Dual-Coding Theory vs. Single-Code Theory

The nature of human cognition hinges on an enduring enigma: what is the fundamental currency of thought? When an individual recalls the face of a childhood mentor, plans a route through a labyrinthine metropolis, or contemplates an abstract ethical dilemma such as justice, what computational or biological substrate carries that mental content? For decades, cognitive science, philosophy of mind, and neurobiology have grappled with whether the mind operates as a universal, amodal symbol processor or whether it relies on distinct, modality-specific representational systems. This profound inquiry is not a mere semantic quarrel; it dictates how researchers model human memory, design artificial intelligence, structure educational curricula, and interpret neuroimaging data.

At the center of this intellectual battleground stand two monumental frameworks: Allan Paivio’s Dual-Coding Theory (DCT) and the Single-Code (Propositional or Amodal) Theory championed by scholars such as Zenon Pylyshyn, Jerry Fodor, and John R. Anderson. Dual-Coding Theory posits that the human mind processes and preserves information through two independent yet functionally linked subsystems: a verbal system specialized for arbitrary linguistic units, and a nonverbal imaginal system dedicated to sensory-rich, analog spatial representations. According to Paivio, our cognitive apparatus does not collapse the richness of our perceptual world into uniform equations; rather, it retains structural traces of the sensory modalities through which stimuli were originally experienced.

Conversely, the Single-Code model asserts that all human knowledge, regardless of its sensory origin or subjective introspective presentation, is codified within a singular, abstract, language-like medium of thought—frequently termed “mentalese” or propositional calculus. From this vantage point, subjective mental pictures are not functional machinery; they are epiphenomena, akin to the heat generated by an engine or the flickering pixels of a visual monitor that do not reveal the underlying binary code executing within the central processing unit. This comprehensive treatise investigates the deep historical roots, mathematical and formal foundations, empirical clashes, neurological correlates, and contemporary computational syntheses of these two titans of cognitive architecture.

1. Introduction: Foundations of the Mental Representation Debate

1.1 Historical Context of the Imagery Debate

The dawn of the twentieth century witnessed cognitive psychology emerge from the sterile behavioral orthodoxies of John B. Watson and B.F. Skinner. Behaviorism had systematically purged the lexicon of psychology of terms such as “mental image,” “introspection,” and “internal state,” relegating cognition to an impenetrable black box that merely mediated between environmental stimuli and observable behavioral responses. Watson famously dismissed mental imagery as a meaningless phantom of introspective psychology, asserting that thought was nothing more than sub-vocal speech carried out via minute muscular contractions of the larynx. However, by the late 1950s and 1960s, the cognitive revolution dismantled this dogma, invigorated by Claude Shannon’s information theory, Noam Chomsky’s devastating critique of Skinnerian verbal behavior, and the burgeoning field of computer science.

As cognitive psychologists reopened the black box, they encountered a foundational problem: how do internal representations exist within the mind? Pioneers sought to establish rigorous quantitative methodologies capable of probing internal mental dynamics without lapsing into unscientific introspection. The rediscovery of mental imagery as an empirical phenomenon brought this issue to the forefront. Alan Paivio began demonstrating through robust, replicable paired-associate learning experiments that the imageability of words was an extraordinarily potent predictor of mnemonic retention, far outstripping conventional behavioral variables.

Simultaneously, the birth of computational theory suggested that computers manipulate abstract, discrete symbols governed by formal syntactic rules. This mathematical view inspired a powerful school of thought arguing that the human mind is functionally analogous to a Turing machine. If the mind is an information-processing system instantiated in neural hardware, then thought must consist of computational operations executed upon symbolic strings. Consequently, the mid-twentieth century set the stage for a historic schism between analog spatial models—which treated mental images as quasi-perceptual, continuous spatial structures—and computational propositional models, which insisted that human intelligence is powered by universal, amodal predicate logic. This divergence polarized cognitive science into an intense debate, primarily embodied by the intellectual clash between Allan Paivio and Zenon Pylyshyn.

1.2 Defining the Core Problem: How the Human Mind Encodes Information

To grasp the crux of the representation controversy, one must clearly disentangle subjective phenomenal experience from underlying cognitive architecture. When an individual closes their eyes and imagines an apple, they introspectively experience a luminous, spherical, crimson object occupying a quasi-spatial arena behind their eyelids. The core scientific question, however, is whether this phenomenological impression reflects the true causal architecture of the human cognitive system, or whether it constitutes an illusion that conceals a completely different representational format.

