Machine Translation: Oversold And Under-delivered, But Who Is To Blame?
We all know the promise of MT: cheap translations, lingua franca, and the ability to keep up with the content explosion. Some folks have been standing on the street corner preaching the end of the world (ok, maybe just translation as we know it) for some time. And just like they do to the guy on the street corner, people have been averting their eyes and stepping around him and the topic for years. (Full disclosure: I am a fellow MT zealot).
The question, though, is why have we not realized the vision? Is it the fault of MT technology providers? Perhaps the universities, where much of this technology is birthed, just aren’t turning out good enough doctoral candidates? No, surely it’s the consumers’ fault; they’re just too demanding!
The real answer, in my not so humble opinion, is: it’s my own fault. Mine and the rest of the translation buyers’. I don’t intend to let the suppliers, academics, developers, and others completely off the hook, mind you; they all have some work to do. However, nobody controls the pace of the technology and its adoption like the buyers.
I wish I had a dollar for every time I heard someone say “MT doesn’t work for us,” or “The quality isn’t good enough for our content.” All too frequently, the answer to my follow-up question (“Have you actually tried it?”) is, “Well, no.” Often it comes down to risk aversion. People are afraid that, if a quality mistake or two creeps in, they will lose credibility. But I haven’t spoken to a single executive who wasn’t willing to accept a quality excursion or two for a dramatic reduction in cost.
And what about the customers? I have personal experience and data that says customers prefer timely data they can use over “pretty” content. Of course, there are always exceptions; certain industries, like medical devices, have very strict quality standards, and certain languages and cultures have less tolerance for imperfect translation.
But on the whole, I have found that MT with post-editing can be used successfully without upsetting the customer or your credibility. In fact, I think there is only upside to MT. I challenge anyone who hasn’t tested machine translation in their ecosystem to take a risk. You might just find yourself on the right side of executive scrutiny.