Copyright © Thomas D. Hedden, 1992-2010

"During the TMI-92 conference in Montreal, Jaime Carbonell gave some details of the contract signed in May 1992 between Caterpillar, the world's largest manufacturer of earth-moving equipment, and the Center for Machine Translation at Carnegie-Mellon University for the development of a fully automatic translation system. The five-year multimillion dollar contract had been concluded after an extensive evaluation by Caterpillar since the 'proof-of-concept' demonstration by the CMU team in June 1991..." (MT News International 3:11 [September 1992]).

News articles such as this show that whether machine translation is already here, coming soon, coming in the distant future, or not coming at all, it is winning big contracts. Therefore, whether we think it is laughable, impractical, a hoax, whatever we may think about it, it is an issue which must be addressed.


Machine translation (MT) means translation using computers. In its broadest sense MT can be understood to include such computer applications as compilers and compression programs, etc., which convert a file in one computer language into a file in another computer language. However, what we are interested in here is natural language processing (NLP). One thing which MT does not mean, but which is sometimes confused with MT, is automatic speech recognition.

There are four basic types of translation, three of which are types of machine translation or machine-assisted (-aided) translation:

Human translation. A human translator performs all the steps in the translation process, using a computer only as a word processor, if at all.

Machine-assisted (-aided) human translation (MAHT). The translation is performed by a human translator, but he/she uses the computer as a tool to improve or speed up the translation process. This is called computer-assisted (-aided) translation (CAT) by people in the field of translation as opposed to the field of MT.

Human-assisted (-aided) machine translation (HAMT). The source language (SL) text is modified by a human translator either before, during, or after it is translated by the computer.

Fully automatic (automated) machine translation (FAMT). The SL text is fed into the computer as a file, and the computer produces a translation automatically without any human intervention. This is sometimes referred to as batch mode. There are two types of fully automatic machine translation: fully automatic high-quality machine translation (FAHQMT) and low-quality machine translation.

The type of MT which people think of when they hear the word machine translation is usually the last type (fully automatic MT). The term MT will be used loosely in this report to mean MAHT, HAMT, or FAMT. Obviously the ultimate goal of MT is FAMT, although its results have also made it the subject of much ridicule, sometimes with good reason. Note that the distinction between HAMT and FAMT is partly conventional, since a FAMT system would be considered a HAMT system if the output is post-edited, and any translation can be checked.

MT is a part of the field of knowledge called artificial intelligence (AI). There are many different definitions about what artificial intelligence is, however a simple way of thinking of it is the attempt to emulate human patterns of thinking and behavior using computer models. Today AI is used in robotics, pattern recognition, etc., as well as in MT.


Apparently the first suggestions concerning MT were made by the Russian Smirnov-Troyansky and the Frenchman G.B. Artsouni in the 1930's. However, the first serious discussions were begun in 1946 by the mathematician Warren Weaver. He and many others were inspired by the success of the Allied efforts using the British Colossus computer to break the German military code produced by the Enigma machine, and the obvious similarity between the task of decoding an encoded message and the task of translation of one language into another. By 1954 there was a MT project at Georgetown University which succeeded in correctly translating several sentences from Russian into English. Soon there were MT projects at MIT, Harvard, and the University of Pennsylvania.

In 1964, after more than $20,000,000 had been invested by the Federal Government in MT, the National Academy of Sciences commissioned the Automatic Language Processing Advisory Committee (ALPAC) to write a study of the status of MT. The committee, headed by John R. Pierce, wrote a now-famous report in which it expressed doubt that a fully-automatic MT system could ever be produced. That report sounded the death-knell for funding of MT research, and MT was neglected for many years afterwards.

