Propositional Representation vs. Spatial Representation

The quest to understand the internal architecture of the human mind stands as one of the most rigorous and enduring endeavors in cognitive science, theoretical neuroscience, and the philosophy of mind. At the core of this investigation is the foundational problem of mental representation: the question of how an organism’s nervous system encodes, retains, manipulates, and renders intelligible the state of the external physical world and abstract conceptual domains. When an individual recalls the layout of their childhood home, deduces a mathematical theorem, plans an evasive maneuver in traffic, or evaluates the logical validity of a syllogism, what is the computational nature of the internal tokens being transformed? Does cognition rely fundamentally upon a universal, amodal, language-like code whose primitives operate through discrete formal rules, or does the cognitive apparatus deploy analog, quasi-perceptual, topologically structured representations that preserve the metric geometries of physical space?

This fundamental divergence between propositional and spatial representation has animated scientific discourse for over half a century. Commonly categorized as the “imagery debate,” the dispute transcends mere disputes over subjective phenomenology. Although human introspection frequently reveals a vibrant, dynamic “mind’s eye” populated by quasi-pictorial entities that can be mentally scanned, zoomed, and rotated, introspection is a notoriously unreliable guide to underlying computational architecture. Cognitive scientists have long recognized that what feels phenomenologically like a picture in the mind could, at the functional level, be generated and sustained entirely by non-spatial, symbolic propositions. The debate therefore pits two radically distinct views of cognitive architecture against one another: the computational-representational theory of mind, which views cognition as digital token manipulation governed by syntactic rules, versus analog and depictive models, which view mental representations as structurally and functionally continuous with perceptual systems.

Examining these competing frameworks requires a multidimensional analysis encompassing formal syntax, chronometric behavioral experimentation, functional neuroimaging, lesion-deficit neuropsychology, and artificial intelligence architectures. Each paradigm offers distinct theoretical advantages and manifests severe operational bottlenecks. Propositional representations excel at representing universal quantifiers, logical negations, semantic inheritance hierarchies, and counterfactual conditional states, yet they struggle with the frame problem and combinatorial explosions when computing dynamic spatial interactions in real time. Conversely, spatial representations provide immediate, computationally efficient access to metric distances, geometric affordances, and physical constraints through isomorphic modeling, yet they fail to capture abstract ontological categories, logical negations, or non-spatial relations without the addition of ad-hoc symbolic machinery. The synthesis of these systems reveals that biological cognition is neither purely propositional nor exclusively spatial, but rather an integrated, multi-tiered computational architecture capable of transmuting modal percepts into symbolic abstractions and decomposing formal propositions into dynamic spatial simulations.

1. Introduction to Cognitive Architectures and Mental Imagery

1.1 Defining the Nature of Mental Representation

A mental representation is an internal cognitive structure that stands in for an external entity, event, relation, or hypothetical state of affairs, preserving information in a format that can be causally transformed by cognitive operations. In the theoretical framework of cognitive psychology, mental representations constitute the primary explanatory entities bridging biological neural hardware and intelligent behavior. Without internal representations, an organism would be bound to pure stimulus-response contingencies, incapable of planning, counterfactual reasoning, memory consolidation, or abstract problem solving. The essential characteristic of any representational system lies in its dual nature: it possesses an intentional content (the semantic aspect, or what the representation is about) and an intentional vehicle (the syntactic or physical format in which that information is physically and computationally realized).

A critical challenge in cognitive science is the categorical distinction between subjective phenomenological experience and underlying functional computation. When an individual engages in mental imagery—such as visualizing an elephant standing beside a mouse—the conscious experience is undeniably depictive, spatial, and visual. The subjective impression is that of inspecting an internal picture exhibiting color, shape, scale, and metric distance. However, cognitive architectures cannot be deduced solely from introspective awareness. Introspection does not reveal the algorithmic code through which information is stored in long-term memory, nor does it illuminate the neural transformations executing the retrieval process. A functional architecture could simulate spatial relationships by executing algorithms over amodal, non-spatial data structures, creating a phenomenal experience that misrepresents the actual computational operations driving the system. Thus, scientific inquiry must separate the phenomenology of the mental image from the functional architecture that executes the cognitive task.

Mental representations are foundational to memory retrieval, deductive reasoning, and perceptual processing. When sensory systems are exposed to external stimuli, they transduce physical energy into neurochemical signals, generating transient perceptual representations. Once the stimulus is removed, the cognitive system must rely on internal stores to reconstruct, interpret, and manipulate that information. Whether this memory store consists of discrete predicate-argument strings or analog spatial layouts governs the efficiency, error profile, and computational limits of cognition. This inquiry ultimately asks whether the human mind functions primarily as a digital computer—manipulating arbitrary, discrete, amodal symbols via formal syntactical operations—or as an analog simulator, constructing dynamic models whose internal topological configurations directly mirror the physical geometry of the represented world.

1.2 The Historical Genesis of the Imagery Debate

The historical roots of the imagery debate trace back to classical philosophical empiricism and rationalism. Empiricists such as John Locke and David Hume posited that the contents of thought consist of “ideas,” which they characterized as faint copies or lingering impressions of sensory perceptions. In contrast, rationalist traditions and subsequent early 20th-century movements, such as the Würzburg School of psychology, identified instances of “imageless thought,” demonstrating that cognitive processes such as judgment, abstraction, and directional intent could occur entirely devoid of conscious sensory imagery. The rise of behaviorism temporarily banished the study of mental representations from academic psychology, treating the internal states of the mind as an unobservable, unscientific black box. However, the cognitive revolution of the 1950s and 1960s restored the legitimacy of internal states, grounded in the newly formulated computational theory of mind.

In the 1970s, the debate crystallized into an intense empirical and theoretical conflict between two primary architects: Stephen Kosslyn and Zenon Pylyshyn. Kosslyn, along with researchers such as Roger Shepard and Jacqueline Metzler, developed chronometric paradigms demonstrating that mental transformations exhibit properties strikingly identical to physical transformations in real space. Shepard and Metzler’s seminal 1971 mental rotation studies demonstrated that the time required to determine whether two three-dimensional shapes were identical was a direct, linear function of the angular discrepancy between them. Kosslyn subsequently introduced mental scanning experiments showing that the time required to scan across a mental image increased linearly with the physical distance separating the scanned points. From these behavioral findings, Kosslyn formulated the depictive-spatial model, asserting that mental images are genuinely spatial representations instantiated in a functional medium operating like a coordinate space.

This analog interpretation was challenged by Zenon Pylyshyn in his 1973 paper, “What the Mind’s Eye Tells the Mind’s Brain: A Critique of Mental Imagery.” Pylyshyn argued that the introspective experience of an image is computationally epiphenomenal—a downstream subjective byproduct that plays no functional role in information processing. He asserted that human cognition is fundamentally symbolic and language-like, operating via discrete, amodal propositions. Pylyshyn claimed that the apparent spatial constraints observed in reaction-time tasks did not reflect an analog cognitive medium, but rather the deployment of “tacit knowledge”: participants unconsciously simulated the physical laws and temporal latencies of the real world because they understood what the experimental tasks demanded. This clash prompted decades of empirical work, shifting from behavioral reaction-time experiments to high-resolution neuroimaging, transcranial magnetic stimulation, and computational modeling.

