Yuqing Gao is bilingual-and so is her computer. At IBM’s Watson Research Center in Yorktown Heights, NY, the computer scientist, role-playing a doctor, speaks Mandarin Chinese into a personal digital assistant. In a few seconds, a pleasant female voice emanating from the device asks, in English, “What are your symptoms?” Gao’s system, designed to help doctors communicate with patients, can be extended to other languages and situations. The ultimate goal, she says, is to develop “universal translation” software that gleans meaning from phrases in one language and conveys it in any other language, enabling people from different cultures to communicate.

Gao’s work is at the forefront of escalating efforts to use mathematical models and natural-language-processing techniques to make computerized translation more accurate and efficient, and more adaptable to new languages. Distinct from speech recognition and synthesis, the technology behind universal translation has matured in recent years, driven in part by global business and security needs. “Advances in automatic learning, computing power, and available data for translation are greater than we’ve seen in the history of computer science,” says Alex Waibel, associate director of Carnegie Mellon University’s Language Technologies Institute, which supports several parallel efforts in the field.