Navy lab helps troops speak the language
The Office of Naval Research is developing translation technology that will help ground troops and ship crews read the writing on the wall; here, soldiers in Falluja, Iraq, use posters to cover up anti-American graffiti.
SPC. Derek Gaines of the Army
Imagine you're a U.S. soldier in Baghdad, and you see an Arabic sign on a wall.
'It might say 'This way to the palace' or 'This way to a public restroom', ' said Joel L. Davis of the Office of Naval Research.
ONR's Division of Cognitive, Neural and Biomolecular Science and Technology is funding development of speech translation technology to help field troops interpret signs and other graphics in foreign lands.
'I always thought this would be mostly for the infantry, whether it be Army or Marine Corps,' said Davis, the scientific officer overseeing the program. But there will be uses in the blue-water Navy as well, such as during interdictions of foreign vessels, he said.
The tools probably won't be ready before U.S. forces leave Iraq, but prototypes created by SpeechGear Inc. of Northfield, Minn., are ready for field-testing. Davis said they would undergo trials during Army and Marine Corps exercises abroad this year.
'Roughly a year from now we'll ship products,' SpeechGear president Robert Palmquist said.
Translation the old-fashioned way'by human beings'requires extensive training of interpreters at an Army language school in Monterey, Calif. 'They are always training for the last war,' Davis said, and adding new languages is a long-term task.
The automated or online translator products now available do not meet the military's needs. The Defense Advanced Research Projects Agency is working on something called a phrase-o-lator, a voice-activated lookup table that spits out one of 500 pretranslated words or phrases.
ONR meanwhile is working with SpeechGear to produce more flexible tools.Using context
SpeechGear's Compadre suite consists of Interact, a bidirectional voice-to-voice speech translator; Camara, which uses a digital camera and optical character recognition software to translate images of written material; and Interprete, a handheld dictionary device to translate from voice or text.
The translation engine in Interact and Camara uses context to determine meaning and to distinguish between homophones such as to, two and too. Interact can do speech-to-text translation to make transcripts of conversations.
Davis cautioned that the suite is not yet at the stage of the Universal Translator that serves the Federation so well on 'Star Trek' TV shows.
'It's taken us at least four years to get to this point, and I don't think we're through yet,' he said.
Accuracy of two-way translation in conversation is hard to quantify, Palmquist said. 'Our metric for Interprete is to accurately communicate the meaning of a sentence,' he said.
For instance the phrase, 'How do I get to the subway?' could be translated as, 'Where is the subway? Please tell me.' Though not a literal translation, it's accurate. But the idiomatic 'Hold the phone' could be translated word for word and miss the meaning completely.
Such misunderstandings are easy to correct by rephrasing in the course of a conversation. 'The Interprete system works quite well for two people who want to communicate,' Palmquist said.
Davis added, 'If they don't want to cooperate, there's very little you can do.'
Camara would be most useful for languages with non-Roman alphabets, because looking up the words in a dictionary would be difficult if not impossible for English speakers.
SpeechGear incorporates translation software from other vendors in its products as needed. Camara now supports about 120 languages, Palmquist said, and Interprete has a much smaller list including European languages plus Arabic, Chinese, Japanese and Russian.
The prototype has a client-server architecture, but Palmquist said initial field tests probably would be on standalone handheld or notebook computers, which would have both client and server software installed to eliminate the need for communication links.
For real-world field use, Davis said, the distributed architecture would be more attractive because soldiers would not have to carry so much computing power.
Given voice and data communications in the field, a soldier could use a handheld device or a phone to digitize and compress voice and transmit it to the translation server, which would return a translation through the phone or the computer's speaker. A similar process could transmit digital images and return text translations.