The reasons for this failure have been described many times, and come down to the fact that the analysis by humans of messages in natural language relies to some extent on information which is not present in the words which make up the message. This led the linguist Yehoshua Bar-Hillel to declare that MT was impossible. The example which he provided has since become a classic, and is now called the Bar-Hillel paradox:

     The pen is in the box. 
        [i.e. the writing instrument is in the container]
     The box is in the pen.
        [i.e. the container is in the playpen or the pigpen]

There are two possible ways that a person could correctly infer the meaning of these sentences. First, if there is a context preceding these sentences, it could make clear which meaning of pen is being used in which sentence. That is, the meaning of the words and information about the context is carried over from one sentence to the next. There is now an entire branch of linguistics, called discourse analysis, devoted to the study of how context affects the meaning of words and sentences. In order to infer in this way the correct meaning of an ambiguous sentence, computers will have to learn how to "remember" a context and make use of it to interpret the correct meaning of words and sentences within that context.

However, in the examples given above, most humans can understand the meaning correctly without any context. In order for a fully automatic MT system to translate these sentences correctly, the following information would have to be available to the computer:

     pens [writing instruments] are smaller than boxes
     boxes are bigger than pens [writing instruments],
        but smaller than pens [playpens, pigpens, etc.] 
     it is impossible for a bigger object to be inside
        a smaller object

Thus, one way or the other, whether the correct meaning of the sentences is inferred based on the context or in isolation, it is necessary for the computer to have information at its disposal which is not included in the message itself. During the early days of MT this realization was enough to make MT seem an impossible task.

Interest in MT revived in the 1980's, following dramatic advances in computer hardware (storage capacity, speed, etc.) and software (LISP, etc.). The need to store and process tremendous amounts of real-world knowledge in order to analyze a single word in the message ceased to be an impediment to design and use of MT systems.


The theory of MT is very complicated, and I will not go into in detail. However, there are several basic theoretical approaches to MT.

The most primitive is called the direct MT strategy. This approach is always between pairs of languages. This approach is based on good glossaries and morphological analysis.

The next most advanced system is called the transfer MT strategy. This is still being used today, although it has a competitor (see below). First, the SL is parsed into an abstract internal representation. Thereafter, a 'transfer' is made into the corresponding structures in the target language (TL). Then a translation is generated. This approach is more advanced theoretically, but also translates between specific pairs of languages. Both the direct MT strategy and the transfer MT strategy can take advantage of similarities between languages. The direct MT strategy has been criticized as theoretically inelegant, although it probably comes closest to modelling how human translators work.

Both the direct MT strategy and the transfer MT strategy have been criticized since they require that separate translation software be written for every language combination. (Actually, from the point of view of someone with experience in the translation industry, these concerns seem trivial, since an overwhelming majority of translation work is done in less than a dozen language combinations: Ten or so packages could be written for these combinations, and other combinations could continue to be translated manually.)

The most advanced system is called the interlingua MT strategy. The idea behind this approach is to create an artificial language, known as the interlingua, which shares all the features and makes all the distinctions of all languages. To translate between two different languages, an analyzer is used to put the SL into the interlingua, and a generator converts the interlingua into the TL. The proponents of this system argue that it reduces the number of analyzers and generators which are required, since only one generator and one analyzer is required for each language, no matter how many other languages there are. While this is true, I think that the proponents of the interlingua MT strategy probably underestimate how complex an interlingua would have to be in order to work as an intermediary among many or even a few unrelated languages, as opposed to among the few related European languages on which most work has been done. If the interlingua is extremely complex, this also means that the analyzers and generators will have to be extremely complex. It is hard to avoid thinking that this approach was inspired by the idea of a universal language such as Esperanto, which has excited certain linguists for centuries.

Other strategies

There is a fairly new approach called knowledge-based machine translation (KBMT), which is similar to the interlingua approach, in that the SL is converted into an intermediate form independent of any specific language. It differs in that the intermediate form is of a semantic nature rather than a syntactic nature.

Writers of MT systems have also explored the relative frequency of the various meanings of words with multiple meanings, and have attempted to include a great deal of real-world information in glossaries. For a more detailed discussion of the theory of MT see Nirenburg (1987).

The four types of translation.

We are not concerned here with human translation, so we will not discuss it.

