As a project manager at ALTA, I’ve seen my fair share of proofreading requests from clients. Sometimes the text has changed slightly; other times, clients just want to double-check on the quality of the translation. Frequently, though, a review of the translation finds the document was generated by some sort of translation software. In my experience, these documents have to be re-translated by a human every time.
Technology has come a long way, but translation software still has far to go before it could be deemed anywhere near trustworthy. Granted, if it worked well, I think I’d be out of a job! But I’m not alone in my opinion.
A recent CNN.com article reviewing translation software suites agrees, and the title pretty much says it all: Translation Software Still Fails – Badly. Even though the point of these products is to save you money, their poor quality, incorrect terminology, and the myriad of other issues plaguing machine-generated translations can end up costing you more in the long run. The bottom line is that translation software is currently incapable of deciphering the cultural and linguistic nuances of a typical business translation.
One example often referred to by my colleagues is how to get a machine to understand and correctly translate text such as “a drug-free workplace?” More often than not, the language translation software will spit back a translation that actually reads, “free drugs in the workplace.” Not exactly what you want in your employee handbook.
That being said, there have been some significant strides in translation software to aid human translators. Many translators utilize computer-assisted translation (CAT) software tools in order to speed up their translation and maintain consistency throughout documents. These tools search for repetition within or between documents, cutting down the time and cost of translation where applicable. And for upcoming projects with anticipated repetition, a memory of the initial translation can be leveraged against future documents.
Overall, though, I think the score reads, machines: 0, humans: 1 (take that, Terminator).