Variations and Selections:

An Anthology of Reviews from the Journal of the Experimental Analysis of Behavior

At the outset, let us note that cognitive phenomenaoften are not clearly differentiated from cognitivist interpretationsof phenomena. Typically, cognitive phenomena are identified with traditional psychological terms (thinking, deciding, remembering, and the like), whereas cognitivist interpretation views such phenomena as mediating processes (e.g., computational or representational ones) underlying behavior. These inferred processes are often discussed mentalistically (although usually without implying mind-body dualism). But activities like thinking and remembering can be taken as behavior to be accounted for rather than as processes with special status. Thus, behavior analysis, which deals with these activities as behavior even if they seem peculiarly internal, offers a nonmediational and thus noncognitivist interpretation of cognitive phenomena. Tension between behavior analytic and cognitivist approaches may be inevitable given the mediational assumptions inherent in the latter. The essays here reflect that tension.

Neisser's Cognitive Psychologywas influential in establishing the cognitivist viewpoint. Salzinger's review sketches cognitivist objections to a behaviorist approach, identifying not only the usual points (e.g., behaviorism as ignoring cognitive phenomena) but also a core disagreement hinging on whether events not contiguous with their effects can be allowed as causes. In detailing differences between cognitivist and behavior analytic assumptions, Salzinger shows how behavior analysis can address phenomena cognitivists have claimed as uniquely their own.

Zuriff's review of Stich's book, From Folk Psychology to Cognitive Science,shows that, though Stich is a cognitivist philosopher, his arguments support long-standing behavior analytic cautions about accepting ordinary-language mentalisms at face value. In Stich's view, folk psychology, while easy to sell and even difficult to avoid, is ultimately an untenable method for accounting coherently for human behavior. The routine acceptance of ordinary psychological terms within cognitivist theory, then, generates "serious but usually unacknowledged problems at the v ery core of cognitivism."

Two pairs of essays, by Shimp and Branch, and by Wasserman and Malone, illustrate how behavior analysis can address cognitive phenomena. Shimp observes that behavior analysts have largely ignored the bread-and-butter phenomena of cognitivist psychology (in particular, those identified with memory). Behavior analytic human research has increased since Shimp's essay, but little of it has been organized around cognitive labels; thus, with a few notable exceptions (e.g., Palmer, 1991) the observation remains valid. The case Shimp makes for more emphasis on behavior organization (and thus traditionally structuralist issues) also argues for traditionally mentalistic terms such as memory. While accepting some of Shimp's observations, Branch asserts they can be adequately addressed without importing cognitivist terms. A key point in the exchange is the term "theoretical." For Shimp, any event not "directly observable" qualifies, and by this argument even rate of responding is a theoretical entity. Branch accepts the point that behavior analytic terms often extend beyond the moment of direct observation, but he distinguishes them from the hypothetical constructs of mediational theory.

Wasserman's review of Hulse, Fowler & Honig's Cognitive Processes in Animal Behaviorseeks to minimize differences between mediational and nonmediational explanations. He surveys the origins and evolution of comparative psychology, where cognitivist concepts have been embraced as bases for cross-species comparisons. While acknowledging differences between hypothetical constructs and intervening variables, he is comfortable with both, provided thorough "objective inference" rather than the "subjective inference" of reasoning by analogy relates terms to experiments and data. He questions the varied mediational terms invoked in comparative psychology (e.g., representation, rehearsal, cognitive map) but doesn't reject them outright. Malone takes issue with Wasserman, casting the issues in contemporary more than historical context. He distinguishes between S-R associationism and cognitivist theorizing, calling attention to problems lurking in the knowledge/performance distinction. Notable is his sketch of an experiment to enable subjective inference with respect to pigeon behavior, thereby confronting Wasserman's arguments for acceptability of objective versus subjective inference.

Wixted examines how cognitivist terms are used, in ordinary language and in Kendrick, Rilling & Denny's Theories of Animal Memory. He critically reviews justifications for terms such as imagery, expectancy and rehearsal, and finds wide variations in the degree to which such terms are related to functional analyses. Anticipating philosophical issues treated in Part IX, Wixted offers a reasoned rationale for examining the functions of cognitivist terms as they occur in natural language, and for accepting their relevance for behavioral research so long as their technical development does not contradict those natural contingencies.

Contrasting viewpoints have often been understood in terms of differing domains of emphasis. Behavior analytic versus cognitivist differences have often fit this pattern, each having favored types of procedures and data. Stolarz-Fantino & Fantino review Rachlin's attempt in his Judgment, Decision and Choiceto portray cognitivist and behavior analytic efforts as complementary by emphasizing domains in which each addresses similar facts. Instead of dwelling upon what each approach counts as adequate explanation, they build upon Rachlin's assemblage of data from behavioral economics and cognitivist decision theory to propose experimental strategies relevant to interpretation from both traditions.

Dreyfus' history of the accomplishments and limitations of the artificial intelligence movement in his What Computers Can't Do: The Limits of Artificial Intelligenceagain allows comparison of the cognitivist enterprise and behavior analysis. Crossman's review reveals that the distinction between contingency-shaped and rule-governed behavior is implicit, but other distinctions also make special behavior analytic sense: e.g., viewing organisms either as biological entities or as mechanisms, and viewing concepts either as fuzzy classes defined by contingencies or as formal categories (cf. Palmer & Donahoe, 1992).

In reviewing Parallel Distributed Processing (PDP),Donahoe & Palmer show that the PDP approach shares many assumptions with behavior analysis: e.g., selection extended over time, a consequent emphasis on historical explanation, functionally produced fuzzy sets rather than formally defined concepts, multiple causation, and awareness as a derivative form of knowing rather than a fundamental process. These shared assumptions place both approaches in opposition to core concepts of mainstream cognitivist psychology. The reviewers note that the postulated characteristics of PDP networks have not been closely patterned upon neurological facts, so that the networks remain metaphorical neuroscience, and suggest how the replacement of back-propagation with a diffuse projection principle would make the selection of behavioral outputs more closely resemble reinforcement in both molecular process and molar function (cf. Stein, Xue & Belluzzi, 1993).

Finally, Saunders and Green's essay carefully distinguishes between formal analyses that define the mathematical concept of equivalence and behavioral analyses identified with the same term. By precisely delineating structural and functional analyses, they illustrate a constructive basis for characterizing differences between behavior analysis and other approaches, and they show how behavior analytic methods and concepts can be applied to logic, inference and conditional choice, all of which are usually seen as cognitive in nature.