MAHT can range from automatic look-up programs to systems which are practically fully automatic, but which require the translator to approve each sentence. Examples of some of the more successful of this type of software are the Translators WorkBench of Trados and INK Tools. Click here or select the appropriate number for a list of a few more MAHT tools.

HAMT also covers a broad range of systems. Human intervention can mean pre-editing the SL text by a person skilled at using the machine translation system in order to make the SL easier for the computer to understand, or it can mean interactive intervention, in which the translator may be asked questions about the meaning of the SL text by the computer. Some of the most irritating MT systems have used this approach, requiring the translator to sit in front of the computer terminal and answer such questions as:

     The word 'pen' means:
        1) a writing pen
        2) a play pen
        3) a pig pen

Human intervention can also mean post-editing to check the translation and fix mistakes made by the computer. It should be noted that the pre-editing and glossary compilation required for HAMT typically require a person who is a trained linguist who can parse the syntax of the sentence, not simply a translator who understands the foreign language and can express it in his/her own language.

Obviously the most primitive is the system which requires pre-editing, since the computer cannot handle the text unless a human converts NL into a semi-artificial language which is easier for the computer to understand. The ideal is when the automatic translation is so good that all that is necessary is to check the translation and change a few details. Interactive intervention can be anywhere in between.

Although there are FAMT systems, and although they may suit the needs of people who have to search through mountains of information and only need to get a very general idea of the contents of a document (a good example is provided by the low-quality needs of the military and the intelligence agencies), high-quality translation of truly natural language which is really fully automatic (automated) hardly exists. FAHQMT systems have requirements either for the compilation of extensive glossaries and/or are restricted to specific subworlds or sublanguages.

Survey of MT systems.

This part of this report is very incomplete. Hopefully it will be expanded later.

Common commercially-available systems

MAHT Systems

Trados GmbH)
	MultiTerm (concept-oriented, multi-lingual database)
	Translator's Workbench

INK International
	dictionaries (glossaries)
	LookUp (a resident program, works with common word processors)
	Texan (TEXt ANalyzer)
	TextTools ($700-$1,000)

	Concern Data Transoft (from Kurt Lücken Vertrieb in Bad Homburg)
	Stylus (can also run in batch mode)

Pan American Health Organization

HAMT Systems


Logos system.
	New Word Search
	Noun Phrase Search
	Revisions Processor
	ALEX (Automatic Lexicographer)
	Never EOS
	Merge and Restore Utilities
	Print Facilities
	Translation Facility
	The Logos system produces quite respectable results,
	but the amount of time which it is necessary to invest
	in building specialized glossaries is large.
	The Logos system can be a successful approach for large
	or continuing projects.

Intergraph's DP/Translator

Canadian Environmental Department's TAUM-MÉTÉO

FAMT Systems

Common "pocket calculator" translators
FinalSoft ($139)
Globalink ($998 per language direction)
MicroTac Spanish Assistant, etc. ($79.95 each)
PC-Translator ($985 per language direction)
Toltran ($498 per language direction)
Translate ($495 Eng-Span only, Span-Eng available soon)
Translator ($69.95 Eng-Span and Span-Eng only)

Experimental MT Systems

DLT (Distributed Language Translation)
SUSY (Saarbrücker Übersetzungssystem)

Partial survey of users of MT systems.

Antler Translations
AT&T's Document Development Organization, Winston-Salem, NC
	(Intergraph's DP/Translator)
Canadian Environmental Department (TAUM-MÉTÉO)
Caterpillar Tractor (working with Carnegie-Mellon University)
Chemical Abstracts
     Abstracts are written of the SL document in a very
     standardized style in the SL, then MT translates
     the abstract into English.
European Community (EUROTRA)
INK Group (INK Tools)
InterGraph (DP/Translator)
Pan-American Health Organization = PAHO (ENGSPAN(TM))
Paul H. Brink International, Minneapolis, MN (DP/Translator)
Siemens (METAL)
	Even Siemens has its own human translation department.
U.S. Air Force
	Wright Patterson AFB uses Systran
Xerox (Systran)

Problems with using MT systems.

