Subscribe Share/Bookmark

Monday, April 26, 2010

Breaking Down Language Barriers on the Battlefield

Breaking Down Language Barriers on the Battlefield

May 2010

By Grace V. Jean , National Defense Magazine

In a scene from the Academy Award-winning film, “The Hurt Locker,” a native Iraqi translator helps a U.S. Army bomb technician communicate with an Arabic-speaking civilian who reveals that he has time-triggered explosives padlocked to his body. Precious seconds tick away as the exchanges between the defuser, interpreter and would-be suicide bomber grow harried under duress.

Hollywood’s depiction of the language barrier problem reflects a very real one on the battlefield. Troops attempt to interact with the local population on a daily basis, but without enough human translators to go around, miscommunications abound and opportunities for mutual understanding are lost.

In recent years, the Defense Department has deployed electronic translation devices to improve the situation. But most of the technologies provide only phrase-based English-to-foreign language communications. Troops said they would prefer to have handheld systems capable of instantaneous two-way translations so that they can hold normal conversations with civilians.

The Defense Advanced Research Projects Agency in 2007 awarded a contract to McLean, Va.-based Applications Technology Inc. to continue developing a portable two-way translator that could be used at vehicle checkpoints in Iraq. The prototype, which the company is readying for commercialization, will allow U.S. troops and Iraqi citizens to converse with each other in their native languages. A soldier can ask a question in English and the phrase will emit from the device’s speakers in Arabic. The Iraqi civilian can answer in Arabic and the translator will play it back in English.

Steve Cook, AppTek’s chief technology officer, says the system integrates speech recognition, machine translation and text-to-speech capabilities.

The machine translation component is a hybrid blend of two processes: statistical translation and rule-based translation. Most systems function on one or the other.

Rule-based machine translation breaks down sentences and applies linguistic rules to generate the translated text, Cook explains. But the process yields word-by-word interpretations that may not “read” well to a native speaker. Statistical translation, on the other hand, relies upon representative texts that a computer uses to correlate sentences and then generate translations. It may drop words or sacrifice accuracy for the sake of fluency.

“By hybridizing the system, what we’re trying to do is overcome those two shortcomings,” says Cook. The combination yields a more informative translation in the end. AppTek’s speech recognition and speech translation engines help to complete the process.

Mike Veronis, the company’s business developer, adds that the software covers about 15 Arabic dialects because no one speaks modern standard Arabic on the street.

Language translation technologies are not only useful for the Defense Department but also for other government sectors.

“We have a huge and growing demand to understand what’s being said and written in other languages,” says L. Paul Bremer, former U.S. ambassador to Iraq, who joined AppTek’s board after experiencing some of the translation problems firsthand.

When he arrived in Iraq less than a month after the government there fell in 2003, there was an explosion of free speech. More than 100 Arabic-language newspapers were being published in Baghdad alone. His staff of 3,000 shared only a half-dozen translators, he tells National Defense. “There was no way in the world that human translators could follow what was being said in 100 newspapers, on various radio stations and eventually television stations on a daily basis,” he recalls. “This was an immediate problem.”

Technology including automatic speech recognition and machine translators can make human translators more effective, he says. “We think the productivity gains are north of 80 percent when you put this kind of technology in place,” says Bremer.

Though machine translators are hitting the market in larger numbers, the intelligence and national security communities are still placing their primary emphasis on hiring thousands of human translators. It’s a $14 billon industry. Relying on human translators at the exclusion of technology is creating problems in many agencies, Bremer says. There’s simply too much information to track in all languages around the globe.

The White House last fall released a paper that called attention to the problem by challenging those in the research community to develop “highly-accurate real-time translation of the major languages of the world.”

To advance the technologies would require millions of dollars in investments a year. But that would be a drop in the bucket compared to the $4.5 billion the government pays annually for translator contracts in Iraq, technologists say.

Human translators feel threatened by the technology because they think it will supplant their jobs. “It’s not going to replace human translators. We’re not at the ‘Star Trek’ moment yet,” says Bremer. But the technologies “can make them much more productive and increase the capability of analysts throughout the intelligence community and elsewhere to cope with this unprecedented amount of information that’s coming across their desks every day.”

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.