NIST Evaluates Speech-Translation Devices for DARPA

Article Details

Global Watchtower

Common Sense Advisory Blogs

Posted by Donald A. DePalma on July 31, 2007 in the following blogs: Translation and Localization, Technology, Market Data
Pages \| \|

Whether you're conversing or transacting internationally, language frequently gets in the way of successful interactions. Translation, human or machine, is the best solution to that problem. But if there's no monitor or keyboard at hand, you need an interpreter by your side or on the other end of a phone line -- or a device that recognizes speech and does the translation for you. This is frequently the case for soldiers in places like Iraq, police, first-responders like firemen and emergency medical technicians, and travelers to places where they don't speak the language. Over the last few weeks there's been some activity in this product category most often associated with science-fiction communicators:

Etaco announced its iTRAVL NTL-9C, a handheld translator that includes speech recognition, speech analysis and voice output technology, and two-way full-text translation for English to Chinese, French, German, Italian, Polish, Portuguese, Russian, and Spanish. Just in case its dictionary of idioms produces an unpleasant result, it also includes a database of medical and legal terms.

The BBC reported that Iraqi-born Amin Ismail developed a wristwatch-based (but not wristwatch-sized) translation device called Ahky (Arabic for "speak"). The first version will be English-Arabic, but future versions could manage other language pairs. Like IBM's somewhat larger Mastor, the device is meant to help soldiers in the field overcome language barriers. Ismail signed on with Civil Defence Supply, a UK military equipment manufacturer, that hopes to begin production by the end of 2007.

On the less life-threatening side of the speech recognition equation, SpeakLike will soon enter a beta test of its speech-plus-translation phone service that utilizes the internet, VoIP, and chat for long-distance and in-room cross-language communication. The system supports English<>Spanish today, but CEO Sanford Cohen tells us that the company will add more languages after the test period. He plans to offer SpeakLike's speech recognition and automated translation technology as a service to companies wanting to support customers in other languages.

How good is the technology that these suppliers offer? While the infamous Vista speech recognition video made light of Microsoft's for-free offering, there's a lot of serious research and product development being conducted by mainstream companies like IBM, NEC, Nuance, and Philips -- and being leveraged by companies like SpeakLike. We always look for formal, systematic testing of such technologies to see what the experts think. In July NIST conducted a series of lab and field tests of prototype handheld two-way translators for U.S. Marines and Iraqi Arabic speakers. According to the press coverage, NIST had the Marines and Iraqis act out scenarios like traffic checkpoints and neighborhood surveys.

The testing methodology relied on verbal communication in which the translation lab rats had to look at each other during the interrogation. Neither side could look at the screen to see the questions, and the Iraqis who could speak English were given earphones to block the sounds of English. NIST controlled background sounds to simulate realistic scenarios -- traffic, garage doors, passing traffic, and radio broadcasts, but reportedly no small arms fire or explosions. To simulate future smaller devices that they might use in their hands, the Marines carried the prototypes in their backpacks so their hands were free.

So which device won the shoot-out? Neither DARPA nor NIST would say (maybe it's about executive privilege?). The agency said that it will use the data it collected as part of its continuing research into how TRANSTAC systems will improve over time, to determine which technologies show the best chance of success, and to accelerate the development of language pairs once the translation devices have been perfected.

High-quality speech recognition, with and without automated translation, is one of those technologies that everyone wants and needs to happen. We're hoping that military budgets drive and accelerate product innovation in the private sector, too. After all, we taxpayers are footing the bill for this technology (and the war that required it).

Categories