The problem can be framed around cognitive persistence: when sensory input fades from the immediate sensory registers, does long-term memory retain the structural properties of sensory modalities, or does it transform every experience into an invariant, modality-neutral code? If human memory preserves modality-specific codes, then visual, auditory, and haptic modalities maintain distinct representational architectures, operational rules, and processing constraints within long-term storage. Under such an architecture, visualizing a rotating geometric object involves the manipulation of an analog, quasi-perceptual representation that preserves continuous spatial metrics, angles, and distances.

If, on the other hand, the mind is an amodal symbol system, all sensory inputs are immediately parsed, interpreted, and filed away as networks of abstract propositions. In this view, mental imagery is purely epiphenomenal—a subjective byproduct of underlying propositional computation that exerts no causal influence on cognitive processing. The debate over whether mental imagery is functional or epiphenomenal remains central to cognitive architecture and artificial intelligence. If human thought requires dual modal codes, artificial systems relying strictly on textual or symbolic tokens can never achieve authentic human-like understanding without grounding those tokens in sensory-analog architectures.

1.3 Rationale for Comparison and Common Misconceptions

A rigorous comparative analysis of Dual-Coding Theory and Single-Code Theory is essential because both paradigms have suffered from persistent misconceptions in academic literature, pedagogical practice, and popular science. The most pervasive mistake is confusing subjective introspective experience with empirical cognitive proof. Naive realists assume that because they “see” a mental picture, the brain must contain an actual internal picture. Single-code theorists correctly caution that equating introspective awareness with functional representation commits a homunculus fallacy: it implicitly presupposes an internal “mind’s eye” observing an internal movie screen, requiring an infinite regress of observers to explain the act of seeing.

A second common misconception is equating Dual-Coding Theory strictly with multimedia learning or educational instructional design. While Richard Mayer and other applied educational psychologists have derived profound instructional methodologies from Paivio’s work, Dual-Coding Theory is fundamentally an epistemological and cognitive architecture hypothesis concerning long-term memory representation, semantic organization, and structural encoding. It is not merely a pedagogical tip that students learn better when diagrams accompany text. Treating it as an instructional gimmick diminishes Paivio’s rigorous theoretical claims regarding functional parallelism, structural independence, and associative cross-activation.

Conversely, Single-Code Theory is widely misunderstood as being synonymous with linguistic, natural language representation. Critics often attack the propositional model by claiming that humans do not think solely in English, French, or Mandarin words. However, propositional theorists such as Pylyshyn, Fodor, and Anderson have consistently stressed that their proposed “mentalese” is an abstract, mathematical, language-like calculus of discrete symbols and truth values that is entirely independent of any natural spoken language. Propositions are amodal; they are neither auditory sound waves nor visual orthography. Clarifying these concepts is imperative before evaluating the structural mechanics of both theories.

2. Overview of Dual-Coding Theory (Paivio)

2.1 Theoretical Architecture: Logogens and Imagenes

Formulated comprehensively by Allan Paivio in his landmark works, Dual-Coding Theory operates on the foundational premise that human cognition is unique in having evolved two functionally and structurally distinct representational systems. These systems are separate yet dynamically interconnected. The verbal system specializes in dealing directly with linguistic inputs and outputs, preserving sequential, linear, and categorical information. The nonverbal (or imaginal) system specializes in handling non-linguistic objects, spatial relations, and multimodal events, preserving continuous, integrated, and synchronous holistic information.

To ground this architecture within discrete psychological units, Paivio introduced two primary concepts: logogens and imagenes. Logogens (a term adapted from John Morton’s work on word recognition) are defined within Dual-Coding Theory as the structural representational units of the verbal system. They represent verbal entities, linguistic tokens, sequential phonological structures, and syntactical sequences. When an individual hears the spoken word “lighthouse” or reads the word on a page, the corresponding verbal logogen is activated. Crucially, logogens operate sequentially; spoken language unfolds over time, where syntax and grammatical hierarchy dictate meaning through a linear temporal trajectory.

