AFTER THE REVOLUTIONS OF GESTALT AND ACTION CONTROL THEORIES: IS THERE A PSYCHOPHYSICS OF TASK-RELATED STRUCTURAL REPRESENTATIONS? A DISCUSSION BASED UPON EVIDENCE FROM VISUAL CATEGORIZATION

A reanalysis of claims of Gestalt psychology reveals that the actual core of its theoretical program is a language-like representational system for perception. The effectiveness of this system resides in its generative power based upon recursive code construction and upon the property that perception expands in time and structure beyond momentary stimulus dependency into memory. Representational issues arising from this view are complex analogues of by now classical psychophysical representation problems. In an experimental section, problems of the joint action of object-related and intentional, demand-related constraints in memory are discussed. Evidence from tasks of structural recognition indicates that demand-related constraints define the relevant sets of traces within which object-related coding proceeds. To portray dynamic aspects of this double determination, findings from experiments are reviewed in which systematic distortions were applied to transformationally generated material. Results suggest the existence of powerful operations of structural extrapolation in perceptual memory. For psychophysicists it is a commonplace that judgmental performance simultaneously depends on quantitative and qualitative stimulus aspects, and on task demands. With regard to stimulus determinants, already Fechner (1860) was considering, beside intensity and modality, compositional factors such as spatial and temporal stimulus contexts. As for taskrelated factors, the influence of scale instructions on judgment has, among others, relatively early been taken into consideration (see Stevens, 1975). Scaling, discrimination, detection and recognition situations (cf. Luce, Bush, & Galanter, 1963) have by now become classical examples of different specific modes of processing. In the course of time, the understanding of the two-sided determination has considerably expanded. On the stimulus side it now includes sensory context effects, situational distinctions such as the distinction between lightness of surface colors and brightness of illumination and, last but not least, it also embraces configural dimensions such as figural goodness. The same holds for intentional task demands. Among the expansions taken into consideration, tasks of cognitive information integration appear of particular interest (cf. Anderson, 1998), since their analysis includes combinatorial inner, "psychological", rules of memory-based valuation and merging. Finally, considerable progress has been made in developing process models of task-dependent psychophysical decision making (cf. Link, 1992). From this brief summary it is only fair to say that psychophysics by studying stimulusand task-related determinants of judgment made a great step forward from a psychophysicallaw-concentrated discipline to an "exact science of body-mind relations" as Fechner envisaged it. Signs of progress should, however, not blind us for weaknesses and deficiencies. Among the apparent shortcomings, one major point is that stimulus-related and intentionally enforced organizational factors have rarely been studied in their joint action. Also, no theoretical view has emerged that would have allowed us to understand their interplay on the basis of general principles and mechanisms of cognition. These are, of course, no failures of psychophysics as such. Modern psychology and cognitive science as a whole have failed to develop an integrative framework. In fact, since the first decades of the last century and the conceptual revolution of early Gestalt psychology followed by the action-centered Piagetian approach and its Russian derivatives, issues of "autonomous" perceptual organization and of perceptual-cognitive organization as servicing action planning and control have been dealt with quite apart from each other or have even been considered as deriving from mutually excluding conceptions. In this contribution, I am going to suggest that a reconciliation is not only possible, but is imperative for answering fundamental open questions of cognitive theory. I will approach this goal from two sides: First, from that of a sketchy outline of theoretical problems of Gestalt psychology reconsidering the state of affairs, and, second, by way of a reanalysis of fairly simple and clear-cut experimental findings in visual categorization. The hypothesis on which both discussions converge is the existence of a common level of memory representation on which structured traces of object information as well as equivalents of intentional constraints project. The Gestalt concept: Basic claims revisited A discussion reconsidering stimulus-based organization in perception and cognition cannot do without a brief glimpse at Gestalt psychology: There is no doubt that the conceptual innovations invented by the Gestalt movement have left lasting traces in psychological thinking and that experimental inquiry