Language Weaver Expands Automated Translation Roster

Language Weaver yesterday released new versions of its bidirectional machine translation software that pairs Arabic with Spanish, Arabic with French, French with Spanish, Danish with English, and Greek with English.

These newest offerings bring to 47 the number of bidirectional language pairs that Language Weaver offers.  A total of 21 languages are offered in those pairings, and at least two more are in the works.

By developing and expanding its bidirectional pairings, Los Angeles-based Language Weaver is changing the paradigm for automated translation. In typical machine translation applications, English is used as a pivot language, meaning that for an Arabic-to-French translation, for example, a company might first translate the information from Arabic to English, and then from English to French. This two-step process slows the translation down considerably, but is the most common.

"Any system can go directly, but traditionally it doesn’t because there is a lot more data available for translating between English and other languages," explains Kirti Vashee, vice president of sales and marketing at Language Weaver.

By contrast, Language Weaver’s approach is to use statistical modeling to translate directly from one language to the other without using English as the bridge. The key is to provide the translation engine with background data surrounding things like inflection, morphology, syntax, grammar, sentence structure, and word order, Vashee says.

For that reason, "first-generation [bidirectional translation] technology tended to work best with Romance languages," he continues. For Language Weaver, succeeding generations have incorporated other European, Middle Eastern, Asian, and several African languages as well. The company is working hard to improve accuracy in all its language pairings.

"None of the technology is 100 percent, because there is just not enough foundational information, but we’re getting there," Vashee says.

Beyond expanding its language pairings, Language Weaver is also working to widen the uses for bidirectional translation software. Though used mostly now for text-to-text translations, the technology is gaining traction in speech-to-speech, text-to-speech, and speech-to-text applications.

According to Vashee, one of the widest uses for the technology will be in real-time monitoring of international TV and radio broadcasts, but it could theoretically be used in call center operations that rely on interactive voice response (IVR) systems. In such operations, the caller could conceivably speak in his native language, and the system would receive a translated version of the utterance; such translations could even show up as a screen pop on a live agent’s computer. The agent could speak or type in her native language, and the called would hear her questions in his native language.

"These are all available in prototype today," Vashee notes. "The challenge is to get them integrated into systems and get the costs down."

Translation systems for IVRs now cost about $10,000, but Vashee anticipates that the costs will come down to a few hundred dollars in the next year or two, especially as companies move the technology from an embedded software mode to a networked server environment.

Companies like Language Weaver are also building out their partner networks to expand the uses for the technology. Language Weaver, for example, has extensive relationships with BBN, SRI, Virage, and others, and is engaged in a number of pilots with the U.S. military and government in the Middle East.

"The world in general is becoming more global, so the need for [this type of technology] is increasing," Vashee concludes.