Conversely, imagenes are the fundamental structural units of the nonverbal imaginal system. Imagenes are not tiny photographs stamped onto cortical tissue; rather, they are modality-specific mental representations that preserve the perceptual and spatial properties of the objects, scenes, and sensory events they denote. Imagenes encompass visual appearances, auditory signatures (such as the distinct timbre of a cello or the roar of thunder), and haptic impressions (such as the texture of sandpaper). Unlike logogens, imagenes are organized synchronously and holistically. When an individual activates the imagene of a lighthouse, the architectural base, cylindrical tower, surrounding ocean waves, and radiant beacon can be apprehended simultaneously in a unified spatial matrix rather than as a temporal sequence of linguistic components.

The structural independence of these two subsystems means that one system can function autonomously without the active participation of the other. An individual can rehearse an abstract verbal formula or recite a poem purely through logogen-to-logogen transmission without generating nonverbal imagery. Similarly, an artisan can mentally sculpt or manipulate a complex geometric shape using pure imaginal processing without ever naming the components or translating them into linguistic tokens. Nevertheless, the true cognitive power of Dual-Coding Theory emerges from the functional interconnectivity between logogens and imagenes, which allows cross-activation and rich semantic retrieval.

2.2 Levels of Processing Within Dual-Coding

Dual-Coding Theory articulates a three-tiered hierarchy of processing that illustrates how information moves through and between the verbal and nonverbal systems. This taxonomy distinguishes between direct sensory impacts, cross-system translation, and intra-system associative branching:

• Representational Processing: This represents the initial, direct activation of the cognitive architecture by environmental stimuli. When a visual object (such as an actual canine) hits the retina, it directly activates its corresponding nonverbal structural unit—an imagene. Conversely, when a printed word or spoken acoustic waveform (“dog”) reaches the sensory apparatus, it directly stimulates its corresponding verbal unit—a logogen. This tier involves no cross-system translation; it is the immediate sensory-to-code mapping that initiates cognitive processing.
• Referential Processing: This critical level involves the cross-system activation linking verbal labels to nonverbal imagery and vice versa. Referential connections bridge the gap between the two distinct architectures. When the auditory logogen “elephant” triggers the mental visualization of a massive, gray mammal with a trunk (an imagene), referential processing has occurred. In the opposite direction, when a visual stimulus (seeing a sunset) prompts an individual to verbally label the phenomenon (“twilight” or “sunset”), the imaginal system has referentially activated the verbal system. This translation is asymmetric, variable in speed, and dependent upon an individual’s cognitive history and stimulus familiarity.
• Associative Processing: This tier involves intra-system activation, where units within a specific subsystem activate other units within the same subsystem without crossing the representational boundary. In the verbal logogenic subsystem, hearing the word “king” might associatively activate logogens for “queen,” “crown,” “monarchy,” or “palace” via linear semantic networks. In the nonverbal imaginal subsystem, visualizing a beach might spontaneously trigger associated imagenes of rolling waves, sand dunes, seagulls, or the physical sensation of heat. These dynamic associative chains permit elaborate daydreams or complex logical deductions to proceed uninterrupted within their respective modalities.

The fluidity of these levels of processing accounts for dynamic cognitive shifts. Depending on the cognitive task, environmental constraints, and intrinsic stimulus characteristics, human thought seamlessly alternates between verbal associative reasoning, imaginal spatial manipulation, and cross-modal referential translation.

2.3 Empirical Evidence and the Concreteness Effect

The foundational bedrock of Dual-Coding Theory lies in an extensive body of empirical research conducted over four decades, centered on the phenomenon known as the concreteness effect. Paivio and his contemporaries repeatedly demonstrated that human memory performance is dramatically superior for concrete nouns (e.g., “horse,” “hammer,” “guitar”) compared to abstract nouns (e.g., “justice,” “truth,” “obligation”). Dual-Coding Theory explains this disparity through the additive memory hypothesis.