Quality. Everyone has heard anecdotes about humorous translations produced by machine translations systems. One of the classic examples is the translation into Russian of the sentence "the spirit is strong, but the body is weak". This was translated literally as "the vodka is strong but the meat is rotten". Another is that the expression "Out of sight, out of mind" was translated as "Invisible, insane". Of course occasional mistakes per se are not a reason why MT should not be pursued, since human translators also make mistakes. There is the famous translation "male water sheep" (for "hydraulic ram"), which was produced by a human translator, not an MT system. The question is how frequent the mistakes are and how much time is required to fix them, vis-à-vis how much time and money are saved by using the MT system.

Ramp-up time. Another problem is the requirement for compiling glossaries before the systems can work. Here again, the requirement in and of itself is not a reason not to pursue MT, since human translators also compile glossaries. However, the requirements for the form of the glossaries are much more elaborate than are the requirements for glossaries which are good enough for a human translators. As was mentioned above, the requirements for glossary compilation are such that they cannot be met by a typical translator. The time spent compiling glossaries has to be weighed against the time saved by using the MT system. Obviously, it is not cost-effective to compile a glossary in order to use an MT system to translate a two-page personnel policy into Spanish. But just as obviously, it could pay off in the translation of a whole series of manuals on the same subject. The process of glossary creation can now be simplified by routines which will help identify those terms which are not in the glossary, and tools for building new glossaries on the basis of existing glossaries.

Reduced standardization and review of terminology. In the long run, glossaries compiled for use in-house cannot be of the same high quality as published dictionaries, since they are not widely distributed and exposed to criticism from thousands of users, the way published dictionaries are.

Free-lancers. A problem for translation agencies is that they rely heavily on free-lance translators, who may not have the software necessary to work on the project. They may not be able to afford it. They may not be willing to learn it. Being freelancers, they work with various agencies, and if every agency requires the freelancer to learn a different package, the translator may balk. Even if translators are given the software and even if they are willing to learn it, they may not have the appropriate hardware. Obviously, such systems can be used to their full potential only in-house. Using MT systems in-house requires in-house translators. Having in-house translators requires a predictable flow of work which can keep a translator busy. Note that this always has been a problem, and continues to be a problem even for translation agencies, not to mention end-users of translations. IBM's and AT&T's in-house translation departments began accepting work from outside sources for that very reason.

Poorly-formed SL text. If the SL original is badly written, as most manuals are, then the system will have difficulty translating it correctly, whereas a human translator can understand the intent of the writer, and produce a translation which is better than the SL original. In order to avoid this problem, some large users of MT, such as Caterpillar, have set up requirements for the style of the SL text. If the original writer did not adhere to these requirements it is necessary for the text to be edited. However, it can take almost as long for an editor to put the source language text into "standard" form as it would for a good human translator to translate it.

Reluctance of translators to use the system. A good case study of how implementing an in-house HAMT system can backfire is when a well-known technical translation company in Southern California adopted a such a system. The system adopted was an extremely tedious interactive system. In the end the best translators quit. (This company was later purchased by one of the largest translation agencies, but some of the original players have formed another company under a similar name.)

Size of translation assignment. Many assignments are simply too short to justify going to the trouble of using an MT system.

No file available. For many assignments, no file of the SL text is available. Although OCR can solve this problem to some extent, this is an additional step which requires additional manpower, additional software, and additional hardware, and must be factored into any calculation of the benefits of MT vs. using human translators.

Secrecy. The competitive advantage offered by a successful system and the enormous investment required to attain that advantage means that successful MT systems and their glossaries may be jealously guarded rather than released and widely shared. Contrast this state of affairs with that of human translators, the best of whom often have a finanacial incentive to compile and publish glossaries, thus improving the over-all level of knowledge in the industry.

Profitability of commercially-available MT systems.

