One the basic problems facing Machine Translation (MT) is getting the language into the MT system. With text, this is rarely an issue, since so much of what we produce is electronic, and scanning can take care of the rest, albeit slowly and with errors. But for interpretation, voice input and output are essential, and here a company called SRI has made progress.
During a recent demonstration of a system SRI called IraqComm, technology took the place of a human interpreter in a simple exchange between an English speaker and an Iraqi Arabic speaker. The system was necessarily limited in its vocabulary, translation capability, and voice recognition capacity, but it worked. And for soldiers in the field, it could be invaluable.
"One of the crying needs in Iraq is overcoming the language barrier," said Kristin Precoda, director the SRI lab that developed the system. The same phrase could be applied to virtually any language pair in the world, but especially Spanish and English in medical and legal settings in the United States, and the major languages in the War on Terror, particularly Arabic, Pashto, Davi, and Farsi. Any system that can fill in even small gaps is welcome.
IraqComm is hardly a universal translator, whether the Star Trek or Hitchhiker's Guide to the Galaxy version. Human translators are in no danger of losing work to such devices, at least not for a long while. At present, SRI and many other firms are focusing on MT technologies to fill a specific niche, thus avoiding the fundamental problems facing MT in general.
Such systems tend to be closed, with limited vocabularies and voice-recognition capacity. They are intended for a particular situation in which the social interaction, vocabulary, idioms, and terminology are relatively stable and predictable. For instance, a history and physical in an emergency room, a court hearing for a suspected thief or drug dealer, or a brief interview with a civilian in Baghdad or Kabul. The systems can and do work, and over the next five years we should expect to see considerable refinement in their accuracy and adaptability, along with reductions in cost.
But human translators and interpreters are still essential. First, they have the depth and breadth of experience to handle a vast range of content. Second, they can do research when they don't know the words or phrases they encounter, and can ask questions of experts to get clarification. Third, they can learn and adapt to new uses of language, including slang, neologisms, dialect, and deliberate obfuscation by a speaker. Finally, interpreters can pick up on all the non-verbal linguistic cues that people give as a part of the process of communication.
Voice-input is a workable technology. I use it to do some of my own work, in order to same wear and tear on my hands. Voice synthesis is improving rapidly, so much so that it is starting to sound almost natural, at least in limited contexts such as calling a bank or hotel reservation system and asking for basic information. Limited-scope MT systems similarly are improving, and will take their place in an ever expanding multilingual world.