martes, 5 de abril de 2011
Military’s Newest Recruit: C-3P0
The Pentagon has spent decades and gazillions of dollars trying to build the perfect translation device. Now, its far-out research arm is looking at a new direction: a robot that can interpret all sorts of languages — and think for itself. That’s right: The Defense Department wants to build a real-life version of C-3P0.
Thursday, Darpa announced its new Broad Operational Language Translation, or BOLT, research initiative — the latest in a long, long line of military interpretation gadgets and algorithms. The United States tends to fight its wars in places where it doesn’t really speak the language. Training up troops in critical languages like Arabic would be difficult, time-consuming and not entirely practical on a large scale.
Enter BOLT, which Darpa has asked Congress to fund at $15 million this year. Once developed, BOLT would act something like C-3P0 from the Star Wars movies, performing a variety of difficult translation feats for troops in hostile territory.
It would go well beyond the array of handheld phrase-translation machines currently in use. BOLT would use language as well as visual and tactile inputs so that it can “hypothesize and perform automated reasoning in the acquired language.” The end result, Darpa’s announcement says, will be a robot with visual and tactile sensors that can recognize 250 different objects “and understand the consequences (pre-state and post-state) of 100 actions, so that it can execute complex commands.”
The bot should be able to conduct both human-to-machine and human-to-human translation. On the human-to-machine end, Darpa wants BOLT to be able to understand human speech in English and one Arabic dialect, such that it can take “complex commands to control a desk-top application” like e-mail or Microsoft Excel.
For person-to-person translation, BOLT is intended to enable ”multiturn, bilingual human-human conversation” between English and Arabic with a success rate of 90 percent. The translation would be ”genre-independent,” meaning translation of language (either Mandarin or an Arabic dialect) regardless of whether it’s in a text message or just plain conversation.
The U.S. military has tried out all sorts of translation gizmos on the battlefield. But devices like the Phraselator and the Voice Response Translator are limited on the battlefield. They can’t translate just any words you’d like to say, though. Instead, they spit out a few key phrases and words in local languages likely to be useful on the battlefield. The blunt phrase exchanges can’t produce the kind of complex communication that the Defense Department would like soldiers to be able to engage in. They can also be downright awkward sometimes.
That’s why Darpa’s currently putting money into more sophisticated devices like BOLT and another Threepio-like translator. The agency asked Congress to fund its Robotic Automatic Translation of Speech (RATS) program increased to $21 billion this year, up from $9 million in 2010. RATS is supposed to be able to pull speech out of “noisy or degraded signals”and identify the language spoken. It’s also intended to sniff out not just the language spoken, but the person behind it by using voice recognition technology to check the person against a most-wanted list.
Whatever comes of Darpa’s attempt to turn BOLT into a military C-3P0, let’s just hope they don’t give it the continuously-piqued accent of Anthony Daniels, the dude who played the protocol droid in the movies. It’s annoying enough in English, it might be worse to hear it in Arabic.