When an individual is exposed to an abstract word like “epistemology,” that word is primarily encoded in a single representational format: the verbal logogenic system. It has no direct, universal physical referent, making it exceptionally difficult to spontaneously trigger a stable, concrete imagene. Therefore, its mnemonic trace relies upon a single code. In stark contrast, when a concrete noun such as “alligator” is presented, it inherently stimulates its verbal logogen while almost instantaneously triggering a referential nonverbal imagene. The concrete noun is stored in memory twice—once in an arbitrary verbal code, and once in an analog spatial-visual code. If the participant forgets the verbal label during a recall task, the imaginal representation remains intact to assist retrieval; if the image fades, the verbal trace preserves the memory. Mathematically, the probability of retrieving at least one of two independent codes is substantially higher than the probability of retrieving a single code.

This additive mechanism is demonstrated further by the picture superiority effect. In standardized recall tests, pictures of common objects are retained and recalled significantly better than the printed names of those same objects. Dual-Coding Theory explains that while printed words require cognitive effort and additional time to trigger referential imagery (and frequently fail to do so in rapid-fire experimental paradigms), visual pictures access the imaginal system automatically and immediately. Furthermore, because humans are culturally conditioned to verbally label their visual environment, observers spontaneously generate logogens for the presented pictures. As a result, pictures consistently prompt dual-coding, leading to robust memory traces.

Empirical validation is further strengthened by selective interference paradigms. Utilizing working memory models consistent with Alan Baddeley’s framework, dual-coding researchers demonstrated that processing two tasks within the same modality simultaneously causes severe cognitive degradation, whereas cross-modal dual-task performance remains remarkably stable. For instance, executing a spatial tracking task (such as pursuing a moving target on a screen) severely disrupts an individual’s ability to retain or manipulate visual mental imagery, yet it exerts minimal interference on simultaneous verbal rehearsal of word lists. These findings point to the existence of distinct, modality-specific cognitive architectures operating with separate pools of processing resources.

2.4 Criticisms and Theoretical Limitations of Dual-Coding

Despite its vast explanatory power and rich empirical support, Dual-Coding Theory has encountered substantial criticism from philosophers, cognitive scientists, and computational theorists. The most prominent philosophical objection revolves around cognitive parsimony and theoretical elegance. Critics invoking Occam’s razor argue that positing two entirely autonomous, highly complex cognitive systems—complete with their own independent structures, retrieval mechanics, and associative networks—introduces unnecessary redundancy into human cognitive architecture. Why would evolution craft two distinct internal languages when a single, universal information format could handle both linguistic categorization and spatial navigation?

A second persistent criticism centers on Dual-Coding Theory’s handling of abstract reasoning. While the theory comfortably accounts for physical nouns and sensory experiences, it struggles when modeling abstract concepts, mathematical operations, and formal philosophical logic. Concrete concepts possess obvious physical referents that seamlessly map onto visual or auditory imagenes, but what is the nonverbal imagene of “contradiction,” “hypothetical,” or “entropy”? Paivio argued that abstract words are grounded in idiosyncratic, situational imagery (e.g., imaging a courtroom scene to represent “justice”) or through extensive intra-verbal logogenic chains. However, critics argue this approach is ad hoc. Idiosyncratic imagery fails to capture the essential semantic core of abstract concepts; visualizing a gavel does not explain one’s comprehension of jurisprudence.

Finally, Single-Code proponents argue that Dual-Coding Theory under-specifies the precise computational mechanism of nonverbal processing. To simply assert that an imagene preserves “analog sensory information” does not explain the formal algorithmic processes through which the brain stores, indexes, searches, and transforms that information. Without precise computational definitions, imagery constructs risk degenerating into homunculus-driven mentalism, relying on internal subjective viewing rather than mechanistically sound information processing.

3. Overview of Single-Code Theory (Propositional / Amodal Model)

3.1 Theoretical Framework and Computational Roots

The Single-Code Theory, frequently designated as the Propositional Theory or Amodal Model, is grounded in the foundational developments of twentieth-century mathematical logic, predicate calculus, and symbolic artificial intelligence. Spearheaded by cognitive philosopher Zenon Pylyshyn, along with prominent cognitive scientists like Jerry Fodor, John R. Anderson, and Gordon Bower, this paradigm posits that the human mind does not possess multiple representational currencies. Instead, all information—whether derived from visual perception, spoken language, haptic exploration, or abstract deductive deduction—is translated and stored within a solitary, universal, abstract representational language: a language of thought.