1.3 Why Propositional and Spatial Frameworks Are Frequently Conflated

In both lay discussions and technical literature, propositional and spatial frameworks are frequently conflated due to pervasive introspective illusions. The human mind falls prey to what philosophers call the “intentional fallacy” or the “phenomenological fallacy”: the mistaken assumption that the structural properties of the object of thought must also be structural properties of the representation itself. When thinking about a green circle, the conscious mind experiences roundness and greenness. Consequently, an introspective observer assumes that the neural representation itself must be round and green, or at least spatial and depictive. This intuitive leap obscures the computational possibility that a non-spatial, non-colored list of semantic predicates, such as SHAPE(OBJECT1, CIRCLE) and COLOR(OBJECT1, GREEN), can generate and sustain the cognitive behavior without containing any spatial medium whatsoever.

A second source of confusion stems from neurofunctional co-activation. Human neuroimaging studies frequently reveal that during complex problem-solving tasks, both verbal-propositional networks (such as Broca’s and Wernicke’s areas in the left hemisphere) and spatial-perceptual networks (such as the parietal cortex and visual areas V1 through V4) are engaged simultaneously. When an individual reads a spatial description—such as “the lamp is to the left of the manuscript”—the brain processes the syntactic structure of the sentence while frequently triggering an automatic, quasi-perceptual spatial representation of the scene. The rapidity and automaticity of this bidirectional translation create an experiential continuum wherein symbolic propositions and spatial configurations feel unified, making it difficult to isolate their distinct computational contributions without precise experimental manipulations.

Finally, semantic ambiguity in cognitive science terminology has contributed to this conflation. Terms such as “cognitive map,” “mental model,” “schema,” and “spatial semantics” are often applied interchangeably across both paradigms. A “cognitive map” can refer either to an analog metric coordinate system realized in grid and place cell networks or to an abstract relational graph of linked topological nodes devoid of metric geometry. Furthermore, hybrid cognitive architectures developed in modern artificial intelligence and computational neuroscience frequently obscure this boundary. Systems that translate vector embeddings into spatial coordinates or convert linguistic prompts into geometric visual tokens blur the distinction between symbolic storage and analog spatial retrieval, demanding precise formal definitions of the underlying representations.

2. Understanding Propositional Representation

2.1 Foundational Principles and Language-Like Syntax

Propositional representation is an amodal, symbolic format that encodes information through discrete, abstract tokens structured according to formal syntactic rules. Unlike sensory representations, which retain the physical characteristics of the stimuli that generated them, propositional representations are completely divorced from perceptual modalities. A proposition represents the semantic core of an assertion, independent of whether that assertion was seen, heard, read, or generated through deductive inference. In this system, the internal mental code functions precisely like a formal logic or an idealized language—often termed the “Language of Thought” or Mentalese, as articulated by the philosopher Jerry Fodor. The basic units do not resemble their referents; the symbol used to denote a dog bears no physical, metric, or topological resemblance to a biological canine.

The formal syntax of propositional representations is commonly modeled on predicate calculus. A proposition is composed of a predicate, which designates an attribute, state, or relation, and one or more arguments, which denote the entities instantiated by that predicate. The canonical notation is structured as Predicate(Argument_1, Argument_2, ..., Argument_n). For example, the spatial scenario of a cat sitting on a rug is formally encoded as:

ON(CAT, MAT)

In this structure, CAT and MAT serve as symbolic arguments denoting conceptual categories or specific entities, while ON operates as a relational predicate specifying the spatial configuration between them. Crucially, the predicate ON does not occupy physical space, nor does it possess spatial extent; it is an abstract operator with specific semantic truth conditions. If the identical situation were expressed as “The mat is under the cat,” the propositional structure would be rewritten as UNDER(MAT, CAT), which shares an equivalent truth value under systematic logical transformation.

A defining characteristic of propositional systems is semantic compositionality. In a compositional architecture, the meaning of a complex expression is entirely determined by the syntactic rules that combine its elementary conceptual primitives. This structure yields systematicity and productivity: an individual who possesses the conceptual tokens BITE, DOG, and MAN, along with the syntactic rules of predicate argument structure, can understand and generate both BITE(DOG, MAN) and BITE(MAN, DOG), recognizing that despite containing identical components, the propositions possess fundamentally different truth conditions. The meaning of the whole depends systematically on how the parts are computationally assembled, providing the cognitive apparatus with an unbounded capacity to formulate novel thoughts from a finite vocabulary of discrete symbols.

2.2 Zenon Pylyshyn and the Epiphenomenon Hypothesis

Zenon Pylyshyn’s critique of mental imagery remains the most formidable theoretical defense of propositional cognitive architecture. Pylyshyn posited that the subjective phenomenological experience of inspecting a mental picture is an epiphenomenon—a downstream, functionally inert byproduct of underlying computations that are non-spatial, amodal, and propositional. To illustrate this concept, Pylyshyn frequently used the analogy of a mechanical clock: the ticking sound emitted by the gears is a real, measurable physical phenomenon, but it plays no causal role in driving the movement of the hands. Similarly, while conscious humans undoubtedly experience mental imagery, Pylyshyn argued that this phenomenology has no causal computational role in cognitive operations. The actual work of memory search, spatial inference, and geometric reasoning is executed through propositional computations operating entirely beneath the threshold of conscious awareness.

To dismantle the empirical findings of Kosslyn and Shepard, Pylyshyn introduced the tacit knowledge hypothesis. He contended that when participants in an experiment are instructed to “scan” across a mental image or “rotate” an imagined object, they inadvertently draw upon their implicit real-world knowledge of how physical objects and physical observation operate. Participants know that in the physical world, it takes longer to travel between two points that are far apart than two points that are close together. Consequently, because they comprehend the task demand characteristics, they deliberately pace their responses to simulate the temporal dynamics of physical visual scanning. The observed linear relationship between distance and reaction time is not an intrinsic property of the cognitive medium; rather, it is a deliberate, knowledge-driven simulation executing over propositional structures.

Pylyshyn formalized this critique through his criterion of cognitive penetrability. A cognitive process or architectural feature is said to be cognitively penetrable if it can be systematically altered or influenced by an individual’s goals, beliefs, tacit knowledge, or explicit instructions. Pylyshyn demonstrated that many mental imagery phenomena are penetrable: if an experimenter changes the instructions or alters the conceptual framing, the supposedly immutable metric constraints—such as the scanning latency—vanish or change their mathematical characteristics. If an operation is cognitively penetrable, it cannot be an intrinsic, hardwired property of the biological cognitive architecture; it must be an operation executed by software-level beliefs and strategies. True cognitive architecture must be cognitively impenetrable, operating independently of the organism’s subjective belief states.

