The Pentagon's Defense Advanced Research Projects Agency (DARPA) has launched a large-scale project that aims to create translation software capable of translating TV broadcasts, phone calls and other forms of speech and text in real-time, and summarize them in English for human intelligence analysis.
The system is called GALE, for Global Autonomous Language Exploitation. It is being backed by some top research firms including the research arm of IBM. And the effort is also being staffed with the top talents from the academic world, including renown researchers from Cambridge University, Carnegie Mellon, John Hopkins, Stanford and top European and Asian universities.
GALE's goal is to deliver, by 2010, software that can almost instantly translate Arabic and Mandarin Chinese with 90 to 95 percent accuracy. This is a very difficult task; the current state of the art for written text (such as newspapers articles) is about 60 percent accuracy. And DARPA's project is far more ambitious than translation of just written text. In translation of speech, for example, project researchers will need to contend with tasks which are considered to be highly daunting to speech-recognition and machine- translation engineers. This includes issues such as background noise, dialects, accents, slang, short words like "on" or "of" that most speakers don't bother to clearly enunciate. Other forms of text translation which are also being targeted by GALE, such as translation of newspaper talkbacks and Internet chats, also hold many technical challenges.
10:04:55 AM 
|
|