Pylyshyn’s critique of mental imagery was launched in his seminal 1973 paper, “What the Mind’s Eye Tells the Mind’s Brain: A Critique of Mental Imagery.” Pylyshyn took direct aim at what he termed the “picture-in-the-head fallacy” and naive mental realism. He argued that the human brain is an information-processing computational engine, not an internal art gallery or stereopticon. Cortical tissue, consisting of complex arrangements of neurons, synapses, and action potentials, cannot literally hold pictures, continuous spatial canvases, or analog shapes. Pylyshyn contended that when individuals report seeing a mental image, they are mistaking their subjective phenomenological experience for the functional, computational architecture responsible for cognitive operations.

The single-code framework found powerful computational formalization in John R. Anderson’s ACT (Adaptive Control of Thought) and ACT-R cognitive architectures. Anderson demonstrated that human memory, pattern recognition, problem-solving, and linguistic performance could be modeled using purely propositional networks. In these computational frameworks, every piece of knowledge is represented as a network of abstract nodes interconnected by labeled, directional associations. These networks operate entirely via amodal symbolic manipulations, obviating any need for sensory-specific or analog formats to explain spatial reasoning, geometric problem solving, or visual memory recall.

3.2 Structure of Propositional Representations

To understand the mechanics of the Single-Code model, one must examine the precise anatomy of a proposition. In formal logic and cognitive science, a proposition is defined as the smallest unit of knowledge that can meaningfully stand as a distinct assertion, capable of being assigned a truth value (true or false). Propositions are not words, sentences, phonemes, or visual icons; they are abstract, formal relational structures typically represented using predicate calculus notation:

PREDICATE(Argument_1, Argument_2, …, Argument_n)

The predicate represents the underlying relationship, attribute, or operational state, while the arguments denote the entities, concepts, or subjects bound by that relationship. For instance, consider the sensory experience of observing a black cat resting atop a mahogany dining table. Dual-coding theory claims this experience is retained via a rich, nonverbal, quasi-spatial imagene alongside an associated linguistic logogen. Single-code theory, however, asserts that the brain extracts the informational content of this scene and encodes it as a bundle of abstract, interconnected propositions:

• ON(CAT, TABLE)
• COLOR_OF(CAT, BLACK)
• MATERIAL_OF(TABLE, MAHOGANY)

These propositions are entirely amodal. They are stripped of every sensory modality. The predicate ON does not possess spatial height, visual hue, lighting parameters, or physical weight; it is a mathematical assertion of a spatial-relational invariant. The argument CAT does not contain soft fur, whiskers, or an acoustic purr; it is an abstract semantic node that points to a wider ontological cluster within a vast semantic network. Whether an individual witnesses the cat on the table in broad daylight, reads the sentence “The black cat is on the mahogany table,” or feels the cat while blindfolded, the resulting internal long-term mental representation is identical: an amodal, propositional structure.

In this architecture, knowledge is structured hierarchically within vast semantic webs. Concepts are stored as interconnected nodes where meaning is derived strictly from their structural positions relative to other nodes. Retrieval does not involve re-playing an internal visual tape or scanning an internal image; rather, it proceeds through spreading activation. When an environmental cue activates a node, electrical and computational activation ripples outward along associative links. If the activation crossing adjacent nodes exceeds an internal activation threshold, the relevant propositional facts are integrated into active working memory, facilitating reasoning and behavioral output.

3.3 The Epiphenomenal Stance on Mental Imagery

If all cognitive content is stored and manipulated as propositional strings, why do humans possess such vivid, compelling introspective experiences of mental visualization? Single-code theorists address this paradox by relegating mental imagery to the status of an epiphenomenon. An epiphenomenon is a secondary phenomenon that occurs alongside a primary causal process, but plays no causal role in the mechanism itself. A classic computational analogy is the activity of a computer’s physical cooling fan: the fan spins and generates an audible hum as the central processor executes code, but the sound of the fan plays no computational role in solving the software’s mathematical equations. Similarly, the flickering pixels on a computer monitor present an image of a spreadsheet to a human user, but the computer itself calculates everything via internal binary code without looking at the monitor.

