Understanding natural language is a good example of something that humans can do almost without thinking but computers find surprisingly tricky (although they’re getting better at it).

If understanding one language is hard for a computer, translating into another language is even harder. Machine Translation (or MT) is the field of computer science that tackles the problem. And it’s proved to be an elusive goal that always seems to be a few years away from being fully realized. Yet in recent years MT has made major strides.

The best-known commercial service is probably Google Translate, which does a pretty respectable job with many language pairs and will happily translate whole web pages more or less instantaneously.

Google Translate uses several techniques to accomplish this, but it ultimately relies heavily on human translators, who provide raw material in the form of previously translated texts. Google Translate cleverly crunches huge amounts of data to ‘learn’ how to translate new sentences better.

Playing to the computer’s strengths

While computers find certain tasks that humans might find trivial pretty difficult, the opposite is also true: computers can make short work of tasks that humans would find overwhelming. Performing various mathematical feats and searching databases containing millions of records, for example.

Computer Assisted Translation (CAT) tools aim to pair up the superior language abilities of humans with a computer’s capacity for storing and quickly sifting through lots of data. In this case, the data are text strings (usually sentences) that a human translator has already translated.

The CAT tool takes the current sentence and looks through its database (usually known as a ‘Translation Memory’ or TM) trying to find a match. It may find a 100% match (i.e. the exact same sentence has come up before), a partial match that the human translator can use as the basis for the new sentence being translated, or no match. Once the translator has translated the current sentence, that gets added to the TM so that if it comes up again, it will show up as a match.

CAT tools such as Memsource and Smartling all work on this principle, allowing translators to work more quickly and with more consistency. What’s more, many CAT tools also leverage some form of MT, giving the human translator an extra tool to work more effectively with.

Which is best for you?

Of course each approach has its benefits, depending on the situation. Running something through Google Translate is great to gain a quick understanding of the content, if you are not concerned about confidentiality, grammar or the occasional mistranslation. However, if you are looking to create a message that speaks on a personal and cultural level, a seasoned linguist using a CAT tool will help to better connect you to the audience in your target country.