Learning From Translation Mistakes
As a former translator and reviser of translations, I find it very difficult to believe that a data processing system is really able to do the same job as a human translator. This is probably due to my lack of knowledge and understanding of how computers work. But whatever my incompetence in that field, I hope the examples I will draw from my experience in translation units will give you an interesting insight into some of the most frustrating problems encountered when transferring ideas from one language to another.
Taking part in the selection of candidates for translator jobs, I have often been amazed by the fact that a number of candidates with a perfect knowledge of both the source and the target languages and an impressive mastery of the relevant field could be very poor translators indeed. Why is that? One of the human factors is the lack of modesty. The translator's personality and intelligence interfere with the very humble task he has to perform. Instead of putting aside his own ideas, fantasies and style to follow blindly the author's, he embellishes, adds or transforms. This kind of problem, I suppose, cannot arise with a machine translator, although, being something of an Asimov fan, I may have my doubts: if machine translation is actually working, it must come close to the capabilities of Asimov's robots.
Anyway, besides humility, candidates must possess two other qualities that may be difficult to develop in machines, however sophisticated: judgment and flexibility.
By judgment I mean the ability to solve a problem through wide knowledge of the field, through awareness that a problem exists and through taking into account the various levels of context.
Wide knowledge of the field. Let's take the phrase to table a bill. The translator must know that if the original is in British English, it means "to submit a bill - i.e. a text proposed to become law -- to the country's legislative body", in French déposer un projet de loi (in Esperanto, submeti leĝprojekton), but that if the author followed American usage, he meant "to shelve", i.e. "to adjourn indefinitely the discussion of the text", in French ajourner sine die l'examen du projet de loi (in Esperanto arkivigi la leĝprojekton).
Here is another example. The word heure in French can mean "hour" as well as "o'clock". To be able to translate correctly the French phrase une messe de neuf heures, you have to know that a Catholic mass lasting nine hours is extremely improbable, so that the translation is "a nine o'clock mass", and not "a nine hour mass". Since the linguistic structure is exactly the same in un voyage de neuf heures, which means "a nine hour journey", only knowledge of the average duration of a mass can help the translator decide.
Awareness that a problem exists. When you become a professional translator, the chief development that occurs in you during your first three or four years consists in becoming aware of problems that you had no idea could exist. If you are transferred to another organization, the whole process will start anew for a few years because the new field implies new problems that are just as hidden as in your former job. Some of the public in this room may know that in the history of international communication there was an organization called International Auxiliary Language Association. Well, if you ask people how they understand that title, you will realize that, for a number of them, it means "international association dealing with an auxiliary language", whereas for others it means "an association studying the question of an international auxiliary language". The interesting point lies not so much in the ambiguity as in the fact that most people are not aware of it. When exposed to the phrase, they immediately understand it in a certain way and they are not at all conscious that the very same words are susceptible to another interpretation and that their immediate comprehension does not necessarily reflect what the author had in mind.
Similarly, most junior translators simply do not imagine that the words English teacher usually designate, not a teacher who happens to be a British citizen, but somebody who teaches English and can be Japanese or Brazilian as well from any English speaking country...
Taking into account the various levels of context. The English word repression has two conventional translations in French. In politics, the French equivalent is répression (in Esperanto subpremo), whereas in psychology, it is refoulement (repuŝo). You might believe at first glance that translating it correctly is simply a matter of knowing to what field your text belongs. If it deals with politics, you use one translation, if with psychology another. Reality is not that simple. Your author may use the psychological sense within a broad political context. For instance, in an article dealing with the Stalin era, you may have a sentence beginning with Repression by the population of its spontaneous critical reactions led to... In this case, although the text deals with politics, the sentence deals with psychology. The narrow context is at variance with the broad context.
I recently revised a text which had me wondering how a computer would deal with the various meanings of the word case. It was about packaging. In a section on wooden cases, it said: Other reasons for water removal important in specific cases are: (1) to avoid gaps between boards in sheathed cases; (2) to (...). A human translator's judgment leads him to a correct understanding of the first case as a synonym of "occurrence" and of the second as "a kind of big box", but how will a computer know? Suppose the text includes such phrases as A case can be made for plastic boxes or the importer complained about the poor quality of the cases. When the case was settled in court (...). Knowing the broad context does not help to choose the right translation if there is no mechanical means to determine that the author switched, in a narrow context, to a different meaning of the word.