Pylyshyn famously argued that mental pictures are simply the phenomenological “read-out” of complex propositional reasoning. When subjects engage in classic imagery tasks—such as the mental rotation experiments of Roger Shepard and Jacqueline Metzler, or the mental scanning experiments of Stephen Kosslyn—their behavioral latencies can be explained without assuming that the brain stores or manipulates analog images. In Kosslyn’s famous island-scanning experiment, participants memorized a fictional map with landmarks (such as a hut, a well, and a tree) and were asked to mentally focus on one landmark, then “scan” across their mental image to another location. Kosslyn observed a linear relationship between the physical distance separating the landmarks on the original map and the time it took participants to mentally travel between them. Kosslyn asserted this proved that the brain uses an analog spatial canvas that preserves physical distance metrics.

Pylyshyn challenged this interpretation, arguing that these reaction-time latencies are driven by tacit knowledge and demand characteristics. Pylyshyn argued that human participants are consciously aware of how physical space functions in the real world: they know it takes longer to travel five miles than one mile. Consequently, when an experimenter instructs a participant to “scan” a mental image, the participant uses their tacit propositional knowledge about the laws of physics and the social expectations of the experiment to simulate the appropriate delay. Furthermore, Pylyshyn showed that when instructional cues are subtly altered to eliminate the command to “scan”—instructing participants simply to report whether a landmark exists—the linear relationship between distance and reaction time frequently disappears. Thus, the time delay is not caused by traversing an internal spatial canvas; it is an artifact of propositional rules, task expectations, and the computational complexity of retrieving intermediate propositional nodes.

3.4 Criticisms and Empirical Challenges to Single-Code Models

Despite the computational rigor of the amodal propositional model, Single-Code Theory faces severe empirical challenges that have prevented it from fully dislodging modality-specific frameworks. The most formidable challenge stems from the robust, highly replicable nature of analog spatial reasoning phenomena. In Shepard and Metzler’s iconic mental rotation experiments, participants were shown pairs of three-dimensional geometric figures rendered from differing perspectives and asked to judge whether they were identical or mirror images. The results were remarkably consistent: response time was a direct, linear function of the angular degree of rotation required to bring the two objects into alignment, regardless of whether the rotation occurred in the picture plane or in three-dimensional depth.

If representations were purely amodal propositions, such a linear relationship is difficult to justify. An amodal program can determine identity by translating the object into a coordinate list and running analytical transformations instantly, or the time complexity would scale with the structural complexity of the parts, rather than strictly tracking the physical angle of rotation. The fact that human minds must pass through intermediate stages of spatial orientation—taking twice as long to rotate an object 120 degrees as it does to rotate it 60 degrees—strongly suggests that the brain relies on an analog computational medium that mirrors continuous physical space.

Furthermore, Single-Code Theory encounters the notorious symbol grounding problem, articulated by cognitive scientist Stevan Harnad. If all concepts are simply abstract, amodal symbols pointing to other abstract symbols in a circular semantic web, how does this computational system ever connect with the external physical reality it is meant to represent? A dictionary in which words are defined purely using other words in an endless loop can never establish intrinsic meaning unless some subset of those symbols is directly grounded in sensorimotor interactions with the environment. By severing mental representations from their sensory-perceptual foundations, the Single-Code model creates an unbridgeable chasm between external sensory physics and internal computational logic.

Finally, modern neuroimaging technologies have presented evidence that directly challenges the single-code epiphenomenal stance. Decades of fMRI and PET studies, spearheaded by Stephen Kosslyn and confirmed by contemporary cognitive neuroscientists, have demonstrated that when individuals engage in high-resolution visual mental imagery, the primary visual cortex (Brodmann Area 17 / V1)—which contains retinotopic, spatially organized neural maps—is robustly activated. If mental imagery were a detached, amodal, propositional epiphenomenon, early perceptual cortices dedicated to processing sensory spatial arrays would not systematically participate in internally generated visual thought.

4. Key Differences Between Dual-Coding and Single-Code Theories

4.1 Representational Modality: Modal vs. Amodal Format

The foundational divergence between Dual-Coding Theory and Single-Code Theory lies in the ontological format of internal representations: modal versus amodal. Dual-Coding Theory is fundamentally modal. It claims that long-term mental representations retain structural, functional, and qualitative features directly tied to their originating sensory channels. The imagene for an eagle retains the continuous spatial geometry, color gradients, and visual topology captured by the retina; the logogen for a spoken phrase retains the rhythmic, phonological, and temporal contours captured by the cochlea. Sensory realism is treated as an authentic, foundational mental state that preserves the perceptual fidelity of human experience.

