Novauris has built in some cultural sensitivity to its system. For example, a blunt phrase stated in one language may be translated into a more polite expression in a language that values politeness, Kim said.
Any current smartphone has enough processing power and memory to run the software, which has been written in versions for Windows Mobile, iPhone and mobile Linux and will soon be adapted to Android. A new language could be added in just a few seconds, Kim said. Novauris is talking with partners, and Kim believes a product may be commercially available next year.
Fluential, which has developed speech translation products with funding from the U.S. Defense Advanced Research Projects Agency (DARPA) and other government sources, showed off a system that uses remote processing to translate a wider range of conversations. Its software can be delivered as a service over a cloud infrastructure, deployed on a workstation in an enterprise's own data center, or by other methods, said President and CEO Farzad Ehsani. The company is working on a smartphone prototype that would work over a 3G network.
"This is not a universal translator ... It handles 80 to 90 per cent of common interactions for a given setting," Ehsani said.
Fluential's software includes both template translation, which handles standard phrases, and statistical translation, which is designed to interpret more open-ended speech. The company has tested it at a hospital in San Francisco, where real non-English-speaking patients used it to describe their ailments to medical professionals, Ehsani said. Nurses were trained for about 90 minutes on how to use it, and patients received about 40 seconds of simple instructions. The system achieved overall translation accuracy of 92 per cent, he said.
In a hospital setting, these types of conversations are typically staffed by a human interpreter, at an overall cost of between US$0.70 and $2.00 per minute, Ehsani said. With Fluential's system, it would cost about one-tenth or one-twentieth of that, he said.
The company is now preparing to bring to market a first implementation of its product, for conversations between nurses and patients, and Ehsani believes it could be on sale in six to nine months. Versions for other settings, medical and otherwise, are also in the works.
Sign up for Computerworld eNewsletters.