Besides judgment, the other quality I mentioned as indispensable to make an acceptable translator is flexibility. This refers to the gymnastics aspect of translation work. Mastering the specialized field and the two relevant languages is not enough, you have to master the art of constantly jumping from one into the other and back. Languages are more than intellectual structures. They are universes. Each language has a certain atmosphere, a style of its own, that differentiates it from all others. If you compare such English expressions as software and, on a road sign, soft shoulder with their French equivalents, you realize that there is a very definite switch in the approach to communication. The French translations are respectively logiciel and accotements non stabilisés. The English phrases are concrete, metaphorical, made up, with a zest of humor, from words used in everyday speech, although this does not contribute to better comprehension: knowing the meaning of soft and of shoulder does not help you to understand what a soft shoulder is. In French, the same meanings are conveyed by abstract and descriptive terms, which do not belong to everyday usage. You don't understand them either, but for a different reason: because they are based on too intellectual, too sophisticated, too unusual morphemes, so that most foreigners have to look up the words in dictionaries.
The difficulty lies in the fact that this difference in approach has to be taken into account at the level, not only of words (a good dictionary may often solve that problem), but of sentences. Consider the sentence Private education is in no way under the jurisdiction of the government. It includes mostly English words of French origin, but common etymology does not imply a common way of expressing one's thoughts. In this case, a good French rendering would be L'enseignement libre ne relève en rien de l'Etat. You will realize the importance of those differences in the approach to communication if you take the French sentence as the original and translate it literally into English. The result would be Free teaching does not depend in any way from the State, which means something quite different, especially to an American.
In order to translate properly, you have to feel when and how to switch from one atmosphere to another. No human beginner, in translation work, knows how to do that, and I wonder how a machine will detect the need to do it, unless its memory is so huge that it includes all the practical problems that translators have had to solve for decades, with an appropriate solution. For instance, when new translators arrive in the World Health Organization and have to translate the phrase blood sugar concentration, practically all of them use an expression like concentration de sucre dans le sang. This is what it means, but this is not how the concept is expressed in French, in which you have to replace those three English words with a single one: glycémie.
Similarly, knowing that the French equivalent of software is logiciel does not help you to translate it by didacticiel when it refers to a teaching aid, which is the word you should normally use in that particular case. French uses narrower semantic fields, and this is something you have to bear in mind constantly.
The problem is that with languages, you never know how you know what you know. (Sorry, I am being self-centered. I never know, but perhaps, with your experience in the computerized analysis of languages, you know.) If, in a text dealing with economic matters, I meet the phrase the life expectancy of those capital goods, I know -- because I feel -- that I have to translate it by la longévité des équipements. I also know that when that same text mentions the consumers' life expectancy, I'll have to say, in French, espérance de vie, because the author for a while deals with a demographic concept which is included in his economic reasoning. But how do I know I know? I don't know. This ability to adjust to the various approaches to reality or fantasy embodied in the different languages, linked to an ability to pass constantly back and forth, is what I call flexibility. This is the quality which is the most difficult to find when you recruit translators.
We can now approach the same field from a different angle, asking ourselves the question: what are the problems built-in in languages that make judgment and flexibility so important in translation work? They relate to the grammar and the semantics of both the source and the target languages.
The more a language uses precise and clear-cut grammatical devices to express the relationships among words and, within a given word, its constitutive concepts, the easier the task for the translator. The worst source languages for translators are thus English and Chinese. A Chinese sentence like ta shi qunian shengde xiaohair can mean both "he (or she) is a child who was born last year" and "it was last year that she gave birth to a child".
In English similar ambiguities are constant. In International Labor Organization, the word international refers to organization, as shown in the official French wording: Organisation internationale du Travail. But in another UN specialized agency, the International Civil Aviation Organization, the word international is to be related with aviation, not with organization, as shown, again, by the French version: Organisation de l'aviation civile internationale (and not Organisation internationale de l'aviation civile). This is legally and politically important, because it means that the organization is competent only for flights that cross national boundaries. It is not an international organization that deals with all problems of non-military flying. However, since the linguistic structure is similar in both cases, no text analysis can help the translator; he has no linguistic means to decide which is which. He has to refer to the constitution of the relevant organization.
The problem is complicated by the fact that most English texts on which a translator works were not written by native English speakers, who might be more able to express themselves without ambiguity. Let us consider the following sentence: He could not agree with the amendments to the draft resolution proposed by the delegation of India. The draft translation read: Il ne pouvait accepter les amendements au projet de résolution proposé par la délégation indienne. I am not able to judge if the English is correct or not, but, as a reviser, I had to check the facts, so that I know that the translator, who had understood that the text submitted by India was the draft resolution, was mistaken. Actually, it was the amendments. In French, you would have proposé if it referred to the draft resolution and proposés if to the amendments. Similarly, in Esperanto you would have proponita or proponitaj according to what refers to what.
