Operations Infrastructure for Real-time Translations
Getting to real time actually accomplishes more than merely getting translations out the door and in front of the reader faster. Paradoxically, at the same time it also promotes both increased quality and lower cost. That is because the real-time goal functions as a sort of guidepost, an organizing principle, towards reengineered processes that inevitably also have a positive impact on quality and cost.
This article was published in MultiLingual Computing magazine #115.
Put simply, with real-time translation sometimes there is not enough time to spend as much money as we do now. And with real-time translation, we are forced to optimize quality processes, both human and automated, in a way that can actually result in higher quality than today's cumbersome processes yield.
A colloquial definition of real time would be instantaneous or at least so fast as to be indistinguishable from instantaneous. But here the technical definition of real time works better: fast enough to be useful. For instance, the computer controlling the antilock brake system in your car has to do its number crunching fast enough to avoid you sliding off the road, but doesn’t need to be any faster.
For translation, we could usefully define a number of levels of real-time-ness, ranging from 250 milliseconds, which could be about right for certain applications, all the way up to weeks in some cases. In this sense, perhaps it would be better to refer to just-in-time translation or JIT, although that could have the unfortunate nuance of "at the last minute". Remember, even Japanese car companies are now backing away from the JIT model after experiencing some nasty production stoppages due to delayed parts deliveries.