| 1. |
General Introduction |
we summarize the key ideas of Gestalt theory (Gestalts as wholes being different from the sum of the parts, emergence, self-organization, the law of Prägnanz) and define the challenge of specifying these notions in ways that are satisfactory in light of modern science |
| 2. |
Holism |
|
| 2.1. |
Holism in Traditional Gestalt Psychology |
we briefly review holism as the “fundamental formula” of traditional Gestalt psychology |
| 2.2. |
Modern Approaches to Holism |
we clarify core Gestalt notions such as holistic properties, emergent features, configural superiority, and global precedence, by providing operational definitions that fit into a more contemporary information-processing framework |
| 2.2.1. |
Garner’s dimensional integrality |
we discuss Garner interference and separable versus integral dimensions |
| 2.2.2. |
Emergent features and configural superiority |
we discuss emergent features (EF), configural superiority effect (CSE), and the Theory of Basic Gestalts |
| 2.2.3. |
Global precedence |
we discuss Navon’s global precedence hypothesis, hierarchical patterns, global-local paradigm |
| 2.2.4. |
The primacy of holistic properties |
we discuss holistic properties, configural properties, emergent properties, and global properties |
| 2.3. |
Interim Evaluation: New Foundations Needed |
the conceptual clarifications and operational definitions of key Gestalt notions have been useful in making further empirical and theoretical progress (e.g., dissociating attention, perception, decision components) but we still 7need stronger theoretical frameworks to provide solid foundations to the Gestalt approach’s major principles
|
| 3. |
A Dynamical Systems Approach |
|
| 3.1. |
Introduction |
we discuss how recent developments in nonlinear dynamical systems theory can solve the tension between inert (equilibrium field forces) and active aspects of perception (spontaneous perceptual organization), between stability (attractors) and change (switching) |
| 3.2. |
Noise-Driven Models |
we discuss the double-well model (two equivalent attractors), internal noise, dwell times, the Gamma distribution, sequential dependencies, systems far from equilibrium, cycles of approach and avoidance, and the “visual sniff” |
| 3.3. |
Dynamical Models |
|
| 3.4. |
Dynamic Synchronization and Complex Adaptive Systems |
we review long-range dependencies between dwell times and discuss their implications for dynamic synchronization and complex adaptive systems we discuss switching governed by fragile attractors in complex adaptive systems consisting of coupled oscillators; self-organized criticality; alternating unstable and stable periods, corresponding to synchronization of high-frequency (gamma range) and low-frequency (beta range) oscillations (coherence intervals)
|
| 3.5. |
Conclusion |
different sources of change (stochastic and deterministic), different types of noise (microscopic and macroscopic), and different kinds of attractors and dynamics were considered some of these were shown to correlate well with known behavioral effects (e.g., dwell times) and recently discovered specific neural signatures (e.g., coherence intervals and self-organized criticality of synchronization of neural oscillations) in sum, a dynamical systems approach can explain how the brain’s capacities for self-organization are ideally suited to balance robustness and flexibility.
|
| 4. |
Principles of Measurement in a System of Sensors |
|
| 4.1. |
Introduction |
the original Gestalt notion that perception is mediated by properties of a global neural field was rejected by empirical evidence and was replaced by an atomistic trend in neurophysiology with a focus on single cells the original attempt to provide a foundation to the many Gestalt principles in terms of a single general principle (such as the simplicity principle or the law of Prägnanz) failed because of its vagueness the focus on simplicity as an autonomous principle intrinsic to the perceiving organism did not allow to capture the important perceptual role of regularities in the natural stimulation these limitations are resolved in a new framework, which combines elemental and systemic outlooks on perception, taking into account both intrinsic and extrinsic utility of sensory measurement
|
| 4.1.1. |
Elementary versus system processes |
the (atom-like) single cells are studied as integral parts of a (holistic) system |
| 4.1.2. |
Intrinsic versus extrinsic processes |
from an economic perspective, sensory measurements are ranked by their utility, which depends both on capacities of individual sensors (intrinsic to sensory systems) and on how useful the sensors are in the current environment (extrinsic to the system) |
| 4.2. |
Unity of apparent motion |
the apparently contradictory findings on apparent motion are unified in a framework derived from basic properties of sensory measurement |
| 4.3. |
Principles of Measurement |
we discuss the uncertainty principle of measurement (i.e., tradeoff of uncertainty between measuring signal location and signal frequency content) and Gabor’s notion of optimal measurement |
| 4.4. |
Systems of Sensors |
we review the consequences of the uncertainty principle beyond individual sensors to the systems level, revealing a unity of results from the statistical and phenomenological traditions in perceptual science |
| 4.5. |
Economics of Measurement by a System of Sensors |
