into the Response Buffer, where they are held temporarily in a state that allows for self-monitoring, including evaluation as to how well they correspond to the optimal output. On the basis of less-than-optimal output patterns, the child can alter weights in the neural network to improve subsequent performance. However, if these sub-symbolic output patterns are qualitatively different from the symbolic input patterns, it is not clear how the learner can evaluate their adequacy. Smith's discussion is interesting, but far too abbreviated. Perhaps the intent was to raise issues for discussion, rather than to provide any compelling answers, and if so, the discussion is successful.Overall, this book is a dense read but well worth it. The database of Zak's phonological development is interesting and detailed, and will undoubtedly be mined by others. The discussion of theory is interesting but terse ; terms are not defined, and ultimately it is difficult to decide whether Smith's arguments are compelling. Despite these weaknesses, the book has many strengths, and it is a worthwhile read. It will undoubtedly attract the same sort of attention as Smith's previous book has had over the past forty years.