Finally, Pylyshyn argued that positing depictive representations leads directly into an ungrounded homunculus fallacy. If the mind truly generates an internal picture, there must be an internal agent—a “mind’s eye” or homunculus—to inspect, interpret, and extract information from that picture. This internal observer would itself require a perceptual and cognitive system, which in turn would require another smaller picture and another observer, initiating an infinite regress of explanatory failures. Pylyshyn asserted that the only way to avoid this regress is to ground cognition in self-interpreting, amodal propositional structures whose semantic content is computationally realized through physical causal mechanisms, entirely removing the need for a subjective viewer.

2.3 Computational Mechanisms and Semantic Networks

At the algorithmic level, propositional knowledge is frequently structured into semantic networks and directed relational graphs. In these networks, concepts are represented as discrete nodes, while the relationships between them are instantiated as labeled, directed edges. Rather than utilizing metric space, where distance is measured in millimeters or visual angles, distance in a semantic network is purely topological and structural, defined by the number of intervening links separating two conceptual nodes. This network structure operationalizes knowledge via formal schemas that can be parsed by computational logic engines, providing an explicit mechanism for human semantic memory.

Information retrieval within propositional networks relies heavily on the mechanism of spreading activation, originally formalized by cognitive psychologists Allan Collins and Elizabeth Loftus. When a conceptual node is accessed—whether via sensory input or internal deliberation—it enters a state of excitation. This activation spreads outward along the associative links to adjacent nodes, decaying across structural distance and time. The speed and probability of retrieving a related concept are determined by the strength of the connecting edges and the topological proximity within the network, not by metric distance in physical space. For instance, activating the node CANARY primes the adjacent nodes BIRD and YELLOW through rapid link traversal, enabling immediate verification of propositions such as “A canary is yellow” without constructing any visual representation of a canary.

Propositional networks natively support Boolean logical operations, enabling operations that analog systems cannot naturally compute:

• Logical Negation: Propositional systems can explicitly mark a condition as false or non-existent using an operator, such as NOT[HAS(BIRD, GILLS)]. Spatial systems cannot draw or depict the pure absence of a feature without depicting an alternative spatial state.
• Universal and Existential Quantification: Propositions can easily encode assertions across entire sets, using quantifiers such as $\forall x (BIRD(x) \rightarrow FEATHERS(x))$ or $\exists x (MAMMAL(x) land AQUATIC(x))$.
• Counterfactual and Modal Reasoning: Propositional syntax can embed clauses inside modal operators, such as POSSIBLE(RAIN) or BELIEVES(JOHN, TALL(MARY)), segregating internal beliefs from objective reality.

Furthermore, propositional architectures achieve exceptional economy of storage through hierarchical inheritance hierarchies. In a canonical semantic network, properties shared across a broader category are stored exclusively at the highest relevant ontological node. For example, the property CAN_BREATHE is stored at the node ANIMAL. The subordinate node BIRD inherits this property automatically, and the sub-subordinate node ROBIN inherits it from BIRD. Long-term memory avoids redundant storage; it is unnecessary to store “can breathe” alongside robin, sparrow, hawk, and eagle. When a specific question is posed (“Can a robin breathe?”), the cognitive architecture conducts a deductive search across the hierarchy, inferring the truth value through syntactic traversal. This algorithmic compression renders propositional systems vastly superior to uncompressed spatial arrays in terms of memory efficiency.

2.4 Advantages and Inherent Limits of Propositional Architectures

The primary computational advantage of propositional architectures is their universal expressive power. A propositional format can encode any relationship that can be articulated in a formal language, encompassing completely non-visual, abstract concepts such as “justice,” “inflation,” “entropy,” “transitive inference,” and “subjunctive conditionals.” It provides a parsimonious, unified cognitive medium wherein perceptual memories, abstract rules, linguistic communications, and meta-cognitive reflections share a single, mathematically rigorous lingua franca. The amodal nature of propositions permits effortless cross-modal synthesis: sensory data from auditory, tactile, and visual modalities are stripped of their physical format and translated into a unified code, allowing unified reasoning over heterogenous inputs.

However, propositional architectures encounter severe computational bottlenecks when executing real-time spatial transformations and geometric inferences. In a purely propositional system, a physical environment containing multiple objects must be represented as an exhaustive inventory of explicit binary and n-ary relations: LEFT_OF(A, B), RIGHT_OF(B, A), ABOVE(C, A), BETWEEN(D, A, B), and so forth. If an object within that environment moves, the system faces an acute manifestation of the frame problem: it must explicitly recalculate and update every individual spatial relation relative to every other entity in the scene. In a complex environment containing $N$ entities, the number of potential relations scales quadratically ($O(N^2)$) or factorially, requiring substantial computational overhead simply to maintain spatial coherence.

Moreover, propositional models face the intractable symbol grounding problem, famously articulated by Stevan Harnad. If the cognitive architecture consists entirely of amodal, arbitrary symbols whose meanings are defined strictly through their relations to other amodal symbols, the entire system remains ungrounded—a closed loop of syntactic manipulation devoid of intrinsic semantic contact with the physical world. A system that defines the symbol HORSE as an ungulate mammal with hooves, and defines HOOF through further symbolic definitions, never establishes an empirical anchor to actual visual forms, tactile surfaces, or motor interactions. Purely propositional networks must ultimately rely upon an interface mechanism that bridges arbitrary tokens with physical reality, exposing an inherent limit of the format.

3. Understanding Spatial Representation

3.1 Foundational Principles of Analog and Depictive Formats

Spatial (depictive or analog) representation is a cognitive format wherein the structure of the internal representational medium directly mirrors the structural, topological, and metric relationships of the represented entity. In a spatial representation, the internal vehicle is not an arbitrary token; rather, it is an analog model whose geometric properties correspond homomorphically or isomorphically to the physical reality it signifies. The defining attribute of a depictive format is that every part of the representation corresponds to a part of the represented object, and the metric distances between the parts of the representation preserve the metric distances between the corresponding parts of the physical world.

Unlike propositional codes, which are amodal and discrete, spatial representations are intrinsically modal and continuous. They retain sensory-specific features: topological orientation, continuous contours, geometric angles, aspect ratios, and spatial gradients. In a depictive medium, there are no discrete, arbitrary tokens like ABOVE or LEFT_OF. Instead, the spatial relation “above” is instantiated implicitly: one portion of the representation occupies a position in the functional medium that is physically or functionally superior to another portion. Spatial relations are not declared via predicates; they emerge naturally from the geometric configuration of the internal model itself. If a depictive representation shows a circle inside a square, the property of containment is not explicitly declared as a linguistic rule; it is an intrinsic geometric fact of the medium.