Conversely, Single-Code Theory asserts that cognitive architecture is strictly amodal. Regardless of whether an event is seen, heard, felt, or conceptualized, the cognitive apparatus discards its perceptual particulars, distilling the information into invariant, discrete, modality-independent symbols. In this framework, sensory details are merely the scaffolding used by sensory organs to extract semantic facts; once those facts are acquired, the scaffolding is discarded. Where Paivio views perceptual fidelity as an integral component of long-term storage, Pylyshyn views it as irrelevant noise that a computational mind must filter out to build clean, abstract, logical models of the environment.

4.2 Cognitive Architecture and Processing Mechanisms

This representational difference leads to radically divergent cognitive architectures. Dual-Coding Theory models the mind as a parallel-processing, dual-channel network. Information can flow simultaneously down two tracks: a verbal track optimized for sequential, serial, grammatical computation, and an imaginal track designed for synchronous, parallel, spatial-relational computation. This parallelism allows humans to process complex visual displays while simultaneously following spoken dialogue, effectively running two representational processors in tandem without fatal systemic interference.

Single-Code Theory, in contrast, models human cognition as a centralized, unified propositional network. Processing is driven by logic, deduction, and spreading activation across an interconnected web of categorical nodes. Spatial relationships are evaluated through the formal verification of logical links (e.g., assessing the validity of a path across nodes) rather than by inspecting a spatial field. Information retrieval is not driven by modality-specific visual or auditory cues, but by the mathematical distribution of activation throughout an amodal semantic matrix.

4.3 Cognitive Parsimony vs. Ecological Validity

The theoretical tension between these paradigms reflects a classic philosophical trade-off: Occam’s razor versus descriptive and ecological realism. Single-Code Theory prioritizes mathematical parsimony. By proposing a single representational medium—the amodal proposition—it unifies all cognitive phenomena under one operational umbrella. It avoids positing multiple representational storage formats, multiple retrieval engines, and complex cross-modal translation modules. It fits cleanly within classical computer science paradigms, where all programs and data reduce to a single binary substrate.

Dual-Coding Theory willingly sacrifices this formal parsimony to secure high ecological validity and descriptive fidelity. Human evolutionary history did not develop from a clean, single-core computer chip designed by an engineer; it arose through biological evolution, building complex layers over millions of years. Primates possessed highly evolved visual and spatial reasoning modules long before the late evolutionary arrival of human syntactic speech. Dual-coding theorists argue that it is biological nonsense to assume the brain threw away hundreds of millions of years of specialized sensory computing hardware simply to translate everything into an internal linguistic calculus. While DCT is architecturally more complex, its proponents argue that this complexity reflects the messy, multimodal biological reality of the human brain.

4.4 Neurological and Neuroimaging Correlates

From a neuroscientific perspective, Dual-Coding Theory finds natural support in the structural and functional lateralization of the cerebral cortex. Clinical neuropsychology has established that the human brain exhibits significant hemispherically specialized processing: the left cerebral hemisphere is typically dominant for sequential, linguistic, and syntactic operations, while the right hemisphere is dominant for holistic, visuospatial, and analog mapping tasks. When patients suffer localized left-hemisphere strokes, they frequently develop severe expressive or receptive aphasia while retaining their capacity to navigate complex spatial mazes, draw landscapes from memory, or recognize subtle geometric patterns. Conversely, severe right-hemisphere damage can precipitate profound spatial neglect and prosopagnosia without destroying the patient’s capacity to engage in sequential, abstract, verbal reasoning. These double dissociations provide strong structural evidence for two largely autonomous, modality-specific representational systems.

Single-code theorists, however, highlight the extensive involvement of the prefrontal cortex and the left inferior temporal lobe in serving as amodal semantic hubs. Modern cognitive neuroscience has identified regions, such as the anterior temporal lobes (often referred to as the brain’s “semantic hub” in the Hub-and-Spoke model of semantic memory), that respond uniformly to concepts regardless of whether they are presented as written words, spoken words, colored drawings, or natural sounds. These amodal hubs integrate diverse sensory experiences into high-level conceptual invariants, providing neurological backing for single-code propositional networks. While primary sensory cortices activate during imagery, single-code advocates maintain that this reflects feedback driven by anterior amodal networks retrieving knowledge, rather than proof that long-term storage is analog.