we explain that the different regimes of apparent motion (tradeoff vs. coupling) occur because the expected quality of sensory measurements varies across the stimulus space we propose how the visual system may allocate its resources according to the expected utility of measurement, determined jointly by the intrinsic utility of sensors and properties of the extrinsic stimulation
|
| 4.6. |
Conclusion |
the theory based on the expected utility of sensory measurement is with a modern incarnation of the Gestalt claim that properties of system elements (the “parts”) are determined by properties of the system (the “whole”) |
| 5. |
A Bayesian Approach |
|
| 5.1. |
Introduction |
the Bayesian approach is presented as a comprehensive mathematical framework in which existing principles are unified and placed on a more principled foundation; specifically:
we apply the Bayesian approach to grouping principles such as proximity and good continuation we offer a Bayesian foundation for core concepts from Gestalt theory such as object formation and Prägnanz we discuss relationships to other frameworks (simplicity versus likelihood, Minimal Model Theory, and Bayesian network models)
|
| 5.2. |
A Bayesian Approach to Grouping Principles |
we discuss proximity and good continuation |
| 5.3. |
A Bayesian Foundation for Core Concepts from Gestalt Theory |
we discuss object formation and Prägnanz |
| 5.4. |
Relationships to Other Frameworks |
|
| 5.4.1. |
Simplicity versus likelihood |
in a Bayesian framework, the central unifying principle of Gestalt theory—Prägnanz—may be identified with the central unifying principle of Bayesian theory—maximization of the Bayesian posterior |
| 5.4.2. |
Minimal Model Theory |
the avoidance-of-coincidence principle holds that interpretations should be preferred in which as few image properties as possible are “coincidences,” such as accidents of viewpoint or configuration |
| 5.4.3. |
Bayesian network models |
|
| 5.5. |
Conclusion |
the Bayesian approach has
offered additional insight into classic Gestalt phenomena such as perceptual grouping and object formation, provided a foundation to core concepts from classic Gestalt theory such as Prägnanz, and established a bridge between likelihood and simplicity
|
| 6. |
Structural Information Theory |
|
| 6.1. |
Introduction |
information theory, connectionism, and dynamical systems theory (DST) use different formal tools to model different aspects but together they may explain how percepts are the result of cognitive processes implemented in the brain starting from the representational coding approach of Structural Information Theory (SIT), such a synthetic, multidisciplinary and typically Gestaltist perspective is sketched -
specifically, we review how SIT deals with three fundamental questions concerning perceptual organization:
how veridical are simple stimulus organizations? (we again specify the relationship between simplicity and likelihood by means of Bayes’ rule but in a different conceptual framework than before) what should be the nature of the visual regularities to be captured to arrive at simple organizations? how are simple organizations computed? this leads to a representational picture of cognitive architecture, which includes connectionist modeling ideas and which honors ideas from neuroscience and DST about neuronal synchronization
|
| 6.2. |
The Veridicality of Simplicity |
we discuss how the likelihood and simplicity principles deal with the veridicality of perception: a Gestaltist visual system that focuses on internal efficiency yields external veridicality as a side-effect |
| 6.3. |
The Nature of Visual Regularity |
we argue that
visual regularities must allow for an easy build-up of internal representations (holographic regularity) and for the specification of hierarchical organizations of the input (hierarchical transparency) three regularities satisfy these conditions: repetition, symmetry, and alternation a holographic model of regularity detection based on this formalization captures human regularity detection
|
| 6.4. |
Cognitive Architecture |
complementing the two preceding sections, addressing what needs to be processed in perceptual organization, we now address how specifically, the way SIT solves the problem of computing the simplest SIT codes of symbol strings also suggests how the brain might arrive at the simplest interpretations of visual stimuli transparallel recoding of hyperstrings of transparent holographic regularities has been implemented in a neutrally plausible algorithm that incorporates three subprocesses in the brain’s visual hierarchy: feedforward feature encoding, horizontal feature binding, and recurrent feature selection
|
| 6.5. |
Conclusion |
SIT offers another perspective on the relationship between simplicity and likelihood, arguing that a Gestaltist visual system that focuses on internal efficiency yields external veridicality as a side-effect SIT specifies the nature of the visual regularities that must be extracted to achieve this efficiency (i.e., transparent holographic regularities) as well as the nature of the cognitive architecture that explains how the simplest organizations are computed (i.e., transparallel processing by hyperstrings)
|
| 7. |
General Discussion and Conclusion |
the Gestaltists failed to provide a thorough specification of the concepts of Gestalts as configurations of parts and wholes, and the mechanisms underlying the law of Prägnanz as based on a neural isomorphism did not work out however, their intuitions protect us from falling back on a naïve mechanistic view, in which perception begins with isolated sensations, thereby denying that phenomenal experience is populated by “Gestalten” as integrated, coherent structures or forms the conceptual and theoretical foundations of Gestalt psychology have been given fresh blood and a solid backbone in both descriptive and explanatory frameworks
