own languages (English, German,
Japanese). (One by-product of this research was mentioned earlier: the rapid-
deployment project for custom-built systems in less-common languages.) The fourth
major effort in speech translation is the long-term VERBMOBIL project funded by

12
the German Ministry for Research and Technology which began in May 1993. The
aim is a portable aid for business negotiations as a supplement to users’ own
knowledge of the languages (German, Japanese, English). Numerous German
university groups are involved in fundamental research on dialogue linguistics, speech
recognition and MT design; a prototype is nearing completion, and a demonstration
product is targeted for early in the next century.
Speech translation is probably at present the most innovative area of
computer-based translation research, and it is attracting most funding and the most
publicity. However, few experienced observers expect dramatic developments in this
area in the near future – the development of MT for written language has taken many
years to reach the present stage of widespread practical use in multinational
companies, a wide range of PC based products of variable quality and application,
growing use on networks and for electronic mail. Despite today’s high profile for
written-language MT, researchers know that there is still much to be done to improve
quality. Spoken-language MT has not yet reached even the stage of real-time testing
in non-laboratory settings.

Comparison of human and machine translation

From this survey it should be apparent that the application of computers to the
task of translating natural languages has not been and is unlikely to be a threat to the
livelihood of professional translators. Those skills which the human translator can
contribute will continue always to be in demand. There is no prospect, for example,
that machine translation could ever attempt the translation of literary or legal texts. By
contrast, for the rough translation of electronic texts on the Internet there is no rivalry
for machine translation – human translators cannot compete in terms of speed, even if
they were prepared to undertake poor quality translation of ephemeral material.
We may compare the relative merits of human and machine translation
according to the categories of need and use outlined at the beginning of this paper. As
far as the dissemination function (production of publishable translations) is
concerned, human translation is more satisfactory and less costly overall whenever it
is a question of translating one particular text in a unique subject domain (whether
scientific, technical, medical, legal or literary). Machine translation demands the
costly investment of dictionary maintenance and updating and the costly involvement
of post-editing. This can be justifiable (i.e. cost-effective) only when large volumes of
documentation within a particular domain are being translated. It is even more
justifiable if translation is into more than one target language (when pre-editing and/or
vocabulary and grammar control of original texts is possible), and when there is
considerable repetition. For such tasks, the human translator would be overwhelmed
by the scale of the task, by the boring repetitiveness and by the need to maintain
terminological consistency. By contrast, the computer can handle large volumes and
can automatically maintain consistency. In brief, machine translation is ideal for large
scale and/or rapid translation of (boring) technical documentation, (highly repetitive)
software localisation manuals, and real-time translation of weather reports. The
human translator is (and will remain) unrivalled for non-repetitive linguistically
sophisticated texts (e.g. in literature and law).
For the translation of texts for assimilation, where the quality of output can be
poorer than that for texts to be published, it is clear that machine translation is an ideal
solution. Human translators are not prepared (and resent being asked) to produce
‘rough’ translations of scientific and technical documents that may be read by only
one person who wants to merely find out the general content and information and is

13
unconcerned whether everything is intelligible or not, and who is certainly not
deterred by stylistic awkwardness or grammatical errors. Of course, they might prefer
to have output better than that presently provided by most MT systems, but if the only
alternative option is no translation at all then machine translation is fully acceptable.
For the interchange of information, there may still in the future continue to be
a role for the human translator in the translation of business correspondence
(particularly if the content is sensitive or legally binding). But for the translation of
personal letters, MT systems are likely to be increasingly used; and, for electronic
mail and for the extraction of information from Web pages and computer-based
information services, MT is the only feasible solution.
For spoken translation, by contrast, there will be a continuing market for the
human translator. There is surely no prospect of automatic translation replacing the
interpreter of diplomatic and business exchanges. Although there has been research on
the computer translation of telephone enquiries within highly constrained domains,
and future implementation can be envisaged in this area, for the bulk of telephone
communication there is unlikely to ever be any substitute for human translation.
Finally, MT systems are opening up new areas where human translation has
never featured: the production of draft versions for authors writing in a foreign
language, who need assistance in producing an original text; the on-line translation of
television subtitles, the translation of information from databases; and no doubt, more
such new applications will appear in the future. In these areas, as in others mentioned,
there is no threat to the human translator because they were never included in the
sphere of professional translation. There is no doubt that MT and human translation
can and will co-exist in harmony and without conflict.

References:

AMTA (1996): Expanding MT horizons: Proceedings of the Second Conference of

the Association for Machine Translation in the Americas, 2-5 October 1996,
Montreal, Canada. [Washington, D.C.: AMTA]
AMTA (1998): Machine translation and the information soup: Third Conference of
the Association for Machine Translation in the Americas… Langhorne, PA,
USA, October 1998: Proceedings, ed. D.Farwell, L.Gerber [and] E.Hovy.
Berlin: Springer-Verlag, 1998. (Lecture Notes in Artificial Intelligence 1529)
Arnola, H. (1996): “Kielikone machine translation technology and its perspective on
the economics of machine translation.” In: EAMT Workshop (1996), 73-88.
Bian, G.W. and Chen, H.H. (1998): “Integrating query translation and document
translation in a cross-language information retrieval system.” In: AMTA (1998),
250-265.
Boitet, C. and Blanchon, H. (1995): “Multilingual dialogue-based MT for
monolingual authors: the LIDIA project and a first mockup.” Machine
Translation 9(2), 99-132.
Brace, C., Vasconcellos, M. and Miller, L.C. (1995): “MT users and usage: Europe
and the Americas.” MT News International 12 (October 1995), 14-19.
Caeyers, H. (1997): “Presentation of LANT technology.” In: MT Summit (1997),
253-254.
Chandioux, J. and Grimaila, A. (1996): “'Specialized' machine translation.” In:
AMTA (1996), 206-211.
EAMT Workshop (1996): EAMT Machine Translation Workshop… Vienna, Austria,
29-30 August 1996. Proceedings. [Geneva: EAMT].

