Whoever ends up becoming the 7 billionth person born on Earth may want to learn a second or third language to be able to thrive in a world connected by fast air travel and even faster Internet communication. Otherwise, growing arrays of mobile translation devices stand ready to bridge the newcomer's linguistic gaps in a crowded global village.
Translation machines already allow businesses to operate global customer services. U.S. soldiers have used various handheld devices to speak Pashto and Dari in Afghanistan. Google Translate makes foreign language websites accessible for anyone with Internet access, and some translation apps even give people speech translation services right on their smartphones. But such gadgets and software still rely on written works painstakingly translated by humans over thousands of years.
"All the modern machine translation systems are built starting from the available data," said Daniel Marcu, chief technology officer for SDL's language technologies division. "They collect vast amounts of data translated by humans in the past, and then learn from that data how to translate new sentences."
The dream of universal translation
That data comes from digital versions of any written work ever translated from one language into another. But reliance upon such data also creates a huge stumbling block to achieving a universal translator on par with "Star Trek" or the fictional Babel fish, because no direct translation exists between most of the world's languages.
"Universal translation is a bizarre dream that simply doesn't correspond to what has been or could happen," said David Bellos, director of the program in translation and intercultural communication at Princeton University.
People did not do intensive language translation for most of human history until the invention of the printing press in the 1400s, as detailed in a recent book written by Bellos, titled, "Is That a Fish in Your Ear?" (Penguin Group, 2011). Even then, people living as far apart as Icelanders and Italians had little reason to engage in direct translations between their languages until the rise of modern global travel and communication.
Between 5,000 and 7,000 forms of speech have been identified as languages in the world, Bellos explained. If humans even tried to do direct bilateral translations among just 5,000 languages, they end up facing the absurd challenge of translating almost 25 million complete language interactions.
As a result, new translation gadgets or software must do what humans have done for millennia; they cheat with a third language.
Praise for the wizard tongue
The cheat makes use of a so-called pivot language, also known as a vehicular language, which acts as a common second language spoken by two different cultures. The Roman Empire ensured that Latin served as a pivot language for Europe for thousands of years. Arabic served as a pivot language for a modern numbers system that started out in India, but ended up reaching Europe through translation into Latin.
Today's world has about 80 languages acting as pivot languages, but English stands out among many machine translation services. The popularity of translations between English and many other languages — for everything from business manuals to the bestselling "Harry Potter" books — has provided mountains of usable data for translation services to go from, say, Vietnamese to English, and then from English to Hebrew.
"The real magic of 'Harry Potter' is that it helped to improve Google Translate's service," Bellos told InnovationNewsDaily. He praised Google Translate as a "quantum leap in translation accessibility" despite its current clunkiness.
Creating a billion-dollar market
Early innovations in machine translation by the likes of Google, IBM, Microsoft and smaller companies such as Language Weaver have relied heavily upon U.S. government funding — especially from the Pentagon's DARPA research arm. Private investment by corporations has only overtaken government investment in the past several years, according to Marcu, the SDL chief technology officer.
The market for machine translation services remains less than $100 million per year, but there is growing demand for customer service translation, Marcu said. Previously, he helped found Language Weaver, a translation technology company acquired by SDL last year.
"Embedding machine translation into solutions aimed squarely at that [customer care] market is something where we see significant growth at Language Weaver," Marcu said.
Such machine translation was still bad enough six or seven years ago to make it more expensive than human translation, Marcu explained. Now, human translators use machine translations to become more efficient in their own jobs.
Steady improvements in machine translation could lead to many more people using mobile translation devices or programs on their smartphones. But first, engineers must figure out how such translation can fit smoothly into the flow of human interactions.
"We're pretty good about using devices to help ourselves, but how to use devices between groups of people is not straightforward," Marcu said. "I think we need some clever thinking and breakthroughs in that area; once we have that, we'll see true speech-to-speech technologies."
Still, even the most magical universal translators in the foreseeable future must rely upon the work of humans translating corporate documents and great works of literature, according to both Bellos and Marcu.
"Behind all of that impressive engineering, human translators have at some point in the past done the original translations," Bellos said.