A team of researchers at the University of Southern California has designed a computer program that takes a novel approach to language translation. While most translation software uses hand-coded rules for recognizing words and phrases, the new program learns on its own—by taking a set of basic translation rules and expanding on them through statistical analysis.



