Language professor doubts machine translation’s future

April 4, 2016

I wrote recently that the capabilities of machine natural-language translators will reach or exceed those of human translators. Not everyone agrees. For example, David Arbesú, an assistant professor of Spanish at the University of South Florida, sees definite limits to machine translation. Writing March 28 in The Conversation, he contends that “…language contains nuances that are impossible for computers ever to learn how to interpret.”

He cites as an example of potential problems the homonym “saw” with regard to mistakes his students have made. But a computer could easily determine that a sentence containing “saw” and “parents” likely refers to a recent visit, not an attempt at dismemberment using a carpentry tool. He further cites Google Translate’s faulty translation of “No es bueno dormir mucho” (it is not good to sleep too much) as “Not good sleep much.” But I’m sure Google Translate will improve. Bing Translator already has.

Arbesú acknowledges that engineers working on machine translation might try to “log all the rules by which languages work.” But, he contends, languages don’t work that way.

But what Arbesú overlooks is that with machine learning, computers don’t work that way either. A machine translator needn’t begin with algorithms—it can begin with data and derive its own algorithms.

Microsoft Research, for example, uses a data-driven approach that makes it unnecessary to write explicit rules. With the data-driven approach, machines learn how to translate using human-translated parallel texts.

In addition to offering the Bing Translator portal, Microsoft recently released a new version of the Microsoft Translator API. The API, the company said, “…is the first end-to-end speech translation solution optimized for real-life conversations (vs. simple human to machine commands)….” Microsoft reports that mobile operator Tele 2 of Sweden has integrated the API into its PBX to support real-time phone-call translations on its cellular network.

Of course, machines will never be perfect translators, but neither will people. As the late Peter Newmark, English professor of translation at the University of Surrey, put it, “There is no such thing as a perfect, ideal, or ‘correct’ translation. A translator is always trying to extend his knowledge and improve his means of expression; he is always pursuing facts and words.”

If it becomes pointless to learn foreign languages for the purpose of communication, students can always focus on learning machine languages. As Arbesú points out, the Florida State Senate had proposed a bill providing that high schools offer foreign-language instruction credits for courses in computer coding, and that the state college system must recognize those credits as meeting foreign-language credit requirements. Parents and students would be warned, however, that colleges outside the state system may not accept the coding credits in lieu of foreign-language credits.

It looks as if the Florida bill will not become law, and that’s fine. It’s good to learn computer code, and it’s good to learn a foreign language—there’s no need to confuse the two. Computer prowess at natural languages won’t negate the benefits of people speaking multiple languages.

MIT Professor Lawrence Susskind, who is teaching an upcoming online course in negotiation at MIT (beginning April 26—click here for more information), says that negotiation requires trust, and one way to establish trust is to travel to meet your counterpart face-to-face despite the existence of email and Skype. Learning your partner’s language can be a further step in building trust.

Arbesú writes, “Needless to say, communication is only one of the many advantages of learning another language (and I would argue that it’s not even the most important one).” I completely agree with him on that point.

Sponsored Recommendations


To join the conversation, and become an exclusive member of Electronic Design, create an account today!