This structural isomorphism dictates that transformations within a spatial medium must operate continuously rather than discretely. When an analog representation is scaled, rotated, or translated, it cannot jump instantly from an initial state to a final transformed state without passing through intermediate trajectory states that correspond to physical motion through space. If a representation of a line segment is rotated 90 degrees clockwise, the computational medium must generate the intermediate representations at 10, 20, 30, and 80 degrees. The representational medium is fundamentally constrained by spatial geometry, imposing an analog continuity that directly shapes human processing latency and behavioral execution.

3.2 Stephen Kosslyn and the Functional Equivalence Hypothesis

Stephen Kosslyn provided the empirical and theoretical foundations of depictive theory, challenging the propositional orthodoxy through systematic chronometric paradigms. Kosslyn formulated the Functional Equivalence Hypothesis, which asserts that visual mental imagery relies on the same internal representational structures and computational mechanisms as visual perception. Although an imagined object lacks an external physical stimulus, the internal cognitive vehicle that instantiates the mental image functions in a manner computationally equivalent to the neural structures activated during actual visual perception.

Kosslyn verified this hypothesis through his classic mental scanning experiments. In his landmark fictional island experiment, participants memorized a map containing various landmarks separated by varying metric distances (such as a hut, a well, a swamp, and a tree). The map was then removed, and participants were instructed to generate a vivid mental image of the island and fixate their attention on a specific starting location (e.g., the hut). When the experimenter named a second target landmark (e.g., the tree), participants were asked to “scan” across their mental image from the origin to the destination, pressing a button the moment their internal gaze reached the target. Kosslyn demonstrated that the reaction time required to complete the mental scan was a direct linear function of the objective, physical metric distance separating the landmarks on the original map.

Crucially, this linear relationship persisted even when the intermediate space was completely devoid of alternative landmarks, ruling out the hypothesis that response latencies were caused by traversing intervening discrete propositional nodes. Kosslyn demonstrated that mental scanning traverses functional metric space: to cross greater distances within an analog medium requires proportionally greater time, exactly as a physical eye would scan across a physical landscape. When combined with Shepard and Metzler’s findings on mental rotation—where response times correlate strictly with the degree of angular transformation required—these data provided empirical evidence that the internal operations were structurally constrained by geometric dimensions.

To formalize these findings computationally, Kosslyn proposed the Visual Buffer Theory. He argued that the mind possesses an internal, active coordinate space—the “visual buffer”—realized in the retinotopically organized areas of the visual cortex. The visual buffer operates as an array of discrete coordinate cells that functions as an analog continuum. Images are generated in this buffer either through bottom-up perceptual input from the eyes or top-down generative processes driven by information stored in long-term memory. Once an image is rendered in the buffer, cognitive processes can “inspect” it using an attentional window, zoom in to resolve fine details, rotate it, or shift its coordinate points. Kosslyn maintained that depictive representations possess unique computational affordances: properties not explicitly encoded during the initial observation can be read off directly from the spatial array post-hoc through visual inspection.

3.3 Neurobiological Foundations and Topographic Organization

The definitive test between propositional and spatial representations shifted toward cognitive neuroscience, specifically focusing on the functional architecture of the primate visual system. A cornerstone of the depictive framework is that the primary visual cortex (Brodmann Area 17, or V1), along with adjacent extrastriate regions (V2, V3), is neurobiologically organized as a series of retinotopic maps. In a retinotopic map, adjacent neurons in the cortical tissue respond to adjacent points on the retina. The primary visual cortex preserves the exact topological geometry of the physical retina, mapping the sensory visual field onto a physical sheet of cortical tissue through a conformal mathematical transformation (specifically, a logarithmic polar transformation). This biological structure functions as an anatomical depictive medium: metric relationships in the visual field are directly preserved in the physical geometry of cortical neural firing.

Using high-resolution functional Magnetic Resonance Imaging (fMRI) and Positron Emission Tomography (PET), neuroscientists have demonstrated that when human subjects generate visual mental images with their eyes closed, these retinotopically mapped cortical areas exhibit robust activation. Kosslyn and colleagues showed that when participants imagine small objects, the neural activation is confined to the posterior portion of Area 17, which processes the central foveal visual field. Conversely, when participants imagine large objects that occupy a wide visual angle, the activation spreads to the anterior portions of Area 17, which map the peripheral visual field. The spatial extent of cortical activation in primary visual cortex directly reflects the imagined metric size of the mental object, providing neurobiological confirmation that mental imagery utilizes a spatial medium.

Furthermore, this spatial architecture is reinforced by specialized regions within the posterior parietal cortex and hippocampal formation. While occipital regions provide the depictive canvas, the parietal lobes compute spatial coordinate transformations, converting viewer-centered (egocentric) coordinates into world-centered (allocentric) frameworks. Within the hippocampus and entorhinal cortex, place cells and grid cells construct a literal, metric coordinate framework for internal navigation. Grid cells fire at regular, tessellated triangular intervals across physical space, providing an analog, metric coordinate ruler that operates during both physical traversal and purely imagined mental navigation.

Neuropsychological lesion studies provide causal evidence supporting this spatial architecture. Patients suffering from visual hemispatial neglect following damage to the right parietal lobe fail to perceive or respond to objects on the left side of physical space. In a classic experiment by Edoardo Bisiach and Claudio Luzzatti (1978), neglect patients were asked to imagine standing in the Piazza del Duomo in Milan and describe the landmarks. When imagining facing the cathedral, patients systematically failed to name the buildings on the left side of the square, describing only the right side. However, when instructed to mentally reverse their perspective and imagine standing with their backs to the cathedral facing the piazza, they promptly named the buildings they had just omitted, while failing to name the buildings on their newly imagined left. Because their underlying propositional knowledge regarding the architecture of Milan remained intact, the selective omission of details from one side of the mental image demonstrated that imagery relies on a structurally spatial, viewpoint-dependent coordinate medium.

3.4 Advantages and Inherent Limits of Spatial Architectures

The primary advantage of spatial representations lies in their direct computational efficiency for geometric, physical, and affordance-based reasoning. In a depictive medium, complex metric relationships are represented implicitly and in parallel. There is no requirement to calculate transitive inferences through formal logic; for example, determining whether point A is closer to point B than point C requires only an immediate spatial comparison of distance vectors across the medium. Affordances—such as whether a sofa will clear a narrow doorway, or whether an obstacle can be cleared by leaping—can be simulated through analog rotations and translations, leveraging the intrinsic physics of the internal model rather than computing thousands of propositional boundary equations.

Moreover, spatial representations support the emergent discovery of novel information. An individual can memorize an image or combine two visual shapes (e.g., an uppercase letter “D” rotated 90 degrees counterclockwise placed on top of an uppercase letter “J”) and immediately “read off” a totally new emergent shape (an umbrella). In a propositional system, this emergent shape cannot be detected unless an explicit, complex inference engine calculates the intersecting line geometries and compares them against thousands of categorical prototypes. Spatial representations allow instant, parallel visual extraction of emergent contours, overlaps, and spatial occlusions.

