Experimental feature

Listen to this article

00:00
00:00
Experimental feature
or

Nicole Kidman may have added a hint of glamour to the job when playing the lead role in the Hollywood film The Interpreter this year but some believe that improvements in technology tasks could soon render the human touch surplus to requirements.

Instead, according to the vision, a computer will be whirring away in the background translating matters of state into perfect sentences spoken in another language.

On Google, the internet search engine, you can often translate foreign language text that comes up in your search into your own language.

You won’t get a word-for-word translation but you will probably get the gist of what the article is about – enough to know whether it is relevant to your search and possibly even enough to answer your question.

“Machine translation [using software] has an accuracy rate of about 60 per cent,” says Alison Crawford, a research analyst. “It is great for getting the gist of an e-mail, so that you know ‘Bob from Germany says the shipment will be four days late’.

“But attention has turned more towards translation memory, a database of phrases that have already been translated correctly. This has an accuracy rate of up to 80 per cent,” she says.

“For a smaller corporation, a vertical translation memory – with phrases from their industry – makes sense. But some large corporations will have their own dedicated memory.”

The systems reduce the number of words that have to be translated by a human. “If there’s an exact match for ‘king-size bed’, you don’t have to pay again to get it translated,” she says.

A translation memory gets increasingly reliable the more it is used and new phrases are added. With customers wanting a full range of translation tools and services, there has been a lot of consolidation among software-assisted translation suppliers, such the acquisition of Trados of the US by SDL, a UK-based software house.

The services they provide are increasingly sophisticated. SDL, for example, offers its largest clients, what it calls knowledge-based translation (KBT). This has three technology-based elements.

Automated translation software translates with a little help from the second element, translation memory, while the final piece of the jigsaw is phrase-finding software that highlights the problem areas of the text that are likely to need input from a human translator.

But, most importantly, a translator, or translation team, reads through the machine’s work and translates the most difficult parts.

Getting the combination right can have a significant effect on costs. SDL says that in work it has carried out for hotels group Best Western International, it saved its customer $2m in localisation costs in the first year and cut the time taken to translate information from 24 months to eight months.

“The only other way to get a large report translated very quickly is to put many people on to the same job,” says Mr Lancaster. “But then you will get inconsistencies in the way words are translated. In part of the text, ‘a carrier’ will be referred to, and this will become – in another translator’s hands – ‘a bag’.”

He says: “Translation costs come down to between 15 and 17 euro cents a word when KBT is used, from about 20 to 25 euro cents a word from a translation agency.”

The march of technology is of course having an impact on the working life of interpreters. And while it might seem that having a machine perform some of the more mundane tasks, leaving more time for translators to address some of the more challenging work, would be welcome, that is not necessarily the case.

“A decorator who paints a whole room may get more job satisfaction – and more money – than someone who just finishes off the tricky corners,” says Mr Lancaster.

But translation software might not be entirely bad news for translators. Mr Lancaster believes that as software increases the speed at which information can be translated and brings the cost of translation down, the size of the market will grow.

In February, IDC, the market researcher, predicted the market for translation and globalisation software would grow to $263m by 2009. But, if US companies translated and localised their websites so they could sell more products overseas, they could generate $7,000bn in global e-commerce revenue by 2007.

Not all information is suitable for the automated translation approach, however. Generally speaking fact-based reports in which common phrases are repeated are easier to translate than fiction or anything that is politically sensitive.

So Ms Kidman may not need to worry about being replaced by a computer just yet.

”When you combine a machine and a human you get some good productivity going,” says Mr Lancaster.

“However, I don’t think you will replace translators with machines in my lifetime.”

Copyright The Financial Times Limited 2017. All rights reserved.
myFT

Follow the topics mentioned in this article

Comments have not been enabled for this article.