Who dares forecast the fates of languages? How much credence do such visions deserve? Can we make linguistic prophesy more useful? The book Megatrends 2000 by J. Naisbitt and P. Aburdene (Morrow, 1990) and a recent forecast of "simultaneous translation of human speech ... in just 20 years" raise these concerns.
"Global lifestyles and cultural nationalism" evolve paradoxically together, says the new Megatrends : "The more homogeneous our lifestyles become, the more steadfastly we shall cling to deeper values - religion, language, art, and literature. As our outer worlds grow more similar, we will increasingly treasure the traditions ... from within" (ch. 4).
Megatrends 2000 sees English as the main force for a global lifestyle, though it says English is supplementing, not replacing, other languages. English is the language of computers, science, international business, diplomacy, and youth culture.
Still, foreign language study is up in U.S. schools, with a trend toward Japanese and Chinese, notes Megatrends . And there is a resurgence of Catalan in Spain, Welsh in Wales, French in Canada. Governments have restricted English in the Philippines, Singapore, and Sudan. The Soviet Union embraced some 100 languages, many of which are being heard again more widely.
This countertrend of difference against the juggernaut of sameness has nonlinguistic symptoms, too. The Vienna accords on human rights (1989) following on those of Helsinki (1975) are among them. Megatrends 2000 predicts further expression of ethnic pride and prowess in 1992, as the single European market comes to life.
Which side of this paradox does an invented world language belong to? Contrary to appearances, I think it falls on the differentiating, not homogenizing, side. Most Esperanto speakers, for example, want not only global communication but also linguistic diversity. So, they have been enthusiastic about the use of their language to help computers translate.
What, then, about machine translation? The Institute of Electrical and Electronic Engineers in March 1991 gave an optimistic spin to a Gallup survey of its members. But only 58% of the 150 respondents expected automatic simultaneous interpretation by 2010; 23% felt there was less than an even chance. The optimists were 66% of the government, 60% of the corporate, and 48% of the academic engineers; the doubters rose from 16% in the first two groups to 36% in academia. Self-interest (professional aggrandizement) may blinker this survey, which didn't address the engineers' linguistic sophistication. Vendors of simplistic translation devices, such as the 30,000-word Spanish Assistant , likewise overlook the many mistranslations their products generate when finding examples for advertising copy.
Machine translation is making progress again with advances in linguistics and artificial intelligence, after initially being hyped and then hamstrung. Though linguist David Crystal, writing in the 1987 Cambridge Encyclopedia of Language , doubts machines will ever replace human translators, he finds exciting progress in computer-assisted translation and doesn't rule out the possibility that useful automatic interpretations of telephone conversation will be realized by 2000. Commercial systems' limits remind us we have far to go, while the high quality (in limited domains) of the best academic projects suggests we can go far.
But automated translation will change human behavior, according to a March 1991 article by Peter Benton in Byte . "Adopting machine translation requires a great deal of learning and dedication. Work flow, job descriptions, and habits must change. The fact that raw translation is now performed by a machine, rather than people, transforms the fundamental nature of the process." Term banks, pre-editing and post-editing, voice recognition, parsing and disambiguation have found their way into the lexicon of inter-language processing. Will facilitated translation in science, technology, business, and tourism make language learning obsolete? Don't assume so - did computers make the study of mathematics superfluous? The forecasts of Megatrends 2000 provide a needed balance. So does the saying: Les traductions, lorsqu'elles sont belles, elles ne sont pas fidèles, et lorsqu'elles sont fidèles, elles ne sont pas belles.
Distributed Language Translation (DLT) is a multilingual machine translation (MT) project. It was developed and implemented between 1984 and 1990 in BSO (Buro voor Systeemontwikkeling), a Dutch software firm. Funding of $10 million came from BSO and the Dutch government. A prototype program, designed to translate texts on airplane maintenance in a simplified English, was completed in 1988. DLT was discontinued when the allotted time (though not the money) ran out without an industrial partner willing to invest in further development.
DLT uses a fully developed intermediate language (IL), a slightly modified version of the planned language Esperanto, as a link between source and target languages. This makes DLT unusual. One advantage of an IL is to reduce the number of translation steps in dealing with more than three languages. For example, if there are four languages, we can model systems with and without an IL as shown in the two figures. Without an IL we need 4'3 = 12 translation steps; with an IL we need only 2'4 = 8. The IL functions like the center point in a wheel linking the spokes. As the number of languages increases, so does the difference between the numbers of translation steps in the two models. For example, an IL reduces the steps from 90 to 20 with 10 languages.
The most difficult problem in translating by computer is the indeterminacy of meaning in natural languages. A given word or word group often means different things in different contexts. And context functions in different ways in different languages to make the meaning more specific. DLT approaches this problem by considering the context associated with any word being translated and by using a knowledge bank of relationships between words.
The sentence cut the cake exemplifies this problem. In DLT, the first step in translating that sentence into a target language is to translate it into Esperanto. But there are at least two ways to translate to cut : tondi and tranchi . The only clue in this sentence is the verb's direct object the cake . This suffices, here, to tell us tranchi is the choice (actually, the imperative form tranchu ). Tondu would be right if the object phrase were hair, for example. Tranch- has the sense of slice and tond- the sense of snip or clip . Initially, the system offers both candidate translations tondu la kukon and tranchu la kukon .