Despite these strengths, spatial representations suffer from strict architectural limitations:

• Working Memory Constraints: Analog spatial representations are computationally expensive to maintain. The visual buffer has a limited resolution, and active mental images degrade rapidly over milliseconds, demanding high attentional and metabolic resources.
• Incapacity to Represent Abstract Negation: A spatial representation cannot portray negative facts without depicting an alternative positive state. One cannot construct a mental image of “there is no dog in the yard” without simply imagining an empty yard, which could just as easily depict the absence of a cat, an automobile, or a house.
• Vulnerability to Cognitive Illusions: Analog images are prone to distortion based on structural biases. Cognitive maps routinely introduce systematic metric errors, such as straightening curved roadways or falsely aligning non-parallel continents (e.g., individuals incorrectly judge Reno, Nevada to be east of Los Angeles, California due to categorical conceptual distortions overriding metric representations).
• Inability to Direct Generalize: A depictive representation must always depict a specific instance with a determined size, orientation, and vantage point. It cannot construct a generic “triangle” without deciding whether it is equilateral, isosceles, or scalene, making it ill-suited for universal mathematical abstraction.

4. Key Differences Between Propositional and Spatial Representation

4.1 Format and Code Structure: Discrete Symbols vs. Continuous Topography

The fundamental distinction between propositional and spatial representations resides in their internal code structure. Propositional systems use an arbitrary, amodal, digital code. The relationship between the signifier (the mental token) and the signified (the external referent) is entirely arbitrary; it is established purely by convention and functional mapping. The word “triangle” or the computational token TRIANGLE bears zero physical resemblance to a three-sided geometric figure. Because the symbols are discrete, they operate as distinct, indivisible digital tokens that are manipulated through algorithmic syntactical rules. The information in a propositional system is categorically partitioned, explicit, and segmented into modular arguments and predicates.

Conversely, spatial representations utilize an analog, depictive code governed by continuous topography. The relationship between the internal vehicle and the external referent is one of structural homomorphism or topological isomorphism. The representational medium does not use arbitrary symbols; rather, the configuration of the medium itself preserves the metric intervals, continuous contours, and geometric angles of the external referent. While a propositional representation represents spatial relationships explicitly through labeled predicates (such as declaring ADJACENT(X, Y)), a spatial representation leaves these relationships implicit within its medium. The adjacency of $X$ and $Y$ in a spatial medium is simply instantiated by their proximity within the coordinate field. It does not require a linguistic assertion; it exists as an emergent geometric fact.

4.2 Computational Transformations and Processing Latency

The operations executed upon these two representational formats yield fundamentally different mathematical properties and processing latencies. In a propositional architecture, information processing proceeds via logical deduction, pattern matching, variable substitution, and syntactic string-rewriting. The computational cost of an operation in a propositional network is determined primarily by structural and graph-theoretic metrics: the number of nodes that must be traversed, the depth of the inheritance hierarchy, the number of logical clauses that must be evaluated, and the branching factor of the inference engine. Metric physical distance in the real world is computationally irrelevant unless it is explicitly encoded as an arithmetic parameter.

In contrast, transformations within a spatial representation are governed by continuous geometric operations: metric translation, coordinate rotation, analog scaling, and spatial interpolation. The processing time required to transform an analog image is directly proportional to the physical, metric properties being simulated. In mental rotation, the chronometric latency is a linear function of the angle of rotation (often proceeding at an invariant rate of roughly 60 degrees per second). In mental scanning, the latency scales linearly with Euclidean distance. In spatial systems, computational cost is bounded by the spatial resolution of the coordinate matrix and the geometric magnitude of the transformation, whereas in propositional systems, cost scales with the syntactic complexity and categorical density of the rule set.

4.3 Neuroimaging Correlates and Neural Substrates

Cognitive neuroscience has demonstrated a marked functional dissociation between the neural substrates supporting propositional operations and those mediating spatial representations. Propositional reasoning—involving formal syntax, language processing, taxonomic categorization, and logic evaluation—primarily engages a left-lateralized network of perisylvian cortices, including Broca’s area (inferior frontal gyrus), Wernicke’s area (superior temporal gyrus), and the left middle temporal lobe. These regions execute the phonological, syntactic, and semantic operations required to parse, generate, and evaluate amodal predicate-argument structures.

Spatial representation and analog transformations, on the other hand, recruit an anatomically distinct, bilateral or right-dominant occipito-parietal visual processing network. The depictive visual buffer relies on early, retinotopically mapped visual cortices (V1, V2, V3, and V4), which fire in patterns preserving the spatial layout of an imagined scene. Dynamic transformations—such as mental rotation, spatial coordinate shifting, and allocentric mapping—heavily recruit the posterior parietal cortex, particularly the superior parietal lobule and the intraparietal sulcus, along with frontal eye fields. To confirm that the primary visual cortex is causally involved in mental imagery rather than merely showing passive correlated activation, Kosslyn and colleagues applied repetitive Transcranial Magnetic Stimulation (rTMS) to Area 17, temporarily disrupting its functional capacity. The resulting virtual lesion caused a significant, equivalent impairment in both perceptual visual discrimination and mental imagery tasks, proving the functional necessity of the early sensory cortex in depictive spatial reasoning.

4.4 Comparative Overview Matrix

To synthesize the theoretical, computational, and neurobiological divergences between these two representational frameworks, the following comparative matrix contrasts their fundamental dimensions:

Dimension / Metric	Propositional Representation	Spatial (Depictive) Representation
Core Format & Code	Amodal, discrete, arbitrary linguistic symbols organized by formal syntactical rules.	Modal, continuous, topologically isomorphic coordinate layouts preserving metric geometry.
Relational Encoding	Explicit: Relational predicates (e.g., BEHIND(A, B)) must be formally declared.	Implicit: Relations emerge naturally from the geometric positioning within the coordinate medium.
Computational Operations	Syntactic rewriting, Boolean logic, spreading activation, deduction, unification.	Analog rotation, metric scanning, coordinate translation, spatial zooming, coordinate clipping.
Processing Latency Determinants	Graph search depth, number of intervening conceptual nodes, syntactic complexity.	Metric Euclidean distance, angular degree of rotation, spatial resolution scale.
Negation & Modality	Native capability: Easily expresses negation (NOT), universal quantification, and conditionals.	Incapable of direct negation; cannot depict the non-existence of an object without ad-hoc markers.
Primary Neural Correlates	Left-lateralized perisylvian network, inferior frontal gyrus, middle temporal cortex.	Occipital retinotopic visual areas (V1–V4), superior parietal lobule, intraparietal sulcus.
Causal Vulnerability (Lesions)	Aphasias, semantic dementia, conceptual agnosia, left perisylvian focal lesions.	Visual hemispatial neglect, visual agnosia, loss of mental rotation from parietal damage.
Storage Economy	High: Uses hierarchical inheritance to eliminate redundant properties across classes.	Low: Requires detailed structural rendering; vulnerable to high working memory degradation.
Primary Behavioral Task Paradigms	Sentence-picture verification, category membership verification, syllogistic deduction.	Shepard-Metzler mental rotation, Kosslyn map scanning, visual size-judgment chronometry.
Typical System Errors	Categorical fallacies, deductive invalidity, failure of relational inference.	Boundary extension, geometric flattening, alignment heuristics, coordinate distortions.

