“Reports of [our] death have been greatly exaggerated.” (with thanks to Mark Twain)
Advances in machine translation (MT) have spawned a wave of popular speculation about the imminent end of the translation profession. Like Mark Twain, we beg to differ on those premature reports of the industry’s death, though we do agree that those advancements have opened up reasonable and productive uses of MT. Here is an insider’s perspective of how to distinguish when MT can be used as a great tool and when it is much better to rely primarily on "human" translators.
Web-based communication is multilingual. Long gone are the days where you could expect a webpage, a Facebook post, or a tweet to be in English (or in another language of your preference). That's where free and generic machine translation services such as Google Translate or Microsoft Bing Translator can be excellent tools for gaining an approximate idea of what that foreign-language communication says.
We have all heard at least one of the many, many mistranslations arising from these kinds of services—and while many of these stories may be urban legends, enough are true to indicate that you'll need to be cautious about the reliability of those "gisted" translations. Still, there are many helpful uses for the casual user.
Any content that is uploaded to these online machine translation services is stored and analyzed by Google and Microsoft as fresh content for later translations.
There's more to MT than these generic systems, though. Since the generic systems are trained by the vast amount of online content, they are often unreliable when used with a specific terminology or a specific style of writing. This is where customized MT that has been trained on a specific data set comes into play.
This training can take two routes depending on what kind of MT is used. If it's a statistical machine translation engine, the training data will consist mostly of monolingual or bilingual texts which the computer analyzes and uses for later translation. If it's a rules-based machine translation engine, the data set will be mostly terminology along with specifications for how to use that terminology and style of language. And increasingly, a hybrid model that combines both of these technologies is being used.
The output of these customized systems can be more reliable, especially if the original texts were prepped for that purpose with relatively short sentences or very controlled terminology. However, the results will typically have to be post-edited manually, unless the content owner feels that its readers will tolerate errors and non-idiomatic style (for example, some large companies use this strategy for their knowledge bases).
It's very difficult to give general answers on how much time is saved with post-editing compared to translating from scratch. First of all, of course, it depends on the quality of the machine translation engine, but it also depends on the language combination: machine translation for more closely related languages often produces more idiomatic translations, while other language combinations are less suited for that kind of language transfer.
Finally, the choice between MT and human translation depends on the kind of text you need to translate. A first pass through machine translation will not make sense for creative texts (such as any kind of marketing materials). It is also hard to make the case for a first MT pass if the style and idiomatic expression of the translated text needs to be impeccable. Still, there are some functional texts and materials in some language combinations that can yield adequate results from a combination of machine translation and human post-editing. Your ATA professionals will be glad to guide you in deciding whether that's the case with your next project.