The demand for language services is growing, but machine translation has yet to match the quality of that produced by a human. However, a new estimate suggests that machine translations will have caught up by 2029. Author, inventor and futurist Ray Kurzweil tells Business Day that in less that two decades artificial intelligence will be able to produce translations that rival a human’s for accuracy. This could be good news, as a recent Common Sense Advisory report has revealed demand for translations and other language services is increasing by between 15 and 20 per cent each year. Translation in the pipeline Writing for the publication, Ian Henderson – chief technology officer of Rubric – notes that in the last year there have been announcements about three new simultaneous translators going into development. Indeed, Language Insight reported last year on Microsoft showcasing a new piece of software that is able to interpret a person as they speak and relay what they have said in the target language – and in their own voice. So, developments are clearly being made, but Mr Henderson points out that machine translation is something engineers have been trying to perfect for years. Previous attempts have seen the linguistic rules of different languages being used as the base upon which to build the tool, while others – like Google – have relied on statistics. The latter approach involves the system tallying the most commonly inputted words and turns of phrase and using this to perfect its output, which means the languages most frequently requested for translation garner better results than those that are rarely used. However, what has held up the development of machine translations, no matter what the blueprint, is that it is not possible for them to think like a human. Part of translation involves using subject knowledge, context, association and common sense, which machines cannot yet replicate. For instance, it was discovered that the likes of Google Translate were sometimes returning translations that could be perceived of as sexist. It was recently revealed translated content frequently suggested certain professionals were men, when in fact they were women. This occurred because the results are based on averages, so if a certain profession was more commonly held by a man, the machine would assume it was in every example. A human would not make such a mistake, as they would recognise the context of the original piece and translate it correctly. The human touch Mr Henderson notes that those who work in the language industry understand that when the human mind is removed from the translation process, the whole thing becomes difficult. He points out that in “many” instances a human is needed to intervene and correct the errors in the machine-generated translation. Due to this, businesses often use a machine to create the initial translation and then have a person look over it and fix any mistakes. Because of these difficulties, Mr Henderson regards the future of machine language services as being computer-aided translation, which takes advantage of both a human’s skill and intelligence and a machine’s speed. Ultimately though, the type of translation a business invests in will depend on the specifics of that individual case. “Circumstances will dictate which form of translation to use. Companies that let their translation efforts be guided by their situation will make the best use of their resources and manage risk better,” Mr Henderson explains. KPCB’s recent Internet Trends Report revealed that there are currently 2.4 billion internet users worldwide. The largest and fastest-growing demographic is Chinese, and people in China now spend longer on the internet than people in the USA. However, English remains the most prevalent language on the internet, with more than half of sites featuring content in this language according to W3 Techs. By translating web content, businesses are giving themselves a good opportunity to beat their competitors. This is particularly true of companies that use the internet for sales, as consumers have been proven to prefer to make purchases on sites written in their mother tongue. If a company knows interest in its products is being shown in another country and translates its website to reflect this, it has a good chance of securing these customers. This is even the case if its competitors are also selling in that country at a slightly lower price – buyers are still likely to choose the e-commerce site written in their native language. Businesses thinking about translating their website content may consider using machine translation as it is cost effective and fast. However, the result could be something that is not only inaccurate but also incomprehensible to the reader. If this is the case, the business will have no more chance of selling in this new market than it did before the translation was done. A human translator working into their mother tongue will produce the best results. At the very least, businesses should make use of the services of a qualified human linguist to edit the content generated by the machine translator they used. There should always be a human involved in the project at some point to ensure the best possible results. While Mr Kurzweil’s prediction that machines will rival humans for their translation skills may yet come true, it is a long way off. For now, human translators have little to fear from the machines.