Machine Translation in Automated Workflow

Machine Translation (MT) was supposed to be the great solution for our translation needs. But how big a role does it have in the current translation industry? As far as I can see, in producing translations MT still has a very small role in the automated translation workflow, which relies much more on translation memories (TMs). But when we move to reading texts in foreign languages, such as web pages, MT has a much bigger role. The "Translate this" button provided by Google and many other search engines has a lot of users.

So at least for now MT is used primarily for getting a gist translation of a web site or a document. In producing a translation to be published, TMs and human translators still rule - perhaps with the exception of very strictly regulated or simplified texts which can be translated correctly enough by machine.

The reason is simple: the result from MT is still mostly so bad that a human translator can produce a full translation faster than what it would take to edit a machine-translated text.

Having said that, I do believe that major changes in MT are about to happen. I refer to speech recognition. In around year 2000, companies tried to build a Hal, which would understand not only what you said, but what you meant. Well, I have known my husband for almost 30 years, and I still do not always understand what he means, so that would have been a tall order indeed. Now, however, we have excellent applications based on recognizing a limited vocabulary - think of the times you have called an 800 number and talked to a machine which has understood you quite well.

MT will find more and more uses also in producing translations, when it will limit itself to a certain topic which it handles well, and combine efficiently with TMs. For gist translations, when one wants to understand e.g. the idea on a web page, MT is already quite useful.