I wonder how a computer solves similar problems. I have been told that it detects the possible ambiguities and asks the author what he or she means. I wish it good luck. All translators know that authors are usually unavailable. Much translation work is done at night, because a report or a project produced during the afternoon session has to be on the desks of the participants to the conference in the various working languages on the following morning. They are not allowed to wake up authors to ask them what they meant.
Or the author is far away and difficult to get in touch with. When I was a reviser in WHO, I had to deal with a scientific report produced by an Australian physician. He mentioned a disease outbreak which had appeared in a Japanese prisoner of war camp. This was before e-mail time, so that we had to write to Australia to know if the disease affected American soldiers who were prisoners of the Japanese or Japanese caught by the Americans. When the reply arrived, it stated that the author had been dead for a few years.
Many mistakes made by professional translators result from this impossibility, in English, to assign an adjective to its noun through grammatical means. When a translator rendered Basic oral health survey methods by Méthodologie des enquêtes fondamentales sur l'état de santé bucco-dentaire, he was mistaken in relating the word basic to survey, whereas it actually relates to methods, but he should be forgiven, because only familiarity with the subject enables the reader to understand what refers to what. The correct translation was Méthodologie fondamentale applicable aux enquêtes sur l'état de santé bucco-dentaire.
My wife teaches translation to American students who come to Geneva for one year. A standard translation task she gives them includes the subtitle Short breathing exercises. Every year, half her class understands "exercises in short breathing", whereas the real meaning is "short exercises in deep breathing". The fact that native speakers of English so consistently make the same mistake, although the context provides all the necessary clues, keeps me wondering. Does a computer have a better judgment than humans? Can a machine discern, compare and evaluate clues?
The fact that, in English, the endings -s, -ed and -ing have several grammatical functions often complicates matters. In the sentence He was sorting out food rations and chewing gum, it is impossible to know if the concerned individual was chewing gum while sorting out food rations, or if he was sorting out two kinds of supplies: food, and chewing gum.
Problems caused by semantics are particularly difficult for human translators. They are of two kinds: (1) the problem is not apparent; (2) the problem is readily seen, but the solution either requires good judgment or does not exist.
An example of the first category is provided by the phrase malaria therapy. Since malaria is a well known disease, and therapy means "treatment", a translator not trained in medical matters will think that it means "treatment of malaria". But the semantic field of therapy is not identical with that of treatment, although this is not apparent if you simply consult a dictionary (Webster's defines therapy as "treatment of a disease"). It would be too long to explain here the differences, but the fact is that malaria therapy should be rendered, not by traitement du paludisme (kuracado de malario) , but by impaludation thérapeutique or paludothérapie (permalaria kuracado) , because it means that the malaria parasite is injected into the blood to elicit a febrile reaction designed to cure the attacked disease, which is not malaria. In other words, it means "treatment by malaria" and not "treatment of malaria".
In the French version, published by Albin Michel, of Hammond Innes' novel Levkas Man, one of the characters complains about les jungles concrètes in which an enormous population has to live. This does not make sense for the French reader. Since some of you understand Esperanto, I can explain the misunderstanding better using that language. Jungles concrètes means "konkretaj ĝangaloj". What the author meant by concrete jungles was "jungles de béton", "betonaj ĝangaloj", i.e. high-rise housing developments made of concrete. This is a case in which the translator was not aware of the existence of a semantic problem, namely that concrete has two completely unrelated meanings: a building material, and the opposite of "abstract".
An example of a semantic problem requiring good judgment -- and, with all my prejudices, I fail to imagine how a computer can exercise that kind of judgment -- is the word develop. It has such a wide semantic field that it is often a real nightmare for translators. It can mean "setting up", "creating", "designing", "establishing" and thus refer to something that did not exist before. It can mean "intensifying", "accelerating", "extending", "amplifying", and thus express the concept "making larger", which implies that the thing being developed has been concretely in existence for some time. But it can also mean "tapping the resources", "exploiting", in other words "making use of something that has been having a latent or potential existence". In all other languages, the translation will vary according to the meaning, i.e. to that particular segment the author had in view within the very wide semantic field covered by the word. To know how to translate to develop such or such an industry, you have to know if the said industry already exists or not in the area your text is covering. In most cases, the text itself gives no clue on that matter. Only the translator's general culture or his ability to do appropriate research can lead him to the right translation.
Such a simple word as more can pose problems, because its semantic area covers both the concepts of quantity and of qualitative degree. What does more accurate information mean? Does it mean "a larger amount of accurate information" or "information that has greater accuracy"?
