Companies seek to break language barrier

Rival approaches to translation vie for emerging market

Government agencies and departments need to make their online Web sites and services available to a growing multilingual population. Few of them, however, have the information technology resources to manage separate non-English versions of their sites.

That deficit has renewed interest in automated ways of creating and maintaining such sites, and translation technology vendors have stepped up to meet that demand.

Will Fleming, president and chief executive officer of MotionPoint, which uses a combination of human translators and automation, said the growing demand stems from the citizen "self-serve" model that governments are promoting for online services.

"Self-serve means a greater dissemination of information," Fleming said. "Ten years ago, people didn't go online for information. They made requests and waited for the printed material to arrive."

The increasing reliance on the Web for information has raised demand for the translation of brochures, documents and other online information, Fleming said. But because computers are not adept at understanding idioms or interpreting the correct meaning of a word with multiple definitions, machine translation earned a reputation for producing garbled text. That gives MotionPoint, with its reliance on human translators, an opening with government organizations.

Independent consultant Robert Kuhns, who works on machine translation with Sun Microsystems and IDC Research, said the overall market for translation in 2007 will range from $6 billion to $10 billion.

"The interest is high and will probably go higher, and that will push the demand up even if there's no quantum leap in the quality of machine translation," Kuhns said.

Orange County, Fla., uses MotionPoint to serve its Spanish-speaking residents, which constitute more than 22 percent of its population.

"Our Hispanic population is already large and growing, plus we interact with a lot of companies in other countries that are allowed to do business here," said Rafael Mena, Orange County's chief information officer. "For us, it's a no-brainer."

MotionPoint created the Spanish version of Orange County's Web site in December 2005 and now maintains it. MotionPoint's technology tools alert the Spanish site's administrators when someone adds new images, links or other elements to the English site so that they can keep the Spanish site synchronized.

Traditional machine translation tools, some more than 30 years old, use rules that determine how a computer translates from one language to another. The quality of those software tools depends on how well the programmers understand the practical aspects of business, popular culture and the various idiomatic nuances involved in both languages, said Toby Bell, a research director at Gartner.

As a result, Bell said, such tools generally are good at providing the site's main idea when that is all that is necessary. "You really need human translators for more nuanced content," he added.

University research is likely to tilt the equation in the next couple of years, Bell said. For example, one advance that has already attracted attention is machine translation based on statistical rules. Computer programs trawl through large amounts of material already translated from one language to another to find best matches of words and phrases. Then they apply that training to new translations.

The statistical approach harnesses today's greater computing power, making other rules-based approaches unnecessary, said Kirti Vashee, vice president of sales and marketing at Language Weaver, a 4-year-old company formed to commercialize research conducted at the University of Southern California.

Research creates new translation tech

New research into statistical machine translation could lead to more accurate automated processes. Statistical translation uses existing text translations to give computers references for new ones. The idea is to more accurately translate idiomatic expressions and other nuances of language.

"Every six months, [this technique] produces meaningful advances in machine translation," said Kirti Vashee, vice president of sales and marketing at Language Weaver. "As the online world advances, we are seeing more and more translations in digital format." Language Weaver is commercializing research from the University of Southern California.

The statistical approach will reduce but not eliminate the need for human translators, said Robert Kuhns, a consultant specializing in translation technology.

— Brian Robinson

NEXT STORY: 'Patriot search'... priceless!