Cost is an obvious consideration in a decision about whether to adopt MT systems. The ones which are cheap are worthless. The good ones range from extremely expensive to astronomically expensive. It takes a fairly high volume of suitable projects to pay off an investment of this magnitude. At least one MT system, the Logos system, charges on the basis of how many words you translate. The license for the smallest annual volume over the course of the first year would end up costing more than $0.20 per word. At higher annual volume the cost would go down to $0.06 per word over the course of the first year. These fees are what would be paid for the software license, and do not include time spend compiling glossaries or post-editing. Obviously, the fees at the end of the first year are not substantially cheaper than the fees of free-lance human translators, although these fees could decline dramatically over time.

The profitability of using FAMT and HAMT depends on size of project. The larger the project, the more profitable it is to do things in this way. Note: a large project does not necessarily mean a large document to be translated at one time, but rather a large amount of material in the same subject matter to be translated for the same client. However, given the lack of loyalty of clients for translation agencies, the reality is that the agency has to receive the assignment on a single contract in order to justify making such a large investment.

One of the advantages offered by MAHT tools over FAMT and HAMT is that it can be profitably be applied to small- to mid-sized projects, whether or not the file available.

Other considerations.

Speed. An HAMT or FAMT system which already has high-quality glossaries and is up and running can translate 75-100 pages per hour. This is obviously worth thinking about when poorly organized clients need multiple manuals translated with ridiculously short turn-around times.

Consistency. One respect in which MT is superior to human translators is consistency. The computer may not choose the correct translation, but it will use the same word everywhere, as opposed to human translators who sometimes try to make the translation more interesting by using first one translation, then another. Consistency is extremely important in translations of software and of software documentation, and also makes it easier to make changes later, should it be decided to change one term to another.

PR value (showoff value). Since MT has a high-tech sound and look to it, it offers a lot of potential to impress gullible potential clients. Unfortunately there are a lot of huge clients who are easily tricked into believing the claims of MT companies. Of course, MT also has a very bad reputation among many people who have experience with it or who have heard horror stories about the poor quality, delays, etc., which can result from relying on MT. There was a horror story printed in NOTIS News in 1992. Thus, boasting about MT to potential clients is not always a good idea. Some agencies advertise that they offer MT, but that they also offer human translations when quality is of paramount importance.

Low-quality market. There is a real market for junk translations. That is not to say that clients want poor quality, but rather that they are satisfied with low quality, and if they can get a low-quality translation done quickly for a low price, that is what they will choose. Agencies which offer low-quality MT can satisfy these clients.

File format compatibility. What are the requirements for compatibility with existing operating systems and software, and are any limitations placed on which applications can be used before/after/during the translation? Do files have to be converted? How much work will this involve? Can existing glossaries be imported into the system's glossaries? How much work will be necessary to convert these glossaries?

Hardware. Some systems, such as Logos, require sophisticated equipment: a fairly powerful UNIX workstation (ca. $20,000). Some less sophisticated systems will work on PCs or Macs; one requires Windows NT.

Compatibility. Can the system be used with a network? Can the system interact with and take advantage of powerful glossary tools such as the glossary produced by the Canadian government Termium? This is especially important since in-house glossaries will never be a match for resources such as Termium. Will the system be able to read and take advantage of existing CD-ROM and on-line glossaries, etc., or will such resources have to be used separately?

Industry standard. Which system will be most widely adopted? (The best system to adopt is the one which will become standard.)

Learning curve. Especially as long as it remains uncertain which package will become the industry standard, it is important that the system not require too great an investment of time to learn.

Format. Are formatting/style codes preserved? (This is especially important if the file has to be converted.) Can the system interface with a common interchange formats? (RTF, FrameMaker's MIF, etc.) Will it be necessary for someone to reformat translations produced by a given system?

Free-lance access to MT. AT&T tried to get free-lancers to turn over their rush projects to them, which AT&T will have machine-translated, and then return to the translator for post-editing. If free-lance translators begin using MT themselves, then there is no point in subcontracting to human translators. However, the fact that almost no good translators actually do this provides fairly good evidence that human translation is actually cheaper than MT (otherwise translators could subcontract to MT vendors, and then pass off the work as their own and reap a profit by charging a higher rate).

