Machines can translate, within limits

Defense Advanced Research Projects Agency

"To have or perhaps I have not."

Hamlet? Sort of. It's actually a computer-generated Japanese-to-English translation of the opening line of the moody prince's famous soliloquy, "To be or not to be."

Columbia University professor Merit Janow, addressing a Justice Department advisory panel several years ago, used that imperfect interpretation to demonstrate the promise and the pitfalls of automated translation.

"If you think machine translation will replace all your human translators, you are going to be disappointed. However, if you use [it] in places where it is very strong, you'll save hundreds of man hours," said Joel Ross, a technical innovations officer with Advanced Innovations Technology Inc. Ross is assigned to the Foreign Broadcast Information Service, which monitors international media and government publications and broadcasts for the CIA.

Interest in automated translation is widespread. In addition to intelligence gathering, Justice Department investigators use it in antitrust cases involving international companies, and the Defense Department is developing military- specific translation systems to improve communications with allies.

Automated translation technology is useful in technical-language applications, which is why it is more widely used in the commercial sector than in government. Xerox Corp., for example, has been using the technology for years to create technical documents for international markets.

Translation technology's other strength—one of more interest to government agencies—is in making rough translations of foreign- language documents.

According to a former Justice Department official who spoke on condition of anonymity, Justice's antitrust division relied on the technology in a case involving a Japanese electronics firm. The department had obtained thousands of Japanese documents but did not know which were relevant. And the cost of hiring translators to read through all of them was prohibitive.

Using automated translators, Justice made word-for-word translations of the documents. The English versions were then scanned electronically for certain key words or phrases to help investigators determine which documents were relevant. Only then were human translators brought in.

Randy Morris, chief executive officer of Translation Group Ltd., Haddonfield, N.J., calls that a "triage process" to let people know "what's important and what's not important."

Morris' company markets the Translator, a Web-based system with a customizable translation memory.

He said automated translation is used by intelligence agencies, but also by agencies such as the Commerce Department and the State Department—"anybody who has to keep tabs on what's going on around the world."

In Kosovo, some U.S. Army personnel carried automated translators in small suitcases, Ross said.

"Let's say they find a paper in Serbian. In the suitcase, there's a scanner. They run it through the scanner, which converts the image into digital text," he said. "They take that text and run it through the Serbo-Croation-to-English translator. It all happens in minutes."

Through the Defense Advanced Research Projects Agency, the Pentagon is developing a Web-based system for collecting and presenting Korean documents in English, said Clifford Weinstein, head of the information systems technology group at MIT's Lincoln Laboratory, Lexington, Mass., where the work is being done.

The system will automatically translate general types of documents, such as newspaper articles, he said. The laboratory is also working on a Web-based program that will locate documents or articles and translate them.

As in the private sector, the technical language of the military lends itself to automated translation. "I think "Hamlet' would be pretty challenging," Weinstein said. "We haven't addressed poetry yet."