|
| 7.1. |
Descriptive Frameworks |
-
the initially vague notion of a holistic Gestalt can be translated into a well-defined concept that allows precise operational definitions and experimentation, in two ways:
as configurations in a feature space as superstructures in hierarchical trees
-
both descriptions are well-defined and capable of suggesting experiments, but neither of them captures the Gestalt concept in its entirety; hence, there are clearly remaining challenges:
it does not have the connotation of globality that is characteristic of true Gestalts; any arbitrary emergent property qualifies, in principle, as configural, separable, or integral it does not have the connotation of a whole being a force that binds, shapes, and resists external influences on the configuration of its parts; there is a difference between the whole having priority in processing and the Gestalt observation that the whole determines the appearance of its parts
|
| 7.2. |
Explanatory Frameworks |
Gestalt theory’s intuition about Gestalts emerging spontaneously from self-organizational processes in the brain was specified by the theory of global electrostatic field forces: systems residing at equilibria of least energy expenditure, which is at the same time also the simplest possible organization, given the available stimulation (i.e., the law of Prägnanz) -
with the dismissal of global field theory as a principle for brain organization, the Gestalt concept fragmented: economy, self-organization, the simplest possible, and “given the available stimulation” each became the starting point for four divergent approaches, each specifying the Gestalt intuition further from a modern perspective:
Section 3 specified self-organization in a dynamical systems approach to perceptual organization, explaining how the system achieves optimality by the complexity of the neural dynamics that help configure the global architecture of the system, given simple mechanisms of neural growth and adaptation; in addition, this approach does justice to perception as part of an ongoing process instead of a delineated process going from static inputs to static outputs, as in naïve mechanistic approaches Section 4 specified the principle of economy in terms of the optimization of available resources; we observed that a system of sensors that work independently at the neural level to minimize its uncertainty, is collectively responding optimally to the available patterns of stimulation Section 5 specified the conditional “given the available stimulation” in terms of likelihood, which motivated a Bayesian approach to perception; this allowed a synthetic view on simplicity and likelihood, a coherent explanation of grouping principles such as proximity and good continuation, and a solid basis to typical Gestalt notions such as objecthood and Prägnanz Section 6 specified simplicity within the context of SIT, stating that patterns are preferred according to the greatest simplicity of their symbolic description, using operators that are based on principled properties of its description language, formalizing the principles of the language, and solving the problems relating to the computational complexity of their encoding framework
-
at the same time, each of these explanatory frameworks is also confronted with remaining challenges:
it remains an open question whether the mechanism is really implemented at the neural level as proposed, and whether it generalizes beyond the realm of motion sensitivity, where it was developed how do we explain the functionality of the system at the level of its behavior? which principle governs the selection of those Gestalts that are functional to the system “given the available stimulation” rather than arbitrary others? how does selection at an evolutionary level interact with the proposed mechanisms of neural growth and adaptation? it remains to be seen how fruitful this Bayesian synthesis of likelihood and simplicity will be in the long run; moreover, all Bayesian theories face the problem of explaining how to select the priors it remains to be studied further how SIT’s encoding algorithms can be mapped onto the way in which the visual system encodes visual information
|
| 7.3. |
Conclusion |
the various theoretical approaches that have been motivated by Gestalt problems have all made considerable progress at certain aspects of the conceptual problems, yet none of them has solved the conundrum of Gestalt in its entirety; hence, there are important remaining challenges: each individual approach motivates additional detailed research questions which can now be addressed fruitfully the further specification of the connections between the frameworks, as we have started to do here, will be essential for a synthesis into the conceptually coherent framework which Gestalt theory once was only together will these frameworks make progress in answering Why do things look as they do? in sufficient detail, regarding all of its aspects: the laws of perceptual organization, faithful to perceptual experience, yet formulated in precise quantitative terms, fully explained in terms of their internal dynamics and ecological validity, spelled out at an algorithmic level and linked to its neural mechanisms, from single neurons to neuronal cell assemblies and whole systems
|
