More than words

When officials at Joint Forces Command got a look at voice-to-voice translation technology that IBM Corp. was developing for use at the 2008 Beijing Olympics, the focus of that work quickly changed.

"We demonstrated the technology, and they immediately said, 'We need that in Iraq in Iraqi Arabic,'" said Gary Ambrose, an IBM vice president who works with Defense Department customers.

So instead of developing the software to initially work with Mandarin Chinese, the focus changed to Iraqi Arabic. The software, called the Multilingual Automatic Speech-to-Speech Translator, or Mastor, enables two-way, free-form, speaker-independent speech translation.

"In the past, translators were speaker dependent, in that they would have to memorize the user's voice," Ambrose said. "They were not free form, because they relied on a set of canned phrases."

IBM's new technology is a step beyond previous voice recognition and translation solutions, he said.

The need for voice translation technology, as well as machine translation of paper documents, is sure to remain strong as U.S. forces continue to have extended deployments overseas. Joint Forces Command is testing the technology in specific environments for specific tasks, said Wayne Richards, branch chief with the Joint Forces Capabilities Division.

Speaking in rungs

Having to work with people that use a wide variety of Arabic dialects, for example, is one of the challenges the technology faces. Translation libraries are built up over time to handle those dialects.

"Not every Iraqi can read or write modern, standard Arabic, so they speak their dialects," Richards said. "Those translation data libraries that we are able to create based on the data we collect improved tremendously the accuracy of the translations."

There has always been a need for translators during conflicts, but the need has never been greater than in the operations in Iraq and Afghanistan, he said. Unlike during the Vietnam War, for example, a large part of the work being done in Iraq is for civil applications such as training police recruits and supporting utility ministries.

U.S. military leaders in Iraq found out quickly that they were short on human translators. Most translators are trained to support efforts with high-level Defense Department officials, Richards said.

"At the tactical level, where human translators were also required, there just weren't enough to go around," he said. "That presented a very significant challenge for our forces."

This was especially apparent when coalition forces set up secure areas in and around Baghdad. Soldiers working checkpoints could not communicate with locals, thus making their dealings with locals feeble or unnecessarily intimidating.

"We had instances where local nationals were shot or where we were shot at, resulting in death or serious injury," Richards said. A speech translation device or human translator may have prevented such instances, he said.

Because the technology is still being developed, U.S. soldiers use it in relatively controlled environments, such as in training Iraqi police and military forces. Those settings have less background noise and stress than do house-to-house patrols, where bullets could fly.

The technology is software and, in most cases, is running on a rugged laptop with low-noise, handheld microphones. It works almost in real time, Ambrose said.

"You speak into it in English, it prints on the screen exactly what you said in English, and it speaks back to you in Iraqi Arabic," Ambrose said. "It prints on the screen in Iraqi Arabic what it translated, and it also prints on the screen in English the literal translation of what it said in Iraqi Arabic, so you can be certain that it conveyed the right thought."

Words into actions

Test results on the software are not available yet, but indications are that it is performing well, Richards said.

And challenges still exist. Joint Forces Command would like a hands-free version that a soldier could wear and use during combat operations.

"The automated speech recognition technology can obviously get better," Richards said. "It's performing very well in its current configuration, but we want to get to a point where recognition of whatever is being spoken is 98 percent right, 99 percent of the time."

As work continues on voice-to-voice translation, there is also a continuing need for machine translation of text documents in places such as Iraq and Afghanistan, said Naquib Hatami, CEO of CiyaSoft Corp., a Manassas, Va., provider of natural language processing solutions.

The company's technology is capable of translating handwritten documents, first by cleaning up document degradation, such as dirt or stains.

Next, it uses character recognition technology to convert documents into text, and then finally translates the text from languages such as Farsi and Pashto.

"We designed the product this way, because the majority of documents that you want to process have some kind of degradation," Hatami said. "The documents might have noise in them, coffee stains and other problems. There are 35 to 50 types of degradation that can occur on a document."

The other challenge is similar to those with voice translation: dialects and nuances with the languages.

"The challenges we're facing are in those regions where writing is not standardized," Hatami said.

In English, for example, a sentence begins with a capitalized word and ends with a period.

Proper nouns are capitalized. There are courtesy titles, such as "Mrs." and "Dr." But in Farsi and Arabic, the rules are different. Most of the time, people do not put a period at the end of a sentence, so knowing where one sentence ends and another begins can be a problem.

The languages also have compound words. And because Arabic and Farsi are influenced by poetry, word arrangement doesn't necessarily change the meaning of a sentence. For example, in English, "I am going to school," and "Am I going to school?" have different meanings. In Farsi, the differing order of these words might not change the meaning.

Despite that, the technology is advanced enough to overcome those challenges, Hatami said. And Richards said the same could be said for voice translation.

"It is very functional," he said. "We didn't believe we could get to the point where we are today."

Staff Writer Doug Beizer can be reached at dbeizer@postnewsweektech.com.