In Star Trek movies and TV shows, translations between Klingons and humans onboard the Starship enterprise happens in real time, with no glitches. Many techies likely thought that by 2013 or 2014 we would have accurate machine-based translators in our pockets.
Translation has never been more important. As our world becomes more mobile and more international, being able to translate news stories and web pages in different languages is a big plus. For companies, accurate translations are a must for deals across borders.
However, even makers of popular machine-based translators such as Google and Microsoft’s Bing agree that replace human translators with machines is a long way in the future. In the past few months, Google’s efforts at translation through machine have gotten a lot of press, and there is no doubt that Google’s current translation efforts are a big jump from the online translators of years past. While previous translators online did only word-by-word translations, today’s machine translators use Statistical Machine Translation (SMT) to analyze databases of human translations online. Algorithms then try to find patterns in groups of words – rather than individual words – to translate sentences and paragraphs.
However, there are lots of problems. For one thing, machine translators rely heavily on online translations already completed – and any errors in the system can affect accuracy of the machine translations. A bigger problem is simply that machine translations and translation programs still can’t understand or interpret nuances, syntax, and various translations of one word.
This means that even Google experts agree that machine translations can be a bad idea in situations where accurate translations are a must – such as medical settings. While machine translations may be handy for getting a very general understanding of a web page, they are not great for really understanding or translating longer documents. It can be even worse in cases where a document relies on an industry heavy language. Translating a document written in legalese, for example, absolutely needs a professional and experienced human translator.
It could be one reason why machine translation is still a $200 million market while human translation is a $34 billion market. In many cases, human translators are needed to clean up the grammar and syntax messes that machine translators leave behind. Many human translators have to start from scratch with a document that has been translated by machine, because the quality can be so bad. That Star Trek ideal of instant, machine-based translation is still far in the future.