So far, so good. Chess problems and the tasks of business management can be expressed in numbers and mathematically understood. The knight in a chess game, productive capacity and market supply, divisions of troops, rye and sugar beet can easily be represented by index numbers with which the machine can proceed to calculate in the ordinary way. But what will happen if everyday words are to be used, like those in the Christmas poem?
The computer can deal with them very easily too. It has two ways at its disposal of processing words, concepts and meanings. It can take in and store words of the customary length in just the same way as it accepts and stores numbers with many digits. There are codes - the teleprinters for example - which permit not only numbers but also letters of the alphabet to be translated into bits. The other possibility is to provide each word with an index number and to work with these numbers. Both methods are used.
You will find an example of this procedure in the translation program we mentioned in the first chapter of this presentation. You give the computer a series of Latin words to store. Each word is given an address, and the equivalent English terms are placed at corresponding other addresses. That is the vocabulary. Then there is the grammar. It is composed of circuit regulations for the program. If the world silva ("wood") does not appear in its basic (nominative) form in the text to be translated, but as silvam, the computer finds by means of a circuit line the reference in its stores confirming that this change is permissible, that it connotes the accusative case, that the word "wood" is the object of the sentence, and that if it is preceded by "in" the latter must be rendered "into."
In accordance with this rule, the computer looks for the English meaning of each Latin word, converts it according to the requirements of grammar, and arranges the words in the proper order. The sentence "Cerva in silvam ambulat" would run, thus translated: "The deer into the wood she walks," and the translation of "Avia cum puellis silvam intrat et puellis herbas monstrat" would be: "The grandmother with the girls the wood she enters and to the girls the grasses she shows." That may sound rather like the lyric of a song, but it isn't English. A further working process must see to it that the sequence of words is so arranged as to meet the requirements of English syntax, and that here and there a word is added or taken away.
Every language has words with several meanings. In English, for example, "bank" can either mean the edge of a river, or a place where money is deposited. Or take the word "fit" - does it mean trained and ready for action, or a convulsive attack, or the size of a pair of shoes? And what does "graze" mean? To feed on grass, or to scratch? How is "set" to be translated, or "trip," or "can"?
Those are problems for a poor translating machine. The human reader would judge from the context of the sentence or of the whole article, just what kinds of banks, fits or grazing are concerned. The automatic translator must do its work in just the same way. Its programmer, teaching it to translate, must think a long way ahead. Every word with two or more alternative meanings must be accompanied when it is put into the store with a stock of related ideas. "Bank" for instance could be referred to such words as "mound," "ridge," "river," "lake," "canal," on the one hand, and on the other with such terms as "money," "counter," "deposit," "account," "robbery." In translating, the computer must then - once more - look around: do any of the associated ideas crop up in the vicinity of the disputed word? If it finds any of them, it can make a fairly reliable guess at the meaning of the doubtful word. If it does not find any of them, then ...
Then: well, the experts are by no means satisfied with the capacity of the translating machines. It is reported from Quebec that an automatic translator translated the sentence "The spirit is willing, but the flesh is weak" from the original Greek into English. The result was: "The whisky is recommended, but the meat is no good." The story is a good one, but every computer initiate knows that it cannot be true. The errors committed are not at all typical of the mistakes made by a translating machine. Still, the anecdote does show, and quite correctly, that astonishing misunderstandings are liable to happen in electronic interpreting - particularly when the text leaves tangible, concrete subjects and moves into higher, abstract spheres. An attempt to translate the works of Oscar Wilde automatically into Russian would in all probability have positively grotesque consequences. Even the prospect of an electronic translation of Agatha Christie into French makes us, not to mention the French, fear the worst.
But the scientists would nevertheless like to find out how far an electronic computer is capable of fabricating grammatically irreproachable or even stylistically beautiful texts.