Book Reviews: Computations from the English

all sequential or serial order. They all presuppose a rulebased system that can be enumerated in a "decidable" sense and that includes hierarchical enumeration during rule-application. Even the discussion of the computational approach presented by Schank and Birnbaum includes nuances of this. There are relevant discussions of how one can acquire language and how it might map the real world into a "usable" representation that permits variability. However, these are presented at a theoretical level. There is not sufficient material included to "make the theory computational". It is very interesting to note that the processor, meaning the human brain, can constrain how the real world projects into a subjective perceptual world, and to provide discussion of same, but the issues of how such a processor might control such a happening are avoided. In contrast, for artificial intelligence, determining the process and the control presents the greatest problem. The psychological perspective provides a brief historical introduction to the issues considered important by psychologists with respect to the language problem. Many of the concerns regarding the function of representations, and the relationship between theoretical and applied work and psycholinguistics are similar to those expressed by the computational discussions. However, the stress on evaluation of theories which is of utmost importance to psychological approaches does not appear of such concern in AI. One area which the psychological discussion merely touches has to do with CONTROL of the processes. While control is mentioned, no suggestions as to how it should become a viable part of language process or theory from a psychological perspective are presented. Instead, all of the theories presuppose a sequential process, at least as they are presented in the included papers. No CONTROL alternatives are discussed, as if the problem is already solved. One other psychological paper discusses whether there are language-specific tasks, i.e., tasks that cannot be "learned" or performed unless one has language. While the issues raised are interesting, the presentation and discussion of the evidence is far from convincing. Perhaps it is available elsewhere, but certainly is just touched upon here. Finally, the discussion within the computational approaches to language is not anything that has not been presented many times before. Basically, two viewpoints are presented, one the semantically driven approach to language processing and the other which demonstrates that there are critical aspects of language understanding that can only be syntactically determined. The arguments presented in Marcus's paper advocating a separate role for syntactic processing were the most clearly described in the book. If anyone is seeking example inputs which require syntactic decision processes that cannot be handled entirely within semantic processing, or canriot even be recognized as a problem if one is doing Computations from the English