Separation between research and reality. In the United States there is almost no connection between the world of theoretical research in MT and the real world of application (the only major exception being the work at Carengie Mellon for Catepillar). Theoreticians are mostly interested in discussions about methodology, syntactic parsers, etc., rather than the requirements of real-world applications and how to satisfy them. Research money and the energy of the best researchers are channelled into theory rather than into practical systems.

Areas which require more research.

Further in-house testing of MAHT tools and more thorough survey of commercially-available systems.

Predictions and recommendations concerning MT systems.

A Gallup survey of the Institute of Electrical and Electronic Engineers made in March of 1991 found that 60% of corporate members expected automatic simultaneous interpretation by 2010 (Lieberman 1992). This shows that, whatever our opinion of MT, in the mind of the public it is feasible. As was mentioned at the beginning of this report, big contracts are being handed out to companies to develop or use MT. Xerox has been using Systran for eleven years. If FAMT can actually be made to work well, it has the potential to put many translation agencies out of business and take over the market. Although most of us think that will not happen, even if FAMT cannot be made to work, it has the potential to put some translation agencies out of business before clients realize that MT is not living up to its promises. This calls for a response from the translation community.

There are four possible responses to the challenge posed by MT. The first is to debunk MT. This means refusing to adopt it and pointing out its problems and limitations to clients. This approach will work with some clients, but has the danger of making the debunker appear to be a Luddite, refusing to acknowledge the wave of the future. The second possible response is to embrace fully and hype MT. This means trying to learn a system well right away, using it as much as possible, and (hopefully) boasting about its successes. Again, this approach will work with some clients, but has the danger of backfiring, since many (potential) clients have had bad experiences with MT. A third response would be to offer both human translation and MT, like a few agencies do. The fourth approach, the one I recommend, involves understanding and being able explain the strengths and weaknesses of all systems. All of the four types of translation will have a niche.

Human translation will always be used for small jobs, exotic language combinations, etc.

MAHT is almost certainly the most practical solution for medium- to large-sized jobs when the source text is available in a file is available.

HAMT will probably be used in-house by large companies such as Caterpillar to translate large volumes of very similar material, when the company has the ability to control the SL text. When large companies give enough of such work to agencies, some large agencies will be able to use systems of this kind, but for many agencies it is unlikely that such systems would pay for themselves.

True FAMT will probably be marginalized as a tool for searching through huge volumes of material when the user only wants to have a general idea of what it says. The user can then have a good translation of the material done after the decision has been made that the subject matter warrants investing in a good-quality translation. A certain number of these systems will be sold as gimmicks to naive consumers. For the time being, FAHQMT remains a pipedream.


Association for Computational Linguistics. 1974- . Computational Linguistics. Cambridge, MA: MIT Press Journals.
Association for Computational Linguistics. 1974- . The FINITE STRING. Cambridge, MA: MIT Press Journals.
Association for Machine Translation in the Americas. (forthcoming). MT Yellow Pages.
Barr, Avron, and E.A. Feigenbaum, eds. 1981. The Handbook of Artificial Intelligence, vol. 1. Reading, MA: Addison-Wesley Publishing Company. [Stanford $27.95]
Carbonell, Jaime, et al. 1992. JTEC Panel Report on Machine Translation in Japan. Baltimore, MD: Loyola College in Maryland. [available from NTIS]
Crystal, D. 1987. The Cambridge Encyclopedia of Language, pp. 350-351. Cambridge, etc.: Cambridge University Press.
Firebaugh, M.W. 1988. Artificial Intelligence: a knowledge-based approach, pp. 262-272. Boston: Boyd & Fraser Publishing Co.
Gazdar, G., A. Franz, K. Osborne, and R. Evans. 1987. Natural Language Processing in the 1980s. A bibliography. (CSLI Lecture Notes, 12.) Stanford-Palo Alto-Menlo Park: Center for the Study of Language and Information.
Hutchins, W.J., and H.L. Sommers. 1992. An Introduction to Machine Translation. London, etc.: Academic Press (Hacourt, Brace Jovanovich Publishers). [Stanford $42.50]
International Association for Machine Translation. 1992- .
MT News International. Newsletter of the International Association for Machine Translation. Washington, D.C.:
Lieberman, E.J. 1992. Language Futures. Esperantic Studies 1992:Summer.3.1-2.
Mendoza, Rick. 1991. Translator's little helpers. Hispanic Business 1991:October.32-33.
Nirenburg, S. 1987. Machine Translation: theoretical and methodological issues. Cambridge, etc.: Cambridge UP.
Resnick, Rosalind. 1991. Language liberators. International Business 1991:December.61-62.
Spark Jones, K., and M. Kay. 1973. Linguistics and Information Science. Academic Press.
Tresman, Ian. 1991. Multilingual PC Directory: A guide to multilingual and foreign language products for IBM PCs and compatibles. Borehamwood, U.K.: Herts.
Winograd, T. 1984. Computer Software for Working with Language. Scientific American. Reprinted in W.l S-Y. Wang, Language, Writing and the Computer, 61-72. (New York: W.H. Freeman and Company, 1986.)


