Prosody is an integral part of human communication, but one that second language (L2) learners rarely learn. Topic shifts, contrastive focus, and even simple question/statement distinctions, cannot be recognized or produced in many languages without an understanding of their prosody. However, 'translating' between the prosody of one language and that of another is a little-studied phenomenon. This research addresses the 'prosody translation' problem for Mandarin Chinese and English L2 learners by identifying correspondences between prosodic phenomena in each language that convey similar meanings. The work is based on comparisons of L1 and L2 prosodic phenomena and the meanings they convey. Computational models of prosodic variation suitable for representing these phenomena in each language are constructed from data collected in the laboratory, with results tested on L1 and L2 subjects. The models are tested in an interactive tutoring system which takes an adaptive, self-paced approach to prosody tutoring. This system modifies training and testing examples automatically by imcremental enhancement of distinctive prosodic features in response to student performance. The success of the system is evaluated via longitudinal studies of L2 students of both languages to see whether the new techniques improve students' ability to recognize and produce L2 prosodic variation. By providing a method and computational support for prosody tutoring, this work will not only enable students to attain more native-like fluency but it will provide a model for training students in other pragmatic language phenomena --- beyond learning the words and the syntax of a new language.
