The Changing Role of the Translator

I just read an interesting op-ed by David Bellos in last weekend's New York Times. In "I, Translator", the director of the Princeton translation program offers a brief overview of machine translation, from its infancy in the U.S. at the start of the Cold War to its current Google Translate iteration and concludes that machine translation cannot replace human literary translation.

A point well taken, but literary translators constitute a fraction of the translation community. Most of us don't work on War and Peace but on such prosaic texts as legal contracts, medical documents or technical manuals. Quite a bit of these documents consist of "boilerplate" or otherwise at least semi-standard text -- that's why translation memory tools have become so useful to human translators. Because of this standardization and repetition, such technical text does lend itself to machine translation, just as the original text is a prime candidate for standard content management systems. This move towards further automation may put both translators and technical writers out of business -- unless we adapt.

As Google Translate and similar systems parse ever-increasing quantities of multi-lingual web content -- including my own bilingual website -- their translation engines approximate human translation ever more closely, at least in relatively common language combinations, such as German-English. But while a business contract doesn't need to be good prose, it does need to be extremely accurate -- approximation just isn't good enough. That's also true for medical and technical texts. So while Google Translate can provide a basic idea of what the parties to a contract agreed upon or how to operate a particular piece of machinery, humans still need to ensure that nothing has been mistranslated and that the end result flows smoothly and is easy to understand.

That's where we human translators come in: as editors to verify and refine the somewhat crude machine translation. To some extent we are already performing this role when we edit the "fuzzy matches" provided by our translation memory tools. Post-editing machine-translated text would be similar, just with a much larger translation memory. The question now becomes: how do we come up with a compensation model where we can still make a living, even though this editing will garner a much lower per-word rate than traditional translation?