Initially, 35 ruggedized laptops implementing the Arabic translation program will be used by such units as the Army's medical, research, intelligence, special operations commands along with Marine Corps units. David Nahamoo, speech technology CTO at IBM Research, said the combination of speech recognition, machine translation and text-to-speech rules allowed IBM to develop a common translation engine that is independent of languages.

When a particular language pair is defined, library elements are developed based on conversational use of the languages.

Core tools were developed in 2001 under DARPA programs like Babylon and Compact Aids for Speech Translation. The goal was to create software that could input speech in one language and generate synthesized conversational speech in another. Translation speed would be virtually the same as a human translator.

Speaking in either of two target languages, human speech is translated along with a text output. The speaker of the other language can reply in the same way. Mastor software can run on a laptop, PDA or tablet computer.

"The performance improvements we have realized in certain domains allow us to have response times of a second or two after the speech input," said Manny Athavale, principal researcher within IBM Technical Collaboration Services. "We think that's practical for a soldier working in real-world environments."

To date, Mastor has been developed for English translation to Iraqi Arabic, modern standard Arabic and Mandarin. Athavale said commercial products for domains such as medicine will be offered soon, and commercial viability will help determine what additional languages are offered.

The version deployed in Iraq can recognize and translate more than 50,000 English words and 100,000 Iraqi Arabic words. While there are no specific milestones, successful use of the 35 laptops may lead to more units being used in Iraq.