14
EAMT Workshop (1997): Language technology in your organization? 1997 EAMT
Workshop, Copenhagen, 21-22 May 1997. Proceedings. [Geneva: EAMT]
Elliston, J.S.G. (1979) “Computer-aided translation – a business viewpoint.”
Translating and the Computer: proceedings of a seminar, London, 14th
November 1978; ed. B.M.Snell (Amsterdam: North-Holland), 149-158.
Flanagan, M. (1996): “Two years online: experiences, challenges, and trends.” In:
AMTA (1996), 192-197.
Frederking, R.E., Brown, R.D. and Hogan, C. (1998): “The DIPLOMAT rapid-
deployment speech MT system.” In: MT Summit (1997), 261-262.
Heyn, M. (1996): “Integrating machine translation into translation memory systems.”
In: EAMT Workshop (1996), 111-124.
Humphreys, L. (1996) “Use of linguistic resources like translation memories in
machine translation systems.” In: EAMT Workshop (1996), 101-110.
Hutchins, W.J. (1986): Machine translation: past, present, future. Chichester: Ellis
Horwood.
Hutchins, W.J. (1993): “Latest developments in machine translation technology:
beginning a new era in MT research.” In: MT Summit (1993), 11-34.
Hutchins, W.J. (1994): “Research methods and system designs in machine translation:
a ten-year review, 1984-1994.” Machine Translation: Ten Years On.
Proceedings of the second international conference…held at Cranfield
University, England, 12-14 November 1994…ed. Douglas Clarke and Alfred
Vella. Cranfield: Cranfield University, 1998. Ch.4.
Hutchins, W.J. (1996): “The state of machine translation in Europe.” In: AMTA
(1996), 198-205.
Krauwer, S. et al. (1997), eds.: Spoken language translation. Proceedings of a
Workshop sponsored by the Association of Computational Linguistics and by
the European Network in Language and Speech, 11 July 1997… Madrid, Spain.
[Somerset, N.J.: ACL].
Lee, A. (1994); “Controlled English with and without machine translation.” Aslib
Proceedings 46(5), 131-133.
Leon, M. and Aymerich (1997): “PAHO systems: SPANAM and ENGSPAN.” In:
MT Summit (1997), 259-260.
Lewis, T. (1997) “MT as a commercial service: three case studies.” In: MT Summit
(1997), 219-223.
McCarley, J.S. and Roukos, S. (1998): “Fast document translation for cross-language
information retrieval.” In: AMTA (1998), 150-157.
McLaughlin, S. and Schwall, U. (1998): “Spicing up the information soup: machine
translation and the Internet.” In: AMTA (1998), 384-397.
Mitamura, T. and Nyberg, E.H. (1995): “Controlled English for knowledge-based
MT: experience with the KANT system.” TMI-95: Proceedings of the Sixth
International Conference on Theoretical and Methodological Issues in Machine
Translation, July 5-7, 1995, Leuven, Belgium; 158-172.
MT Summit (1993): MT Summit IV: international cooperation for global
communication, July 20-22, 1993, Kobe, Japan. [Tokyo: AAMT.]
MT Summit (1995): MT Summit V, Luxembourg, July 10-13, 1995. Proceedings.
[Brussels: SEMA.]
MT Summit (1997): MT Summit VI: past, present, future. 29 October – 1 November
1997, San Diego, California. Proceedings, ed. By V.Teller and B.Sundheim.
Washington, D.C.: Association for Machine Translation in the Americas.

15
Oard, D.W. (1998): “A comparative study of query and document translation for
cross-language information retrieval.” In: AMTA (1998), 472-483.
O’Neill-Brown, P. (1996): “JICST Japanese-English machine translation system.” In:
AMTA (1996), 257-260.
Ǿrsnes, B., Music, B. and Maegaard, B. (1996): “PaTrans – a patent translation
system.” Coling 96. Proceedings… Copenhagen August 1996, 1115-1118.
Pedtke, T.R. (1997): “U.S. government support and use of machine translation:
current status.” In: MT Summit (1997), 3-13.
Schaeler, R. (1996): “Machine translation, translation memories and the phrasal
lexicon: the localisation perspective.” In: EAMT Workshop (1996), 21-33.
Schütz, J. (1996): “European research and development in machine translation.” MT
News International 15 (October 1996), 8-11.
Senez, D. (1996): “Machine translation feasibility study at the European
Commission.” In: EAMT Workshop (1996), 63-70.
Theologitis, D. (1997): “EURAMIS, the platform of the EC translator.” In: EAMT
Workshop (1997), 17-32.
Toole, J., Turcato, D., Popowich, F., Fass, D. and McFetridge, P. (1998): “Time-
constrained machine translation.” In: AMTA (1998), 103-112.
Van der Steen, G. and Dijenborgh, B.J. (1992): “Online collection and translation of
industrial texts.” Translating and the Computer 14… 10-11 November 1992,
London. (London: Aslib), 135-164.
Vasconcellos, M. (1993): “The present state of machine translation usage technology;
or: How do I use thee? Let me count the ways.” In: MT Summit (1993), 47-62.
Yang, J. and Lange, E. (1998): “SYSTRAN on AltaVista: a user study on real-time
machine translation on the Internet.” In: AMTA (1998), 275-285.

16