Machine Learning Based Translation Solution for Military Forces Abroad

Machine Learning Based Translation Solution for Military Forces Abroad

machine learning

This post is also available in: heעברית (Hebrew)

A new digital system will supply translation services to US military expeditionary forces around the world. The US military recently revealed the Machine Foreign Language Translation System (MFLTS) system to 700 users. According to c4isrnet.com, commanders are planning this summer to deploy the software across the Project Manager Distributed Common Ground System.

Mike Doney, the product manager for MFLTS, explained that given the Army’s expanding expeditionary role, “there is no way to train a sufficient number of linguists in so many relatively uncommon and obscure languages. So we’re trying to augment the current two-legged capability with a modest level of capability to as many soldiers as possible.”

The system combines voice recognition and a speech synthesizer to enable rapid translation. The new version will incorporate the ability to translate scanned text. Presently, the software can accommodate the two spoken languages most commonly encountered by Army personnel in the field, Iraqi Arabic and Pashto.

Although two users can converse fluently through the system, it wasn’t designed to deliver a high level of sophistication. With conversational fluency defined on a zero-to-five scale, human linguists operate at a four, whereas MFLTS works at a one.

That may very often prove sufficient for soldiers’ needs, said Tracy Blocker, MFLTS product director lead within Army Training and Doctrine Command’s Capability Manager-Biometrics.

“At a checkpoint you can say: ‘Do you have any weapons? Get out of the vehicle.’ If the foreign language speaker complies with that, we consider that a success. Our currently released product meets those requirements,” Blocker said.

“There is a lot of information that needs to get looked at. There are a lot of things intelligence analysts want to use to contribute to situational understanding of the operational environment,” Doney said. “These might be papers captured in the field, or electronic documents taken from web pages or social media. Whatever the source data is, the text to text translator will process that input and put it out in English.”

The paper-translation capability has proven an engineering challenge. “The hard task was ingesting the hard copy document through some scanning or photographic process that yields a clear enough image that an optical capability could turn that image into text,” Doney said. “We have not yet fully deployed that capability, but we are very close.”

The translator relies on machine learning, a sub-discipline within the general field of big data analytics. The development team will be leaning heavily on that capability as it aims to roll out speech and text translation capabilities for more than 60 languages over the next several years. The end goal is a massive online downloadable database, which would enable soldiers to select from a portfolio of language packages, downloading from a web portal as needed.