To choose the better translation, we consult the knowledge bank. It reveals words that combine with each of the two candidate translations of cut . The word kuko (cake) may be in the knowledge bank in the direct-object relation with tranchi . If so, tranchu is selected as the better translation. If there is no precise match in the knowledge bank with either candidate translation, an algorithm known as SWESIL (Semantic Word Expert in the IL) calculates the semantic proximity of the words in the knowledge bank to the words such as kuko which have a specific syntactic relationship to the word being investigated. Even if tranchi kukon is not in the knowledge bank, perhaps such pairs as tranchi panon (cut bread) or tranchi gorghon (cut a throat) are found there. Linked with tondi , on the other hand, we ought to find words like papero (paper) and herbo (grass). What basis does SWESIL have for deriving the conclusion, intuitively felt by the human translator, that the words in the knowledge bank with tranchi are more like kuko than the words with tondi ? In DLT, several different methods of automating such intuitions about meaning were tried. The version of SWESIL ultimately used in the prototype calculated the semantic proximity of any two words on the basis of the number of words in the knowledge bank related to both of them. This often leads to the correct result. On the other hand, SWESIL ignores contexts larger than two-word combinations. The so-called Bilingual Knowledge Bank, developed after the prototype, has as its basis of knowledge entire texts rather than word pairs and thus offers the hope of dealing with larger contexts. It remains to be seen if this avenue of development will be pursued in future attempts to automate solutions to the language barrier.
[Dan Maxwell is a theoretical linguist, specializing in syntax. He works on computational applications of linguistics in the software company Buro voor Systeemontwikkeling, Baarn, Netherlands. He is a co-editor of Metataxis in Practice: Dependency Syntax for Multilingual Machine Translation (Foris, 1989). For more about DLT, see Victor Sadler, "Machine Translation Project Reaches Watershed", Language Problems and Language Planning , 15 (1991), 78-81.]
Students and teachers of second languages are often frustrated by linguistic interference, that is a speaker's knowledge of one language interfering with the correct production of another. Yet, despite the widespread awareness, little practical guidance is available to ameliorate the problem.
Invented languages aren't immune to interference. In cooperation with Professor James Cool and his students in two intermediate levels at the 1991 Intensive Esperanto Program at San Francisco State University, I did a preliminary study of interference with Esperanto. After a week's immersion, the students' written exercises were free from errors attributable to other environments, primary activities, or "being rusty" in Esperanto. Therefore, I was surprised that interference errors comprised the second largest group, after mistakes with the accusative case. Since students at this level think predominantly in their native language, mental mistranslation is a factor. But how can we characterize interference types, and what is the best pedagogical approach for minimizing these errors? In an attempt to better understand the phenomenon, I divided the errors into seven categories based on the linguistic factors and on distinctions made by students:
1.One language has greater precision than the other - a word in one language does the work of two or more in the other language.
2.Different concept mapping - several words in each language describe similar concepts, but the divisions between concepts differ, so no word adequately translates a word in the other language.
3.Differences in tense sequence and verb usage.
4.Sound or spelling similarity.
5.Omitted words - obligatory words in one language that are optional or absent in another.
6.Direct translation of idiomatic expressions.
7.Interference from previous second languages.
I suspect interference is even more common among Esperanto students than among students of other languages. The rapidity with which students can begin to express themselves and the regularity of the language encourage students to attempt to express ideas well beyond what is possible with the language elements that they have "mastered". If their communication efforts are fairly successful, they are likely to continue to push their limits. It is likely that this accelerates the learning process and student satisfaction. Teachers must recognize that this will result in more errors, and apply proper correction in a way that will diminish error production, without dampening the student's enthusiasm.
In combating interference, teachers should distinguish errors from mistakes . We can define "errors" as faults due to a lack of knowledge and "mistakes" as momentary deviations from understood principles. Students often knew this difference and resented teachers who explained in detail things they already knew. If these students are representative, this presents a major challenge to teachers, who must be attentive to interference as a common cause of errors and apply proper correction for both mistakes and errors, without frustrating the student.
[Derek Roff is the learning technologist for the Modern and Classical Languages Department at the University of New Mexico, where he develops multi-media learning materials. Write us for a report of this ESF-funded study.]
The world, of course, is multilingual. The number of languages in use is estimated as two thousand to seven thousand, depending on which speech varieties count as "languages". With the number of countries about two hundred, linguistically diverse countries are the rule, not the exception.
But for most people it's frightening to think about getting prepared for life in a multilingual world. How many languages do I need to learn? Which ones will be the best bets? Can I really make myself understood? Will they speak slowly enough for me? Will they laugh at my accent?
Do we protect ourselves linguistically from this linguistic anxiety? Consider how we talk about languages. Radio journalists in the United States introduce translated voices with "Speaking through an interpreter, XYZ said that ...". Even if the speaker is a Venezuelan speaking Spanish in Caracas, or a Russian addressing the United Nations General Assembly in Russian, one hears the same phrase. A U.S. official speaking English in Paris is never described as "speaking through an interpreter". If journalists said "Heard through an interpreter ...", the suggestion would be that we hearers need help to hear, because we don't understand the speaker's language. The current phrase suggests the opposite: the speaker needs help to speak.
You can find other examples if you look. People are said to "not speak the language", as if they were alingual and it were their duty to learn "the language" rather than the duty of those serving them to learn theirs. Languages are called "dialects" and the idea that it might be appropriate for us to learn them is thereby disparaged. "English" is used as a synonym for "language", helping us forget that the ordinary language is, for most people, something else.
The suggestion is that we are central, others are peripheral, and it's their job to adapt to our language. This is the linguistic analogue to our habit of describing some Philippine village as "remote" or the customs there as "exotic", when for those living there New York may be remote and 5:00 p.m. traffic jams may be exotic.
Illusions are comforting, but they can inhibit the recognition and hence the solution of problems. Unexamined phrases, in turn, can subtly promote illusions. As a small but significant step toward solving the problems associated with world multilingualism, we can teach ourselves to talk more impartially - less self-centeredly - about languages.
Esperanto Studies and Interlinguistics.
Esperantic Studies Foundation.