Ratliff, Evan. 2006. Me Translate Pretty One Day. Wired Magazine, 14.12 (December 2006).


American Society for Information Science (ASIS). 19??- . Address: 8720 Georgia Avenue, Suite 501, Silver Spring, MD 20910-3602.
Asia-Pacific Association for Machine Translation (formerly Japan Association for Machine Translation). 1992- . Address: 3F, Shibakoen Sanada Bldg, 3-5-12 Shibakoen, Minato-ku, Tokyo 105-0011 Japan, e-mail
Association for Information Management (ASLIB). 19??- . Address: Information House, 20-24 Old Street,London EC1V 9AP England tel. +44/71/253 4488; fax +44/71/430 0514
Association for Computational Linguistics (ACL). 1962- . PO Box 6090, Somerset, NJ 08875, USA. Tel. +1 (908) 873-3898, fax +1 (908) 873-0014. Email
Association for Machine Translation in the Americas. 19??- . Address: PMB 300, 1201 Pennsylvania Avenue, MW, Suite 300, Washington, DC 20004, USA, e-mail
Center for Machine Translation. Carnegie Mellon University.
Center for the Study of Language and Information (CSLI). CSLI/SRI International, 333 Ravenswood Avenue, Menlo Park, CA 94025; CSLI/Stanford, Ventura Hall, Stanford, CA 94305; CSLI/Xerox PARC, 3333 Coyote Road, Palo Alto, CA 94304.
European Association for Machine Translation. 19??- . EAMT Secretariat, c/o TIM/ISSCO, Université de Genève, Ecole de Traduction et d'Interpretation, 40 blvd du Pont d'Arve, CH-1211 Geneva 4, Switzerland. Email: or
International Association for Machine Translation. 19??- . Address: c/o AMTA.
Japan Electronic Dictionary Research Institute, Ltd. 19??- . Address: Mita Kokusai Building, 4-28 Mita 1-chome, Minato-ku, Tokyo 108, Japan.
Linguistic Data Consortium. 1992?- . University of Pennsylvania.
Microelectronics and Computer Technology Corporation (a consortium). 19??- . Austin, TX.
Microsoft NLP Research Group. 19??- . Address: Microsoft Corporation, One Microsoft Way, Richmond, WA 98052.
Click here for more interesting links


(see also Tresman 1991)

[Note: The following list was compiled in the early 1990's, and is now out of date.
The Compendium of Translation Software from the EAMT is more up-to-date and complete.]

