Translation Trouble

Experts on mechanical brains are fascinated by the problem of designing a translating machine. A few attempts have been made, but the only "language" translated so far is a few words of artificial baby talk.

In the American Scientist, Dr. Yehoshua Bar-Hillel, lately of M.I.T., now of Hebrew University, Jerusalem, explores the difficulties of true mechanical translation and finds them enormous. He gives up all hope that any machine will be able to make an acceptable translation of a real work of literature. The job would call for cultural and artistic abilities that cannot be built into a machine.

Dr. Bar-Hillel aims lower: at the translation of technical prose, which is not normally written for literary effect and does not require a literary translation. But even this problem is a tough one. A large German dictionary, says Bar-Hillel, contains about 400,000 entries, and since each entry has an average of four or five derivatives, the total number of German words is about 2,000,000. Each of these words, in turn, can be translated into English in about ten ways. So a German-English mechanized dictionary would contain about 30 million words.

Scientific papers, of course, do not need the full vocabulary, but all existing languages are full of grammatical oddities that would be hard on a machine's digestion. In German, for instance, a prefix is often widely separated from the verb whose meaning it changes. Dr. Bar-Hillel points out that the sentence "Paid gibt Trunkenheit vor" (Paul simulates drunkenness) might be translated mechanically "Paul gives drunkenness before." He has no solution for this problem except to make writers of German use an "operational syntax" that will not perplex the machine.

No syntax will help the machine with idiomatic expressions. Literally translated, "Das gibt sich schon" (This will subside in time) would come out as "This gives itself already."

Even the biggest and cleverest machine, Bar-Hillel concedes, will not turn out finished translations. It is bound to fall into errors and ambiguities that will have to be straightened out by a human "post-editor." He admits rather sadly: "It is only when the situations are naturally of a.low degree of complexity, or are artificially arranged to be so, that the rigid mechanical brain can exhibit superiority over the flexible human brain."

This file is automatically generated by a robot program, so reader's discretion is required.