Military Tests Portable Translators

Team Infidel

Forum Spin Doctor
USA Today
December 20, 2006
Pg. 8

Mandarin Chinese, Arabic challenge language machines
By Richard Willing, USA TODAY
Scientists have labored since at least the 1960s to build a machine that can perform real-time translation.
This device has been particularly elusive when it comes to translating between English and complex languages such as Arabic and Mandarin Chinese.
"You're talking about (languages) where everything is different — the words, the structure, even spelling and punctuation," says David Nahamoo, chief technical officer for IBM Research in Yorktown Heights, N.Y. "This is difficult stuff here."
The military, with $22 million in this year's budget for speech translation research, is taking the lead.
This year, the military's Defense Advanced Research Projects Agency (DARPA) began field-testing portable devices aimed at allowing soldiers to carry on conversations with Iraqis without relying on a human translator. The machines are designed to translate spoken English or Iraqi Arabic and play the translation in real time through a software-driven speaker mounted on a laptop computer or a small handheld device.
The devices, made by IBM, SRI International and Carnegie Mellon University, rely on speech recognition software, speech simulators and a vocabulary of about 150,000 English and Iraqi words.
Using statistical probability tables built into their software, they sort through likely word combinations to form phrases and full sentences.
DARPA also continues to test two other machines that use prerecorded phrases that can be tailored to a specific mission. The machines, built by VoxTec International and Integrated Wave Technologies, are the size of a personal digital assistant. They cannot carry on two-way conversations. Both have been used in Iraq.
In a related project, DARPA renewed funding in October for a second year of research aimed at translating vast amounts of foreign language data that are readily available — broadcasts on Al-Jazeera, website postings and online chats — in a format military and intelligence officers can query. The project, called Global Autonomous Language Exploitation (GALE), addresses a shortage of translators and the explosion of foreign-language material available to military and civilian intelligence officers.
"The volume of data defies any resource you can put against it, translators included," says Cliff LaCoursiere, co-founder of CallMiner, whose technology for analyzing call center traffic has been licensed by In-Q-Tel, the CIA's venture capital firm.
The GALE project has two major challenges:
•It is trying to develop software capable of translating material written or spoken in both the standard Arabic used by television reporters and in the regional dialects employed by interview subjects or in chat rooms.
•It is working to produce translated text that can be searched by intelligence officers looking for things such as real names, aliases, relationships and activities of groups whose names may have a number of different spellings.
To be useful, says John Makhoul, chief scientist at BBN Technologies in Cambridge, Mass., GALE cannot be "just a key word search" that brings back hundreds of unrelated matches.
"This must be a super-duper Google search that includes meaning," says Makhoul, whose company performs some of the GALE research.
In addition to Arabic, GALE researchers are developing a translation program for Mandarin Chinese.
Some of the technology has undergone tweaks. Integrated Wave Technologies of Fremont, Calif., added "Keep your dogs away, or we'll shoot them" to the repertoire of prerecorded phrases its voice response translator (VRT) can speak. Company President Timothy McCune says the phrase was added after soldiers in Iraq reported recurring problems with swarming pets.
The two-way translation machines had difficulty sorting out background noise from the voices that were supposed to be translated, says Wayne Richards, a top program officer at the Pentagon's Joint Forces Command in Norfolk, Va. They'll undergo a second round of testing in "benign" environments such as hospitals and offices before they are approved for use in the field.
"We've got to manage expectations," Richards says. "We're three, four years from solving it, but we can be pleasantly surprised."
GALE, too, is undergoing some growing pains. Researchers struggle to write software that recognizes when a speaker switches from formal Arabic to a dialect, says Alan Black, associate research professor at Carnegie Mellon University in Pittsburgh and a GALE project researcher.
Black says the goal — "to be not just accurate but culturally accurate" — is made more difficult by the great variety of spoken dialects and by large structural differences between English and Arabic and Chinese. Chinese nouns, for instance, have no gender. In Arabic, which does not use capital letters, many proper names are also common nouns or descriptive phrases.
"GALE can already get the gist of most television broadcasts," Black says. "But we're really far away from (successfully translating) a telephone call or a natural conversation. And we're nowhere near being able to understand three people in a meeting."
 
Dictionary.com is alright but it's got a lot of problems. You notice it when you try translating between English and another language you already know.
 
Translation is hard in that you also have to translate figures of speech. i.e... something as innocent as "Catching a cold." In Korean it is in fact expressed as "getting hooked by a cold."
In German you'd have "Wie gehen sie?" which would translate directly as "how goes you?"
And those are the easy bits. And consider this. In German your main verb comes at the END of the sentence which means this "live" translator couldn't possibly translate it well until the sentence was finished.
"Ich soll zu die Markt gehen." = "I should to the market go." But what if it was "to run?" "Ich soll zu die Markt laufen." And what if the translator assumed it was "to go" and not to "run"? And mind you this is just English to German which happens to be a very closely related language.
So in other words, you can't really do a "LIVE" without knowing firsthand what the guy's going to say.
Personally I don't know HOW the guys at the UN and stuff do it because they seem to be doing it live. Maybe there's a certain way to talk that will not confuse the translators.
Again, I say the issue is, the key word that needs translation does not appear when it needs to be said in English, which makes a real live translation REAL tricky.
What you COULD have is a center full of dudes who speak the language and basically whenever a unit on patrol is in need of a translator, they call up that unit and have a translator assigned to them until the conversation ends. That's current technology and a lot of tired bilingual people though.
 
Back
Top