Name Matching between Roman and Chinese Scripts: Machine Complements Human

There are generally many ways to transliterate a name from one language script into another. The resulting ambiguity can make it very difficult to “untransliterate” a name by reverse engineering the process. In this paper, we present a highly successful cross-script name matching system that we developed by combining the creativity of human intuition with the power of machine learning. Our system determines whether a name in Roman script and a name in Chinese script match each other with an F-score of 96%. In addition, for name pairs that satisfy a computational test, the F-score is 98%.