Computers: Why cross is your chicken?: The Internet is set to generate a huge demand for translations. Mike Hewill investigates whether machines will able be to do the job

Click to follow
The Independent Online
It is said that in the 1960s the Americans built a supercomputer to translate English into Russian and vice versa. As an experiment, a technician fed through the expression 'Out of sight, out of mind'. The machine whirred for a few seconds and disgorged the Russian equivalent. This was then returned to the computer to be translated back into English. And the result? 'Invisible idiot,' said the print-out.

While the story is probably apocryphal, it does show the sort of obstacles computer software has to overcome. Language translation is rarely a simple matter of A = B. Actual meaning can be context-related, for example, or masked by idiom. How do you meet both your needs and your grandmother within the same piece of text? Rencontrer or realiser? And what would a computer make of 'Let's not beat around the bush'?

Globalink, a US company, reckons it is on top of the problem. Recently it announced a machine-based solution which translates text files sent over the Internet global information network. Translations are made into French, Spanish or German and vice- versa. The fee is 5 cents a word, with a minimum charge of dollars 5 ( pounds 3.25) a translation. So how good is it?

Globalink says its system provides 'draft quality'. In other words, if you only need to convey the major points of a message, the software will oblige. However, users requiring top-quality translations should be prepared to edit the output themselves. 'Just as word-processing software doesn't make a person a writer,' says Globalink, 'translation software doesn't make you a translator.'

To maximise accuracy, Globalink offers a few pointers. First the text should be grammatically accurate. Do not use slang or literary terminology. Ideally, sentences should be short and, if possible, avoid phrasal verbs: ie use 'find' instead of 'come across'. Finally, if your text includes specialised terminology, invoke Globalink's additional subject-specific dictionaries. This is done by including the appropriate abbreviation in the subject line of your message header. For instance, (com) switches on the computer dictionary, and (leg) the legal dictionary.

The electronic mail (E- mail) address depends on the translation required. For English to Spanish, for example, upload to engspn@glnk. com. Translated documents are automatically returned to the sender. But if you want to cc them to someone else, you include the person's E-mail address in the header.

For my test, I composed a simple business letter and copied a short passage of straightforward technical documentation. And, just to show that I was not pussyfooting around, I also uploaded some jokes together with the full text of President Lincoln's Gettysburg Address. In all cases, the text went into the foreign language and then back into English.

And the verdict? A damn sight better than it had any right to be. The simple business letter looked as if it had been tackled by a fairly proficient GCSE student. In other words, it had minor errors of construction, but was still quite understandable.

The technical translation was adequate, in that terms such as Lan (local area network) and hard disk were correctly converted. But it slipped up on words like workgroup and bus. In the latter case, no distinction was made between the computer and double-decker variety. The jokes, on the other hand, went rather well: Mon chien n'a obtenu aucun nez. Comment est-ce qu'il sent? Terrible. ('My dog has obtained no nose. As smells he? Terrible.') And Por que cruzo el pollo el camino? A conseguir al otro lado. ('Why it crossed the chicken the road? To obtain the other side.')

But most impressive was the Gettysburg Address. The famous first line became in Spanish: Fourscore y hace siete antildeos nuestros padres trajeron adelante sobre este continente una nacion nueva, concibieron en la libertad y dedicaron a la propuesta que todo de hombres se crean iguales. ('Fourscore and seven years ago our parents brought forward on this continent a new nation, conceived in the freedom and devoted to the proposal that all of men are believed equal.') Perhaps not as Lincoln would have said it, but intelligible nonetheless.

Globalink is showing quite respectable results already. In five years, such systems could really give human translators a run for their money. They may have to. According to research, the industry average for manual translation is only seven A4 pages a day a person. As the Internet expands and millions more non-English speaking business users log on, machine translation may turn out to be the only way to cope with the glut.

IndyBest product reviews are unbiased, independent advice you can trust. On some occasions, we earn revenue if you click the links and buy the products, but we never allow this to bias our coverage. The reviews are compiled through a mix of expert opinion and real-world testing