PAUL THAGARD Coherence in Thought and Action Cambridge, mA: Massachusetts Institution of Technology Press, 2000, 328 pages (ISBN 0-26220131-3, us$32.95, Hardcover) Reviewed by WILLIAM E. SMYTHE This volume is a comprehensive presentation of the latest work of Paul Thagard and his research group on coherence theory, an area of inquiry to which the author has made ground-breaking contributions for over a decade. Coherence theory is a peculiar synthesis of philosophy and cognitive science that approaches problems in terms of the satisfaction of multiple constraints within networks of highly interconnected elements. The basic ideas of coherence theory would seem to have broad appeal across a number of areas of contemporary thought. They clearly resonate with the meaning holism of Quine's two dogmas of empiricism in philosophy, with notions of distributed cognition and constraint satisfaction models of decision-making in human cognition, with the current popularity of neural network and connectionist models in the cognitive and neurosciences, and, more generally, with the growing interest in nonlinear dynamics and related phenomena across a wide range of disciplines. The main aim of the present work is to extend the reach of coherence theory beyond its usual applications in cognition and epistemology to questions of ontology, ethics, politics, emotion, social consensus, and probabilistic reasoning. Thagard makes a persuasive if not wholly convincing argument that all of these domains can be thought about in roughly similar ways. At the heart of Thagard's main argument is a surprisingly simple conceptual structure. One begins with a set of elements - propositional and other types of representations - that are linked via numerically weighted relations of coherence (positive constraints) and incoherence (negative constraints). The goal of coherence-based reasoning is to partition the elements into two disjoint sets - accepted and rejected elements - in such a way that overall coherence within each set is maximized. Connectionist algorithms can be applied to compute coherence in an iterative fashion for any completely specified network of elements. The underlying simplicity of this explanatory framework is at first striking, given the wide range of problems to which Thagard subsequently applies it. This initial impression of simplicity quickly evaporates, however, as one reads beyond the introductory chapters. Thagard's basic conceptual structure requires substantial embellishment to deal with various specific issues and areas of application. Thus, we learn that epistemological issues require one to distinguish among five distinct types of coherence: explanatory, analogical, deductive, perceptual, and conceptual. Ethical reasoning adds considerations of deliberative coherence. Understanding emotion requires supplementing relations of cognitive coherence with valence relations and introduces considerations of metacoherence. Consensus formation is based, not just on coherence, but on the vicissitudes of communication and information exchange. Then there is the matter of probabilistic reasoning which, as Thagard shows, can be accommodated within coherence networks only with considerable difficulty. Despite the complexity of the issues it treats, this is not at all a difficult book to read. Thagard has the knack for explaining difficult ideas in readily understandable language. However, his frequent crossing of disciplinary boundaries makes it a challenging task at times to evaluate his theory. Is it primarily a theory of human cognition, of philosophical epistemology, or is it mainly intended as a new procedure for addressing philosophical questions? It is perhaps all of these at once. This kind of blending of psychological and philosophical issues is a familiar feature of the work of contemporary philosophers writing about cognitive science. It is also somewhat reminiscent of the style of thought of many of the great philosophers of the past, who presented their epistemologies as part and parcel of their theories of mind. …