I am a big fan of machine translation, as long as it is used for the right purpose and reasonably. In this article, I want to explain something that some folks choose to ignore, when they rely on MT. These include translation agencies that slash vendors’ rates by 50% or more, just because they insert MT output into bilingual files before sending them to translation vendors.
Here is the problem: Most of the text, say 90%, in any technical translation is quite easy to translate. It is the remaining 10% that is tough. The bulk of translation is just as easy for a human translator to translate as it is for MT. A translation pro can process the 90% very quickly, sometimes almost unconsciously, since there is little intellectual effort involved. The main effort goes into translating the “challenging details.” For this reason, MT does not always give the performance improvement some people have come to expect, such as a 50% increase in a translator’s productivity. Whereas with a simple text, MT can indeed save 50% of the time required, it provides much less help in more challenging translations.
Here are three common areas where MT cannot play its part, leaving the main burden on a translator’s shoulders:
Number 1: Creative texts
Imagine translating a slogan. This is usually not a translation, but a transcreation, a cultural adaptation. MT fails miserably, since a machine cannot think creatively. It relies on patterns it found in previous translations, which are useless in this case.
Number 2: Short sentences
An Android app localized into Russian on my phone says “Освежающий,” whenever a page is reloading. The original was “Refreshing,” meaning reloading, but the translation, obviously provided by MT, says “Refreshing” as in “Refreshing juice.” This is a good example of a challenging original. Dealing with such challenges takes up most of a translator’s time.
Number 3: Ambiguous idioms
It is difficult for an MT engine to tell a literal phrase from an idiom, such as to “put a stake in the ground.” You need a human to assess the context correctly and choose the right meaning. For more information about why context is important, read the article “Context Is Everything.”
Another way to look at this problem is comparing it to revising other people’s translations. Incompetent translators can usually translate the easy parts pretty well, but fail to translate the “challenging details” properly. And, again, these parts are the ones that take up the most time. As a result, many translators are reluctant to revise translations done by someone else. They know it is they who will have to deal with the challenges.
The bottom line is that MT is good where it is easy for anyone to be good, even for bad translators. But when it comes to tackling real linguistic challenges, MT is of little help. In this sense, MT efficiency is overrated. Expecting MT to be equally effective with any kind of text is unrealistic. Asking a translation vendor for a 50% rate decrease does not reflect how effective MT really is.
Do you agree that in translation, the bulk of text is simple, and most of a translator’s time is spent on “challenging details”?