initiated by it has led to important discoveries (see Klix, 2001). For the relevance of its topics today it is certainly indicative that themes of Gestalt perception have become one of the most favored subjects of modern neurosciences and of related psychological research (see Müller et al., 2001). Unfortunately, it is also true that a restatement of Gestalt concepts cannot do without noting the misfortune of Gestalt psychology in that it often was misunderstood in its basic claims. One of these misunderstandings roots in popular slogans through which it used to be propagated. Thus it became famous for its "Laws of Gestalt" in connection with the notion of Gestalts as wholes "that are more than their parts." Due to this definition, a list of stimulus factors influencing the perceived grouping of objects was put into the center whereas the concept of Gestalt itself was reduced to the non-additive character of wholes appearing as a somewhat mysterious and ultimately superfluous appendix. This, however, turns truth upside-down. To be sure, Gestalt psychology, like any other modern approach toward perception, understood the perceptual outcomes as reflections of real-world situations in terms of objects, their states and interrelations (Wertheimer, 1923). The specificity of the Gestalt claims, however, consists in the form of statements about properties of these reflections which are being considered apart from their correspondences in the real world, not even requiring that such correspondences necessarily exist. Relative to inner representations of structured entities, this is a turn similar to that invoked by the introduction of internal metrics in psychophysics. In this property, constructs of Gestalt not only account for contents of perception for which there are unique correspondences in reality, but potentially also for purely internally generated phenomena such as cognitive contours or even hallucinations. Specifically, Gestalts have been characterized by an expansion of what von Ehrenfels (1890) had called Gestalt qualities. There is nothing mysterious about these "qualities." The claim is that for statements on percepts global relations like "belonging to the same unit", "being part of", or "being caused by" do have the same "real" status of perceptual attributes as any other "classical" attributes, say, seen color or lightness of a sheet of paper or the perceived inclination of a bar. This notion of Gestalt qualities is not restricted to universal relations of this type. More specific perceptual relations such as those between points seen as forming circles or straight lines represent Gestalt qualities as well. Perhaps, the greatest step done in further explicating the Gestalt view by elaborating upon von Ehrenfels' concept is certainly that Gestalt qualities have come to be understood as in principle open to recursive combination and nesting. Departing from simple demonstrations such as the hierarchy in a percept like "a seen triangle of squares of circles", this innovation goes in its consequences far beyond a descriptive list of percepts: In contrast to even modern concepts of structure drawing upon ecological constraints, this view conceptually furnishes perception with the potential "supercapacity" of a language which not only is capable of encoding objects or events, never seen or heard of before, but also gives perception a guiding role in the construction of artifacts which nature is incapable of inventing. The "true" laws of Gestalt psychology are thus rules of an assumed quasi-language. It is this language-like structure to which the invention of discrete constituents is crucial and which necessitates and enforces their introduction, and not the phenomenal appearance of percepts as it may have been intimated by some representatives of Gestalt schools. Similar to the introduction of metrical relations in a multi-dimensional manifold representing subjective space, the generative representation of structured objects requires definition of proximities. Such a language-like system with its capacity for generating a practically unlimited number of possible structural expressions requires, in addition, tools for comparative valuation of equivalent expressions. To account for the latter, Gestalt psychology introduced a valuation dimension, in German termed "praegnanz", for which English equivalents are goodness, self-consistency or, most to the point, degree of singularity (Goldmeier, 1982). To put it somewhat more generally, the core of Gestalt psychology is the scientific goal of constructing such a system in conformity with known facts on perception of possible objects, their states, mutual relations, changes and their transformations into one another. If this is an adequate idea about perception, there follow important consequences for perceptual processing. Analogous to the absence of a "principle of induction" for reasoning according to modern epistemology (cf. Popper, 1989), i.e. of a general rule for attaining a solution to every empirical problem by induction, there should be no feed-f