The comparative analysis demonstrates that neither representational format is universally superior. Propositional representations provide computational efficiency for abstract, logical, and categorical tasks, while spatial representations offer high-throughput, parallel evaluation of geometric, metric, and physical relationships.

5. Similarities and Conceptual Overlap

5.1 Shared Cognitive Objective in Environmental Simulation

Despite their fundamental architectural divergence, propositional and spatial representations share an identical ultimate objective: constructing internal models that simulate the external environment to guide adaptive behavior. Both formats are tools deployed by the human central nervous system to decouple cognition from the immediate sensory present. By constructing internal models, organisms avoid costly trial-and-error behaviors in physical space, using mental simulations to evaluate risks, project trajectories, formulate counterfactual hypotheses, and plan actions before committing motor output.

Furthermore, both systems draw heavily from identical long-term memory structures to assemble internal working models. Long-term semantic knowledge informs spatial imagery: when an individual visualizes a giraffe, long-term conceptual knowledge dictates that the neck must be disproportionately long relative to the torso. Conversely, perceptual and spatial experiences continually feed bottom-up data into the formation of propositional abstractions. Both systems are subject to strict human cognitive bottlenecks: attentional selection, operational limits in working memory capacity, and decay over temporal intervals. Both systems do not function as isolated systems, but rather as interdependent components within a broader computational cognitive architecture.

5.2 The Dual-Coding Framework and Bidirectional Translation

The theoretical framework reconciling these two modes was formulated by Allan Paivio in his landmark Dual-Coding Theory (1971). Paivio proposed that human cognition operates through two independent, yet structurally linked and interacting representational subsystems:

• The Verbal System (Logogens): A linguistic, amodal, propositional system designed to process, retain, and manipulate arbitrary linguistic tokens, abstract symbols, and syntactical relations.
• The Non-Verbal System (Imagens): A modality-specific, depictive, analog system specialized for generating, processing, and manipulating structural, spatial, and sensory representations.

Paivio demonstrated that although these two systems can function independently, they are connected via rich bidirectional referential pathways. An individual can translate from an imagen to a logogen via descriptive processes—such as verbally articulating the appearance of a complex visual landscape (analogous to computational image captioning). Conversely, an individual can translate from a logogen to an imagen via generative visualization—such as reading a descriptive passage of text and immediately generating a vivid internal spatial layout of the scene.

A central finding of Dual-Coding Theory is that human memory retention and problem-solving capacities improve dramatically when information is encoded simultaneously across both systems—a phenomenon known as the dual-coding advantage. Words that denote concrete, easily visualizable objects (such as “apple” or “locomotive”) are retained and recalled significantly better than purely abstract words (such as “epistemology” or “inherence”). Concrete words automatically trigger both a logogen (the linguistic symbol) and an associated imagen (the visual representation), establishing redundant memory traces that increase retrieval probability. This bidirectional synergy demonstrates that human cognitive performance peaks when propositional and spatial representations are deployed in concert.

5.3 Hybrid Mental Models and Intermediate Formats

The boundary between purely propositional and purely spatial representations is further bridged by the concept of mental models, comprehensively developed by cognitive scientist Philip Johnson-Laird. In his 1983 work *Mental Models*, Johnson-Laird proposed that human deductive reasoning relies neither on formal rules of propositional logic nor on fully rendered depictive images, but on an intermediate representational structure: the mental model. A mental model is structural and analog in its topological relationships, yet schematic and abstract in its internal detail.

Consider a spatial deduction problem: “The fork is to the left of the plate. The knife is to the right of the plate. Where is the fork relative to the knife?” A formal propositional logic system would solve this via abstract inference rules such as transitivity and spatial axioms. A depictive imagery system would construct a fully rendered sensory image of a specific fork, plate, and knife, complete with colors, textures, and lighting. A mental model, however, simply constructs an abstract array of tokens preserving only the relevant spatial topology:

[FORK] [PLATE] [KNIFE]

This internal model is spatial because the relative positions of the tokens correspond to the spatial relations of the objects; however, it is not an image, because it omits all visual, metric, and sensory details. Reasoning occurs by inspecting this model, noting direct spatial relationships, and attempting to construct counterexamples that falsify the deduction. This hybrid architecture explains why humans excel at deductive reasoning when problems can be mapped onto spatial configurations, while struggling to execute logically equivalent syllogisms expressed in purely abstract propositional variables.

This intermediate format is also evident in human cognitive maps, as demonstrated by Edward Tolman, Barbara Tversky, and other cognitive geographers. Human navigation does not rely on a single, uniform Cartesian coordinate sheet, nor on a disconnected list of verbal directions. Instead, cognitive maps are mixed, hierarchical structures: they retain topological and schematic spatial relations (such as network connectivity and route sequences) while embedding discrete propositional categories (such as neighborhood boundaries, regional classifications, and landmark labels). This hierarchical categorization routinely introduces distortions into spatial reasoning, demonstrating how symbolic categories warp and reconfigure underlying analog spaces.

6. Application Contexts: Practical Guidance and Cognitive Utility

6.1 When Propositional Representation Excels

Propositional representations are computationally optimal whenever a task demands the manipulation of abstract, non-spatial, logically complex, or rule-governed structures. In formal domains such as mathematical theorem proving, computational programming, and deductive logic, spatial models are insufficient and often misleading. Proving that an equation holds true across all prime numbers requires the deployment of universal quantification, variable substitution, and algebraic rewriting—operations that cannot be visually depicted in an analog medium without introducing accidental geometric biases.

A second major application domain is the management of abstract conceptual ontologies and legal frameworks. Concepts such as “constitutional liability,” “intellectual property,” “corporate fiduciary duty,” and “inflationary monetary policy” possess no physical shape, color, or metric coordinates. Attempting to model these domains through spatial configurations results in ambiguous visual metaphors. A propositional architecture provides an expressive, rule-governed syntax capable of modeling nested conditional structures, regulatory permissions, obligations, and systemic exceptions:

PERMITTED(INDIVIDUAL, ACTION) $leftarrow$ LICENSED(INDIVIDUAL) $land$ $neg$REVOKED(LICENSE)

This logical compositionality enables legal and regulatory reasoning to operate with clarity and absolute truth-evaluability.

Finally, propositional formats excel in large-scale knowledge organization, database design, and semantic webs. In enterprise data systems, information must be compressed, indexed, and retrieved across billions of categorical relations. Implementing knowledge bases via propositional Resource Description Framework (RDF) triples—structured as Subject-Predicate-Object—allows relational search, inference queries, and taxonomic inheritance hierarchies to run with computational speed. The amodal economy of propositional representations prevents storage inflation, enabling systems to maintain millions of invariant conceptual rules without the memory footprint demanded by high-resolution spatial models.

