Translation: Letting Computers Do It

Scientists have long predicted that computers would one day help speed up the arduous task of translating texts — and now that time has come. Systems designed in the United States, Japan, Eu­rope and South America follow one of two basic approaches. The "direct" method uses rules of syntax and grammar to translate one languageinto another, while the "interlingua" approach employs an internal intermediate language as a bridge between the "source" and "target" languages.

The direct approach is the most common, though not always the simplest, method of computer translation. It requires a sepa­rate program — including a dictionary of thousands of words and phrases, as well as rules of syntax, grammar and conjugation — for each pair of languages involved in a given translation.

Last summer, a small Tokyo firm, Bravice International, Inc. began marketing the country's first commercial Japanese-to-Eng- lish translation system. Unlike word-based systems, this one breaks a Japanese sentence into phrases that are translated and then reconfigured into an English sentence. The company says the system can translate Japanese into passable English at a rate of up to 3,000 words an hour on simple texts.

Bravice's system is based on the translating methods pio­neered by Weidner Communications Corp. of Northbrook, Illinois. Partly owned by Bravice since 1982, Weidner offers direct-method systems that translate English into French, Spanish, German, Por­tuguese and Arabic. There are also systems that translate French, Spanish and German into English. Only six months ago, Weidner introduced translation software that works on an IBM Personal Computer and does rough translations at between 1,200 and 2,000 words an hour. (Previously, the system was compatible only with more powerful minicomputers made by Digital Equipment Corp., which operate at 5,000 to 8,000 words per hour.) Weidner's new software allows an operator to define unfamiliar phrases before translating the full text.

A program developed by Automated Language Processing Sys­tems (ALPS) of Provo, Utah, calls for the continuous assistance of a human translator. The system presents an operator with sen- tence-by-sentence adaptations that can be changed immediately. The computer "learns" as it proceeds through a text; once a term has been translated to the user's satisfaction, it will be changed the same way when it appears again.

The interlingua method is most effective for texts that re­quire translations into several languages. One such program is Systran, marketed by World Translation Center, Inc., of La Jolla, California, and used by the Commission of the European Commu­nities. Researchers at the Group Study for Automatic Translation in Grenoble, France, continue to refine their 10-year-old system that makes use of a "neutral language" called Ariane, which stan­dardizes the syntactic organization of a language before the actual translation.

A unique approach has been developed by Bolivian computer scientist Ivan Guzman de Rojas. He found that Aymara, a lan­guage spoken by 3 million Indians in the Bolivian highlands, con­tains a simple yet rigid structure that could be easily reduced to al­gebraic equations suited to a computer. These equations form the core of a mathematical "language" that can be applied to virtually any Western language. Already, the system can translate English, Spanish, French or German into the other three languages at a rate of 600 words a minute. Guzman de Roja says the dictionary capa­city still must be expanded to arrive at a passable accuracy rate of 90 percent.

Logos Corp. of Waltham, Massachusetts, uses a modified interlingua method in its German-English program. It incorpo­rates a universal language that analyzes structure and content, while using the direct method to complete the translation.

Still, difficulties remain. Editors still must refine the transla­tions by interpreting some idioms and deciphering unfamiliar words. But the programs are improving, and the companies in­volved are optimistic. "In 10 years' time," says the head of one Japanese firm, "more than 80 percent of all translation work will be done by machines."

Текст 7

Intelligence and IQ

What is intelligence? For many years people have wondered about it. How do people define it? What affects intelligence more: enviromment or heredity? How do people measure intelligence? Does intelligence mean success in life?

Teachers and psychologists have known for a long time that some people learn some things faster than others. Some people learn certain subjects better than they learn others. One person may be good at maths but poor at languages. Another may be good at languages but poor at music. Some people may learn new things fast. Other people may learn them slowly. Questions about learn­ing are part of the study of intelligence. Psychologists have made tests called IQ tests to measure intelligence, and to understand the IQ scores.

The definition of intelligence is not universal. Nobody agrees on a definition of intelligence. One definition says intelligence is the ability to deal with new situations. Other definitions include the ability to solve problems. Some say intelligence comes from en­vironment. Environment is the situation people live in. Some defi­nitions say people are born with intelligence. Some say intelligence is a combination of the two: environment and natural ability. Peo­ple use the words intelligence and intelligent in many ways. The ex&ct definition is not clear.

Intelligence is difficult to define, but psychologists try to measure it. Information from the IQ test may help some people. Al­fred Binet, a Frenchman, began testing intelligence in 1905. Binet and Simon, another psychologist, wanted to find students who would probably succeed and those who would probably not succed. Their test was revised later by psychologists at Stanford Univer­sity. The test commonly used today is the Stanford-Binet test. This test is one of the most famous tests of a person's intelligence quo­tient (IQ). Another popular test today is the Wechsler Intelligence Scale for Children (WISC), and the Wechsler Adult Intelligence Scale (WAIS). There are other intelligence tests. Most of these tests measure memory, definitions, reasoning, drawing analogies, calculation, classification, and several other areas. The scores are calculated in several ways. On the major tests, the average score is 100. If a person scores between 90 and 110, this is usually consi­dered average. Above 130 is usually called gifted. Below 85 usually indicates some learning disability.

What do the scores actually mean? This question is very diffi­cult to answer. Many times the tests show learning problems. Then the students can get help. However, there are several problems with understanding and using the test scores.

First, intelligence changes. A child of five may have a diffe­rent score than at age ten.

Another problem with the test is the language of the test. If the student's native language is not the same language as the test, the student's score may be low.

An additional problem is culture. American tests are best for middle- and upper-class people. Therefore, other people may have lower scores.

Another problem is "labeling". If a child scores high, people be­lieve that the child will do well in school. If the child does not do well, some people may wrongly believe that the child is lazy or bored. There may be other reasons why the child is not doing well. The op­posite problem is also true. Sometimes a child scores low. From that time, some people may expect the child not to do well. Therefore, the child begins to believe that he or she will not do well.

In conclusion, it is difficult to define intelligence, but psy­chologists try to measure it with IQ tests. These tests are often very helpful because they identify students who may have prob­lems or who may be very intelligent. However, the scores may not be easy to understand. People must be very careful in using the scores. Careful use of the scores can provide a lot of information. Misuse of the scores may cause many problems.

Приложения

Наши рекомендации