ln the 1950s, the surging Cold War demand for English versions of Russian technical documents figured into an exiting belief, held by many computer scientists, that machines could be designed to produce accurate automatic translations. Today, several software programs claim to have accomplished this feat. But just how realistic is their claim?
Three decades ago, under contract from the U.S. government, a company called Systran pioneered the field of machine translation. Prototype models were “direct systems,” producing literal translations by looking up each word or phrase in a lexicon and substituting an equivalent in the target language. This technology owed much to a simplistic analogy advanced by American mathematician Norbert Wiener. Computers were used during the war to help break enemy codes; decoding is a matter of transforming a set of symbols; language translation could be the same. But when scientists actually fired up their machines, the results were less than encouraging.
One huge snag for machine translation, or MT, is word order. The latest products seek to overcome this by incorporating grammatical rules for figuring out what words are performing what functions in a sentence. The programs construct a “parse tree,”a diagram showing the grammatical function of each word in a sentence. The result is transferred to the target language with the aid of a second set of rules governing grammatical combinations in that language.
The latest results? That depends on whom you ask. When faced with criticism of their products' translations, MT vendors tend to invoke the “talking dog” ― as in, “Don't be picky; it's amazing that a dog can talk at all.” And in all fairness, outright mistranslations are far fewer than one might expect.
The real imitations of MT have less to do with gravimetrical complexity than with the fact that computers don't have any common sense. Language is full of ambiguity that correct syntax goes only a short way toward sorting out. Is a “bank” a place to put money, or is it the edge of a river? A five-year-old child can grasp the difference, but getting a computer to do so is another matter.
In other words, semantics is the key to MT, and semantics is a matter of a let mere than linguistics - it requires real-world knowledge. Indeed, one can produce near-perfect translations with just about any system by limiting its vocabulary to a narrow, specialized area, Canadian radio stations, for example, use software to translate weather bulletins from English to French. With a lexicon of just a few hundred words, the program achieves an accuracy rate of more than 90 percent.
A major effort now under way in MT circles is to develop elaborate classifications of meaning that will duplicate the knowledge that allows human beings to know which of various meanings a speaker intends, This, of course, is virtually equivalent to the challenge of artificial intelligence ―creating a computer that thinks, or at least comes so close to human thinking that we can't tell the difference.