6.2 When Spatial Representation Excels

Spatial representations are computationally essential when tasks involve real-time physical interaction, dynamic geometric calculation, and embodied spatial navigation. In real-time robotics, autonomous vehicle navigation, and spatial terrain mapping, relying strictly on propositional deductions creates fatal computational latencies. An autonomous drone navigating through a dense forest cannot afford to compute linguistic propositions regarding the spatial coordinates of every branch. Instead, it must utilize spatial representations: continuous occupancy grids, depth point clouds, and topological vector fields that map the metric proximity of obstacles in parallel, enabling real-time trajectory updates and obstacle avoidance.

A second critical domain is structural engineering, architectural design, and mechanical analysis. When evaluating the physical equilibrium of a suspension bridge, the clearances within a multi-axis combustion engine, or the structural stress distribution of a load-bearing column, spatial and depictive representations are superior. An engineer utilizes mental imagery or computer-aided analog simulations to observe how geometric components interact under rotational and translational stresses. The analog medium captures collisions, geometric interferences, and leverage points that would require millions of discrete propositional equations to represent symbolically.

Furthermore, spatial formats excel in data visualization, multidimensional pattern recognition, and cartography. When complex statistical data is projected onto two- or three-dimensional spatial plots, the human visual system uses its retinotopic cortical architecture to instantly detect clusters, trend lines, outliers, and correlations. A cognitive system reading a spreadsheet of propositional data points must process the numbers sequentially and deductively; projecting those same data points into an analog Cartesian coordinate space allows human perception to parallel-process relationships across thousands of variables simultaneously.

6.3 Implications for Artificial Intelligence and Knowledge Engineering

The dialectic between propositional and spatial representations directly mirrors the historical and contemporary divisions within the field of Artificial Intelligence (AI). The early paradigm of AI—known as Classical AI, Symbolic AI, or Good Old-Fashioned AI (GOFAI)—was built entirely upon propositional representations. Systems such as the General Problem Solver, expert systems, and large-scale ontological knowledge graphs (e.g., Douglas Lenat’s Cyc project) modeled intelligence through formal symbolic token manipulation, production rules, and propositional predicate calculus. While these architectures demonstrated competence in constrained, logic-heavy domains (such as chess or medical diagnosis trees), they proved brittle when exposed to the continuous, noisy, physical realities of the visual and sensorimotor world.

In contrast, contemporary AI paradigms—including computer vision, deep learning, and spatial computing—rely heavily on analog and geometric representations. Modern systems map complex data into continuous, high-dimensional vector spaces (embeddings), where semantic relationships are represented as metric distances and directional vectors (e.g., using cosine similarity). In computer vision and robotic Simultaneous Localization and Mapping (SLAM), agents maintain continuous metric grids and spatial coordinate frames that directly instantiate the principles of Kosslyn’s visual buffer.

The cutting edge of artificial intelligence centers on the integration of these two paradigms: Neuro-symbolic AI and Multimodal Foundation Models. Modern large multimodal models (such as GPT-4o, Claude 3.5 Sonnet, or Gemini) bridge these architectures by coupling discrete language tokens with spatial coordinate patches and continuous latent vector spaces. These models translate written linguistic prompts into visual coordinate bounding boxes, convert visual input scenes into propositional semantic segmentations, and execute visual reasoning tasks. Moving beyond the limitations of purely symbolic or purely connectionist architectures requires artificial systems that synthesize the abstract reasoning of propositional networks with the intuitive geometric modeling of spatial representations.

7. Frequently Asked Questions

7.1 Did Stephen Kosslyn or Zenon Pylyshyn Ultimately Win the Imagery Debate?

The imagery debate did not conclude with a simple victory for either Stephen Kosslyn or Zenon Pylyshyn; instead, it evolved into an integrated synthesis that validated components of both positions. Kosslyn successfully established the biological reality and causal computational necessity of depictive representations. Modern neuroimaging (fMRI, PET) and neurostimulation (TMS) have demonstrated that the primary visual cortex (V1) and extrastriate areas are retinotopically activated and functionally necessary for visual mental imagery, disproving Pylyshyn’s radical assertion that depictive representations are purely epiphenomenal. Functional brain architectures clearly recruit metric, topographic coordinate sheets to execute analog spatial transformations.

However, Zenon Pylyshyn’s contributions profoundly reshaped cognitive science by exposing the dangers of the homunculus fallacy and demonstrating the pervasive influence of tacit knowledge. Pylyshyn proved that mental imagery is not an unfiltered internal video feed; it is deeply permeated and bounded by conceptual, propositional knowledge. When an individual imagines an object, they cannot visualize properties they do not conceptually possess, and their internal simulations are routinely altered by conscious expectations and verbal framing. Ultimately, modern cognitive science views Kosslyn and Pylyshyn not as combatants in a zero-sum game, but as theorists who mapped distinct levels of a dual-processing cognitive architecture: Pylyshyn delineated the amodal propositional control structure, while Kosslyn delineated the analog spatial simulation medium.

7.2 How Does Aphantasia Inform the Distinction Between Formats?

The phenomenon of congenital aphantasia—a neurocognitive condition first systematically characterized by Adam Zeman in 2015, wherein individuals possess no capacity for voluntary visual mental imagery—provides a natural experiment validating the functional independence of propositional and spatial representations. When asked to close their eyes and imagine a sunrise or a friend’s face, individuals with aphantasia report a complete absence of subjective visual phenomenology; their conscious mind is, in their own terms, entirely dark. Introspectively, their internal mental landscape is entirely non-depictive.

Remarkably, empirical studies have shown that individuals with aphantasia perform nearly as well as neurotypical controls on many spatial reasoning tasks, including mental rotation and spatial navigation. When tasked with determining whether two rotated three-dimensional objects are identical, aphantasics achieve accurate results, though they frequently employ non-visual, abstract, coordinate-based strategies or structural deductions rather than subjective visual simulation. This clinical dissociation confirms two critical theoretical principles: first, that conscious depictive visual phenomenology is not strictly identical to functional spatial computation; and second, that cognitive systems can bypass depictive visual imagery by executing spatial tasks through propositional tracking and topological coordinate manipulation.

7.3 Can Propositional and Spatial Formats Truly Coexist in the Same Mind?

Propositional and spatial formats not only coexist within the human mind; their ongoing coexistence is fundamental to human cognitive flexibility. Human cognition operates as a multi-tier representational system. While reading a schematic architectural diagram, the left perisylvian language regions process the alphanumeric labels and symbolic specifications (e.g., dimensions, materials, and safety codes) in a propositional format, while the right occipito-parietal areas construct an analog, spatial layout of the room configurations and physical flow.

This functional coexistence is sustained by continuous, bidirectional information exchange. When working memory is loaded with high volumes of verbal information, human subjects routinely offload cognitive load into the visual sketchpad by converting lists into spatial mental diagrams. Conversely, when an analog spatial representation becomes too dense or begins to degrade over time, the mind compresses the visual information into discrete propositional chunks (e.g., collapsing a complex visual shape into the categorical label “star-shaped”). This fluid translation demonstrates that human intelligence relies heavily on switching between symbolic precision and spatial simulation as task demands change.