Applications Technology, Inc. (AppTek)
Corporate Headquarters:
6867 Elm St. Suite 300
McLean, VA 22101
Tel: +1 (703) 821-5000
Fax: +1 (703) 734-5001

Basis Technology Corp.
150 CambridgePark Drive
Cambridge, MA 02140-2322 USA
Tel: +1 (617) 386-2000
Fax: +1 (617) 386-2020

Buro voor Systeemontwikkeling = BSO (Dutch software firm)

Concern Data Transoft [Stylus-also from K. L¨cken Vertrieb]
Utmarksvägen 33
S-802 91 Gävle
tel +46/26/11 63 60
fax +46/26/10 16 60

FinalSoft Corporation (Translate)
3900 N.W. 79th Avenue
Miami, FL 33166
tel 800/232-8228

Globalink, Inc.
9302 Lee Highway
Fairfax, VA 22031
tel 703/273-5600

Globalink Marketing, Inc.
6034 West Courtyard Drive, Suite 305
Austin, TX 78730
tel 800/324-2150; 512/338-2150
fax 512/338-2151; 512/338-2152

INK Tools
INK International
Prins Hendriklaan 52
1075 BD Amsterdam

INK INternational V.B.
Gebouw "De Amiraal"
Baarjesweg 224
AA Amsterdam
tel. +31/20/164 591
fax +31/20/163 851

Intergraph Corporation
Natural Language Group
Huntsville, AL 35894-0003

Kurt Lücken
Vertrieb für EDV-Literatur und EDV-Zubehör
Emmerichshohl 6
6380 Bad Homburg 6
06172/4 73 87
06172/4 59 76 3

Language Engineering Corporation
385 Concord Avenue
Belmont, MA 02178 USA
Tel: +1 (617) 489-4000
Fax: +1 (617) 489-3850

Language Weaver
4640 Admiralty Way, Suite 423
Marina Del Rey, CA 90292 USA
Tel: +1 (310) 437-7300
Fax: +1 (310) 437-7307

Logos Corporation
One Dedham Place, Suite 4
Dedham, MA 02026
tel 617/326-1595
fax 617/326-9341

TRADOS Corporation
113 South Columbus Street, Suite 400 
Alexandria, VA 22314
Tel     +1 (703) 683-6900 
Fax     +1 (703) 683-9457 
(For contact information outside the United States,
click here.)

MicroTac Software, Inc. (Spanish Assistant, French Asst., etc.)
4655 Cass Street, Suite 214
San Diego, CA 92109
tel 1-800-423-3556; 619/272-5700

Linguistic Products (PC-Translator)
The Woodlands, TX

Polygon Industries, Inc. (Translator)
New Orleans, LA

1, rue du Cimetière
95230 Soisy-sous-Montmorency
Tel: +33 (1) 39 34 97 97
Fax: +33 (1) 39 89 49 34

North America:
SYSTRAN Software, Inc.
9333 Genesee Avenue, Plaza Level, Suite PL1
San Diego, CA 92121-2112 USA
Tel.: +1 (858) 457-1900
Fax: +1 (858) 457-0648

Toin America Corporation
Atlanta Financial Center, Suite 1120
3353 Peachtree Road N.E.
Atlanta, GA 30326
tel 404/240-4110
fax 404/240-4111

Toin Corporation
3-9-1 Meguro Meguro-ku
Tokyo 153
tel 03/5721-3016
fax 03/5721-3261

Toltran Ltd.
Barrington, IL

Trados GmbH


[Note: The first MT conference took place at MIT in 1952.]

Annual Meeting of the Association for Computational Linguistics. Annual.
ATA Conference. Annual.
COLING. International Conference on Computational Linguistics.
Conference on Applied Natural Language Processing.
European Conference on Artificial Intelligence.
International Conference on Current Issues in Computational Linguistics.
International Conference on Theoretical and Methodological Issues in Machine Translation. Bi-annual.
International Workshop on Natural Language Generation.
MT Summit. Annual.
MT World.
Translation and the European Communities Conference.


American government.
Defense Advanced Research Projects Agency (DARPA).
Dutch government (10M to BSO)
Japanese government.
Japan Key Technology Center


Distributed Language Translation (DLT)
"Fifth Generation Project"
Electronic Dictionary Project (will cost > $100,000,000)
Cyc (pron. "psyche", from encyclopedia)

Click here to return to Thomas Hedden's home page.

This page is viewable with any browser.

Valid XHTML 1.0!

Valid CSS!