We’re talking with the person responsible for the original idea and the linguistic analysis for VICTOR Translator. A linguist and translator for the European Commission in Brussels for more than 25 years, she was the inspiration for this English-to-Spanish machine-translation project commissioned by the translation company, Berca Translator and developed by MTC Soft in Granada, Spain.
Question: There are many machine translators on the market, some of them developed by industry giants. Why was it necessary to invent another one? What made you think that you were the right person to undertake this adventure? How did you expect to achieve with the help of a tiny Spanish software company what some of the world’s most important corporations hadn’t achieved with unlimited resources?
Answer: I’ve got 25 years’ experience in professional translating, 10 of them translating into Spanish original texts from English, French and Italian using some of the finest machine translating systems on the market. I participated in the development of some of those systems. For that reason I am well acquainted with the problems and limitations of the existing systems.
That’s what prompted me to develop an analysis which, concentrating on the specific stumbling blocks in the process of English-to-Spanish translations, would permit programmers to overcome those problems from its very conception. That is to say, an application capable of overcoming the obstacles which machine-translation programs face every working day:
- Functional-analysis faults (frequent in the English-Spanish language pair, since the process proceeds from a Saxon language to a Latin one.
- Syntactic-analysis deficiencies (identification of a sentence’s true subject, or the semantic or syntactic links between the different parts of the sentence.
- The necessity to foresee, for a single English Word, different translations according to the context.
- The necessity to give adequate treatment to idioms of the source language, avoiding literal translation while substituting them for their corresponding idioms in the target language (this both in expressions and in verbs)
- The necessity, particularly in an English-to-Spanish translation system, to obtain a syntactic reformulation in specific cases, such as the addition of linking conjunctions or prepositions, or relative pronouns, the conversion of a sentence from the passive mode in English to reflexive in Spanish, etc.
Our greatest advantage in this project is that we’re starting from a lengthy practical experience which not even the universities (for their principally theoretical approach) nor the great computer-science companies (for their mainly statistical approach) seem to have counted on until now. It’s our intention to use statistical information only to solve the problems which exhaustive linguistic analysis is not capable of resolving. No linguistic analysis can determine if “Harry’s bar” is the bar which belongs to Harry or the famous “Harry’s bar” in Venice. In this case we are also obliged to resort to statistics.
Presumably one doesn’t go lightly into a project which is going to last for years. What inspiration moved you to start? What final objectives did you have from the beginning?
My objective from the outset was to achieve a system which would resolve all of the “automatic” part of the translation process, which constitutes—I know from experience—a very high percentage of my daily work. Then there is the advantage that once a machine has translated an expression correctly, or produces a correct analysis, it will always do it the same way. I think the effort we dedicate to achieving a program capable of carrying out this task must necessarily be more productive than dedicating day after day resolving repeatedly the same problems in successive texts to be translated.
How much time have you invested in creating the linguistic analysis necessary to create this program?
Approximately five years for the abstract conception and another five for the practical part, working with the programmers, and adapting my work to theirs, often on a trial-and-error basis as they went along developing the program.
Please explain to us a bit of your procedure in creating your analysis. What were the first issues to overcome? Why hadn’t they been solved before by the creators of earlier machine-translation programs? I mean, machine translation has been around for a long time, and a lot of money has been spent on it. What happened?
I think our program has various strong points:
- A very efficient functional and syntactic analysis, based on logic, but also on practice (compatibility, probability, tendency of the speaker to obviate ambiguity, etc.)
- The possibility for the massive, automatic creation of specific glossaries. That is to say, of groups of set expressions.
- The posibility to convert English idiomatic verbs into the corresponding Spanish idiomatic verbs. I don’t believe any other program does that today.
- The proper management of English phrasal verbs, extremely complex and subtle.
- The translation or ad hoc addition of the different prepositions according to the specific translation which Spanish assigns to the original verb: that which we refer to in VICTOR Translator as
“subordinated” and “added” prepositions. - Finally, the possibility to condition the translation of any Word (from among all possible translations) to the context in which it appears, with the option of delimiting that context almost exhaustively.
How did the contact with Fernando Moreno-Torres y MTC Soft come about? Coincidence, friendship, or a systematic search?
The origins of our cooperation are in mutual confidence, both personal and professional.
What are the usual complaints that professional translators have against machine translators?
The use of automatic translators began around 10 years ago. For a professional translator the principal problem with today’s machine translators is that sometimes it takes more work to “fix” the phrase that the machine proposes than to do the translation yourself.
Today’s machine translators, conceived, I think, from lower levels of complexity than those of the process of translating (perhpas because the lacked translating experience from the beginning) cannot resolve all of the problems presented in real translation experiences. When I started working on this new translator I started out from the point of view that a program which offers functional and syntactic analysis is contributing something valuable to the human translator. If it also obtains the appropriate terminology, a correct linking of the different parts of the sentence, correct prepositions and correct idioms, the material which the automatic translator offers to the human professional translator will contribute to an important saving in time and resources.
How has your collaboration with MTC Soft gone? Was it easy for them to translate your natural-language linguistic analysis into a computer program?
No, it hasn’t been easy, but MTC Soft has put two magnificent analysts with many years of experience on the project, and they have been able to translate the linguistic logic which underlies the program into computer language. They have quickly captured the problems and contradictions which have arisen and contributed elegant computer solutions.
How does one feel seeing the VICTOR project all the way through to the beta version?
Sometimes I feel impatient, because there’s still work to do on the next version which will incorporate statistical analysis. But, seeing the really brilliant results achieved up till now, I feel a tremendous satisfaction.
Do you hope to see your translator colleagues using VICTOR in their daily work?
Yes, I hope so.