A goal of machine learning is to use natural language to process information, and to do so with a minimum of human involvement to get as much information as possible and necessary into a condition to be useful in computation. Unlike supervised learning, in which data is labeled, semisupervised learning turns unlabeled data into useful information for the computer to process. Abney (U. of Michigan-Ann Arbor) gives students and researchers a strong background in semisupervised learning (as well as supervised and unsupervised learning), including theory and linguistic applications. He covers such concepts as self-training and co-training in machine learning and goes immediately to their applications and relevant classification, then describes boundary-oriented methods and their the mathematics. He describes clustering, generative models, agreement constraints, propagation methods, and spectral methods and the their mathematics. Annotation ©2008 Book News, Inc., Portland, OR (booknews.com)