7.4 How Does Modern Neuroimaging Distinguish Between the Two Systems?

Modern cognitive neuroscience employs high-resolution functional Magnetic Resonance Imaging (fMRI), magnetoencephalography (MEG), and event-related potentials (ERP) to empirically differentiate propositional from spatial representation based on distinct metabolic and electrophysiological signatures. When tasks recruit spatial representations—such as mentally scanning across an imagined map or rotating a geometric polygon—fMRI reveals retinotopically organized activations in early visual cortices (Brodmann Areas 17 and 18) and the intraparietal sulcus. Advanced multi-voxel pattern analysis (MVPA) algorithms can decode the physical, geometric contours and orientations of an imagined object directly from the spatial distribution of voxels in Area 17, demonstrating that the activation preserves the metric physical structure of the stimulus.

In contrast, when tasks engage propositional representations—such as categorical verification (“Is a robin an animal?”), syllogistic logic, or semantic comprehension—the early retinotopic cortices remain inactive, while blood-oxygen-level-dependent (BOLD) signals spike within the left inferior frontal gyrus (Broca’s area), the angular gyrus, and lateral temporal cortices. Chronometrically, MEG and ERP data show distinct temporal dynamics: semantic-propositional tasks reliably elicit language-linked components such as the N400 (reflecting semantic integration costs), whereas spatial transformations elicit slow parietal positive shifts whose temporal duration correlates directly with the metric degree of spatial rotation or scanning distance.

8. Conclusion and Future Horizons in Cognitive Science

8.1 Synthesis of the Dialectic: An Integrated Cognitive Framework

The protracted debate between propositional and spatial representations has propelled cognitive science past simplistic, reductionist accounts of the mind. Today, cognitive science rejects the extreme formulation of the propositional hypothesis—which dismissed the functional reality of spatial imagery as mere epiphenomenal fantasy—while simultaneously rejecting the naive pictorial view that treated mental images as literal, uninterpreted photographs displayed inside the brain. What has emerged is a sophisticated, integrated cognitive framework that recognizes both formats as vital, complementary subsystems of human cognition.

Human intelligence is uniquely characterized by its computational agility: the mind moves fluidly between discrete, amodal symbolic networks and continuous, analog spatial representations. Propositional representations provide the scaffolding for abstraction, logical quantification, counterfactual conditional reasoning, and general taxonomic categorization, freeing thought from the constraints of immediate physical perception. Simultaneously, spatial representations supply an embodied simulation engine that preserves the geometry, topology, and physical affordances of the real world, allowing organisms to interact intuitively with dynamic environments. Human thought achieves its unique power not through the dominance of one format over the other, but through their bidirectional integration.

8.2 Emerging Research Frontiers in Embodied and Artificial Cognition

As cognitive science advances into the 21st century, the dialectic between propositional and spatial representations continues to inspire novel research across embodied cognitive science, spatial computing, and artificial intelligence. The paradigm of embodied cognition has challenged the traditional supremacy of amodal propositional models, asserting that even abstract concepts are grounded in modal, sensorimotor, and spatial simulations. Conceptual metaphors—such as conceptualizing time as a spatial line (the past behind us, the future ahead) or social status as vertical elevation (high-ranking versus low-ranking)—suggest that the human brain repurposes its ancient evolutionary spatial and coordinate representations to structure abstract propositional thought.

Simultaneously, the rapid development of spatial computing, extended reality (XR), and geometric artificial intelligence is transforming how humans interact with externalized representations. Digital interfaces are shifting away from purely symbolic text and windows-based desktop metaphors toward immersive, metric spatial environments where digital data is organized spatially in three dimensions. In artificial intelligence research, the integration of geometric deep learning—which encodes symmetries, group theory, and topological manifolds—with discrete symbolic reasoning architectures represents the leading edge of machine learning. By unraveling how biological brains synthesize propositional codes with analog spatial models, cognitive scientists and computer engineers are constructing artificial architectures that approach the versatility, depth, and flexibility of human intelligence.

References

• Bisiach, E., & Luzzatti, C. (1978). Unilateral neglect of representational space. Cortex, 14(1), 129–133. https://doi.org/10.1016/S0010-9452(78)80016-1
• Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82(6), 407–428. https://doi.org/10.1037/0033-295X.82.6.407
• Fodor, J. A. (1975). The Language of Thought. Harvard University Press.
• Harnad, S. (1990). The symbol grounding problem. Physica D: Nonlinear Phenomena, 42(1–3), 335–346. https://doi.org/10.1016/0167-2789(90)90087-6
• Johnson-Laird, P. N. (1983). Mental Models: Towards a Cognitive Science of Language, Inference, and Consciousness. Harvard University Press.
• Kosslyn, S. M. (1980). Image and Mind. Harvard University Press.
• Kosslyn, S. M. (1994). Image and Brain: The Resolution of the Imagery Debate. MIT Press.
• Kosslyn, S. M., Ball, T. M., & Reiser, B. J. (1978). Visual images preserve metric spatial information: Evidence from studies of image scanning. Journal of Experimental Psychology: Human Perception and Performance, 4(1), 47–60. https://doi.org/10.1037/0096-1523.4.1.47
• Kosslyn, S. M., Pascual-Leone, A., Felician, O., Camposano, S., Keenan, J. P., Thompson, W. L., Ganis, G., Sukel, K. E., & Alpert, N. M. (1999). The role of area 17 in visual imagery: Convergent evidence from PET and rTMS. Science, 284(5411), 167–170. https://doi.org/10.1126/science.284.5411.167
• Paivio, A. (1971). Imagery and Verbal Processes. Holt, Rinehart & Winston.
• Paivio, A. (1986). Mental Representations: A Dual Coding Approach. Oxford University Press.
• Pylyshyn, Z. W. (1973). What the mind’s eye tells the mind’s brain: A critique of mental imagery. Psychological Bulletin, 80(1), 1–24. https://doi.org/10.1037/h0034650
• Pylyshyn, Z. W. (1981). The imagery debate: Analogue media versus tacit knowledge. Psychological Review, 88(1), 16–45. https://doi.org/10.1037/0033-295X.88.1.16
• Pylyshyn, Z. W. (2002). Mental imagery: In search of a theory. Behavioral and Brain Sciences, 25(2), 157–182. https://doi.org/10.1017/S0140525X02000043
• Shepard, R. N., & Metzler, J. (1971). Mental rotation of three-dimensional objects. Science, 171(3972), 701–703. https://doi.org/10.1126/science.171.3972.701
• Zeman, A., Dewar, M., & Della Sala, S. (2015). Lives without imagery – Congenital aphantasia. Cortex, 73, 378–380. https://doi.org/10.1016/j.cortex.2015.05.019