A word like tape is just as tricky. If it refers to sound recording, you translate it into French as bande or cassette (provided you know which kind of recorder was used). But if it refers to the gluing material, as in Scotch tape, you have to render it by ruban adhésif, since in that particular case, the French word bande evokes the bandaging of a wound.
Often, a problem arises -- without being always apparent -- because a word has a special semantic value in the particular milieu in which the author works; in that case, an underlying concept is frequently unexpressed, since the author addresses persons working in the same field and used to the same kind of compact expressions. In the sentence WHO helped control programs in 20 countries, only knowing that in WHO parlance control program means "a program to fight a disease and put it under control" may make the translator suspect that the author meant "WHO granted its assistance to help fight the relevant disease in 20 countries". The junior translator who understood it as meaning "it helped to control the programs" was grammatically justified, since in English the verb to help can be construed without the particle to in the following verb and, in such a sentence, nothing enables you to know if control is used as a noun or as a verb.
However, most of the difficulties that human translators meet relate to the different ways in which various languages cut up reality into differentiated semantic blocks. I use the word block on purpose, because very often reality is continuous, as well as concepts, whereas language is discontinuous. Blue and green are what I call "semantic blocks", whereas in the spectrum there is perfect continuity. Very often, a concept that exists in a language has no translation in another, because peoples cut up the continuum in different sizes and from different angles.
In a number of cases, it does not matter. The fact that for the only French word crier English has to choose among shout, scream, screech, squall, shriek, squeal, yell, bawl, roar, call out, etc., does not pose serious problems in practice.
But how can you translate cute into another language? The concept simply does not exist in most. Conversely, the French word frileux has no equivalent in English, so that a simple French sentence like il est frileux cannot be properly translated. Still, you can say he feels the cold terribly or he is very sensitive to cold. Although those are poor renderings, they are acceptable. What most resists translation is the adverbial form: frileusement. How can you translate il ramena frileusement la couverture sur ses genoux? You have to say something like He put the blanket back onto his knees with the kind of shivering movement typical of people particularly sensitive to cold. To those of you who might think that this is literary translation, something outside your field of research, I have to emphasize that descriptions of attitudes and behavior are an integral part of medical and psychological case presentations, so that the above sentence should not be considered unusual in a translator's practice.
An enormous amount of words, many of them appearing constantly in ordinary texts, present us with similar difficulties. Such words as commodity, consolidation, core, crop, disposal, to duck, emphasis, estate, evidence, feature, flow, forward, format, insight, issue, joint, junior, kit, maintain, matching, predicament, procurement and hundreds of others are quite easy to understand, but no French word has the same semantic field, so that their translation is always a headache. Dictionaries don't help, because they give you a few translations that never coincide with the concept as actually used in a text; in most cases the translations they suggest do not fit with the given context.
Another case in point is provided by the many words that refer to the organization of life. You cannot translate Swiss Government by Gouvernement suisse, because the French word gouvernement has a much narrower meaning than the English one. (Interestingly, although the semantic extension of both words does not coincide exactly, you can translate it into Esperanto by svisa registaro, because the Esperanto concept is wide enough). In French, you have to say le Conseil fédéral or la Confédération suisse according to the precise meaning. The French word gouvernement designates what in English is often named cabinet. The English word government is one of the frustrating ones. You may render it by l'Etat, les pouvoirs publics, les autorités, le régime or similar words, evaluating in each case what is closest to the English meaning, and you have to bear in mind that at times it should be sciences politiques (for instance in the sentence she majored in government, in which the verb major is another headache, because American studies are organized in quite a different way from studies in French speaking countries).
The Russian word dispanserizacija illustrates a similar problem. It designates a whole conception of public health services that has no equivalent in Western countries. If you want your reader to understand your translation, you should, rather than translate it (it would be easy enough to say dispensarisation), explain what it means.
As you see, each one of the problems I mentioned makes the translators' task very arduous indeed. Problems caused by ambiguities, unexpressed but implied meanings, and semantic values without equivalent in the target language require a lot of thinking, a special knowledge of the field and a certain amount of research -- as for instance when you have to find out if an industry being developed already exists or not, or if secretary Tan Buting is a male or a female, which, in many languages, will govern the correct form of the adjectives and even the translation of secretary (Sekretär? Sekretärin?) . Such problems take up 80 to 90% of a professional translator's time. "A translator is essentially a detective," one of my Spanish colleagues in WHO used to say, and it is true. He has to make a lot of phone calls, to go from one library to another (not so much to find a technical term as to understand how a process unfolds or to find basic data that are understood, and thus unexpressed, among specialists) and to tap all his resources in deduction. I do hope that computers will free the poor slaves from those unrewarding tasks, but I confess that, with my incompetence in data processing, I am at a loss to imagine how they will proceed.