“And the heart is hard to translate/ It has a language of its own” —Florence + The Machine (“All This And Heaven Too”)
OVER THE YEARS, people have called me a great many things. Music obsessive. Mr. A (a high school nickname). Technology dilettante. Customer experience evangelist (I even got paid to be called that!). Naïve. Punk. Absurd little man (that one, flung at me in a bar in the East Village in New York City, hurt a bit. C’mon, I’m slightly above average height for an American male! Your tears and donations to my therapy fund are welcome.) Suave heartthrob with Cary Grant’s debonair flair. OK, maybe only the voices inside my head called me that last one.
But one thing I have never been called is a Luddite. Although I am definitely—and often defiantly—skeptical of the techno-utopian inclinations of Silicon Valley, I try to remain open-minded about the capabilities of technology, the pace of technological advancement, and tech’s potential for good. Yet there is one technology that always seemed to promise more than it would ever be able to deliver. A technology so cool—if it would only work as advertised—that its presence would make us forget those jetpacks we were all promised. A technology for which my incredulity could well have gotten me labeled a technology antagonist. That technology: real-time machine translation.
As an inveterate traveler, the idea of being able to seamlessly communicate with pretty much anyone on the planet holds more than a little appeal. In fact, it would be one of the few “Take my money now!” technologies—if it succeeded. About 10 years ago, I spent a lovely meal with some data scientists and linguists from IBM who tried to convince me that we were on the cusp of conquering language barriers in narrow, constrained contexts such as translation of medical jargon. Even that slimmed-down promise did not really go anywhere, and I remained persuaded that we were all stuck learning other languages the hard way if we wanted to communicate across cultures. About two and a half years ago, I even wrote a column in this very publication that essentially urged caution when considering machine translation for projects such as chatbot dialogues.
Yes, this is another column in which I recognize that I might have been wrong. It wouldn’t be the 12,383rd time. Let me be clear: We ain’t there yet. Today, I cannot call a company in Budapest, Hungary, and clearly communicate with a salesperson while I spoke English and they spoke Magyar—and as a crazy person who tried to learn Magyar, I can’t tell you how much I wish I could.
But I am seeing real glimmers of hope. And, ironically enough for a technology touted as machine translation, that hope comes in the form of human labor. One vendor now touts its solution as “AI-Powered, Human-Refined Translation,” and the inclusion of human judgment is what is providing me a slight feeling of anticipation. Human editors review, hone, or outright fix the translations. That human work—often powered by a gig-like labor model—then feeds back into a machine learning system to improve future translations. This human-in-the-loop model allows for a much wider array of domains to be successfully translated and brings us closer to good—and maybe even great—machine translation.
We’ve got a ways to go before these early efforts, often focused on “near-real-time” use cases, turn into the universal Babel Fish for which I’ve always yearned. But I am buoyed by even the minutest sliver of hope here! Now if I could only get the voices inside my head to start complimenting my abundant charms in Swahili…
Ian Jacobs is a vice president and research director at Forrester Research.