Even Microsoft gets it wrong (“Skype Translator”) September 29, 2014
Posted by Jill (@bonnjill) in Random musings, Uncategorized.trackback
From the latest ATA News Briefs:
New Translation Technology No Threat to Professional Translators
After 15 years of research, Microsoft has unveiled Skype Translator, a voice translator that will convert speech from one language to another in almost real time. The service—dubbed by the media as the “Star Trek Translator”—will be available for Windows 8 by the end of this year. Microsoft chief executive Satya Nadella describes the system as a neural network that “learns” from data, much like the human brain. “It’s not,” Nadella says, “just about daisy-chaining speech recognition, machine translation, and speech synthesis.” But according to Andy Way, associate professor of computing at Dublin City University, the hype promises more than can be delivered. Way says, “You’re more likely to have everything else in Star Trek before you ever get a universal translator.” Philipp Koehn, chair of the Machine Translation School of Informatics at the University of Edinburgh, agrees with Way. “Automatic spoken translation is a particular problem because you’re working with two imperfect technologies tied together—speech recognition and translation,” he says. Despite their imperfections, industry analysts say Skype Translator and other automated translation programs are here to stay. They believe globalization has driven the demand for translation beyond the availability of translators. European Commission Language Officer Angelique Petrits says that her organization translates two-million pages into 24 different languages every year. “The organization wouldn’t be able to fulfill its mission without up-to-date translation technology.” Petrits does not view machine translation as a replacement for human translators. “Technology is a tool that helps dealing with the scarce resources of translators by speeding up their work and allowing them to concentrate on the essentials. It also contributes to the consistency of terminology, crucial in EU texts,” Petrits says. Way, Koehn, and Petrits all insist that technology is not about to replace translators. As Way observes, “There is just so much translation to be done—people have estimated that only around five percent of what needs to be translated actually is—that good translators will never be out of a job.”
From “Tech Is Removing Language Barriers—but Will Jobs Be Lost in Translation?”
Guardian (United Kingdom) (09/19/14) Williams, Martin
Well, considering they keep talking about “spoken translation” I don’t think we have anything to worry about. Come on, engineers, if you are programming it you should AT LEAST know the difference between interpreting and translation. Wow.
Musings from an overworked translator 
I can’t wait to see what they do with the Thai language.
FWIW, I’d say that “speech translation” is probably more correct in this context than “interpreting”, because what the computer programs essentially do is use speech-to-text recognition to analyze the spoken words and output text, then “translate” the analyzed text into the other language, and then output that text again as speech. They don’t interpret like a human person between spoken words, there’s always the analysis/”translation” of written text as an intermediate step via the usual MT algorithms.
There was an interesting opinion piece in the FT the other day (September 25th) by Peter Thiel, former PayPal CEO. He argues that we shouldn’t be afraid of robots/computers: you don’t have to agree with him unreservedly that “Robots are our saviours”, but there’s no doubt we couldn’t translate the way we do today without computers. What really struck me, though, was the following:
“Men and machines are good at different things. People form plans and make decisions in complicated situations. We are less good at making sense of enormous amounts of data. Computers are exactly the opposite: they excel at efficient data processing but struggle to make basic judgments that would be simple for any human.”
That just about says it all as far as MT is concerned. Because it’s a dumb technology that uses brute computing power as a proxy for human intelligence, it can only ever play a supporting role for the incredible ability of the human brain to make sophisticated judgments. Data processing power is no substitute for intelligence (though you might think otherwise if you read what some of the MT evangelists claim). Translator-controlled MT might really work, if given a chance.
Good point, Robin. Thanks for bringing that up. It will never be good enough to stand on its own. They will never be able to get rid of us completely.