The technology could help the military overcome a major hurdle in Iraq, which is the inability of most soldiers to speak Arabic beyond basic phrases, and a shortage of interpreters, International Business Machines Corp and military officials said.
IBM says it has delivered 35 notebook computers with the voice recognition software to be initially used by medical personnel, US Special Operations forces and the US Marine Corps. It will be used to ease communication in medical situations and with Iraqi security forces and citizens.
For now, however, it will not be used in combat or conflict situations that require split-second communications and decision-making, according to IBM.
“Our goal is to enable units operating in areas where human interpreters are scarce to communicate effectively with speakers of different languages in real-world tactical situations,” said Wayne Richards, branch chief of the US Joint Forces Capabilities Division.
IBM of Armonk, New York, has long been developing speech recognition and translation technology for use in commercial, consumer and military applications.
The technology being deployed in Iraq, called multilingual automatic speech-to-speech translator, or Mastor, has been in development since 2001, said David Nahamoo, chief technology officer for human language technologies at IBM’s research business.
A language barrier separates
“In those situations where the US military has to interact with the Iraqi forces or citizens, this language barrier is really affecting their performance,” Nahamoo said.
Using a Mastor-equipped laptop or a hand-held computer, a user speaks into a microphone and the software recognises and translates the speech, then vocalises the translation for the other person to hear, Nahamoo said.
The technology differs from existing translation software in that it is not limited to pre-programmed phrases, IBM said.
Instead, it recognises the way people speak, with variations in grammar, word order and sentence structure, Nahamoo said.
Because no technology can flawlessly translate languages, IBM’s Mastor can suggest up to three possible interpretations on a text screen first. That gives users the ability to prevent wrong translations, but it means words are not translated instantaneously.
The effect is like a conversation in which an interpreter waits for a speaker to complete a sentence, then translates it for the listener, Nahamoo said.
The IBM technology can translate more than 50,000 English words and 100,000 words in Iraqi Arabic, IBM said.
Eventually, Nahamoo said, the technology could find its way into commercial settings where many languages are spoken, such as banking, aerospace and defence, and law enforcement. Tourists could use the technology as well, he said.