Comparative Dimension	Dual-Coding Theory (Paivio)	Single-Code / Propositional Theory (Pylyshyn, Anderson)
Core Representational Substrate	Dual: Modality-specific verbal units (logogens) and nonverbal sensory units (imagenes).	Single: Amodal, abstract, symbolic units structured as formal logic networks (propositions).
Representational Format	Modal: Retains continuous perceptual, analog spatial, phonological, and sensory metrics.	Amodal: Stripped of sensory characteristics; invariant language of thought (“mentalese”).
Status of Mental Imagery	Functional: Mental images are genuine cognitive representations that directly drive reasoning and memory.	Epiphenomenal: Mental images are byproducts or illusions accompanying propositional computation.
Processing Architecture	Parallel, dual-channel: Simultaneous visual-spatial processing and sequential verbal tracking.	Centralized, serial/network spreading activation: Deductive parsing of propositional webs.
Information Organization	Synchronous and holistic for imagenes; sequential and linear for logogens.	Hierarchical, relational networks based on predicate-argument relational structures.
Explanation of Memory Advantages	Additive Coding: Concrete stimuli are encoded simultaneously in both systems, doubling retrieval pathways.	Propositional Density: Concrete stimuli activate richer, more interconnected semantic networks.
Handling of Abstract Thought	Associative verbal chains supplemented by context-specific, idiosyncratic sensory metaphors.	Native strength: Naturally models abstract, logical, mathematical, and non-sensory truths.
Computational Roots	Perceptual systems, biological analog modeling, and evolutionary neuroanatomy.	Predicate calculus, mathematical logic, Turing computation, and symbolic AI.
Primary Vulnerability	High cognitive redundancy; difficulty formalizing non-metaphorical abstract propositions.	The symbol grounding problem; difficulty accounting for continuous mental rotation latencies.

When to Use Which Framework

Determining which theoretical paradigm to apply depends largely on the level of cognitive analysis, the specific domain of study, and the goals of the researcher or practitioner. Neither model holds a total monopoly on cognitive explanation; rather, each excels within specific contexts:

Apply Dual-Coding Theory when:

• Designing Educational Materials and User Interfaces: When the objective is to optimize human learning, reduce working memory bottlenecks, and maximize long-term information retention, Dual-Coding provides the most actionable, ecologically valid guidelines. It clearly delineates how visual diagrams and auditory narration should be synchronized to prevent cognitive overload.
• Investigating Working Memory and Perceptual Attention: When modeling the real-time execution of split-attention tasks, spatial tracking, or cross-modal sensory interference (such as navigating while listening to verbal directions), DCT’s dual-channel assumptions accurately predict human behavioral limits.
• Designing Clinical Neuropsychology Interventions: When developing cognitive rehabilitation programs for stroke patients, aphasics, or individuals with traumatic brain injuries, utilizing dual-coding techniques—such as tapping intact nonverbal imagery to compensate for damaged verbal circuits—yields highly effective, practical therapeutic strategies.

Apply Single-Code (Propositional) Theory when:

• Developing Symbolic AI, Knowledge Graphs, and Ontologies: When building automated reasoning systems, formal expert architectures, semantic search engines, or database schema where unambiguous truth-values, logical consistency, and formal inferences are required, the propositional framework remains the gold standard.
• Analyzing Complex Deductive Reasoning and Formal Logic: When modeling how the human mind resolves syllogisms, navigates counterfactual scenarios, verifies grammatical syntax, or manipulates abstract mathematical equations that have no physical or perceptual reality, the amodal propositional calculus provides the necessary formal machinery.
• Modeling Unified Long-Term Semantic Storage: When attempting to formulate a parsimonious, computationally implementable architecture of universal long-term memory that can explain how diverse sensory inputs are synthesized into invariant, non-redundant conceptual knowledge, amodal models offer significant theoretical advantages.

5. Similarities and Theoretical Overlap

5.1 Shared Explanatory Goals in Cognitive Psychology

Despite their long-standing theoretical opposition, Dual-Coding Theory and Single-Code Theory are engaged in the same foundational enterprise: constructing a scientifically robust, mechanistic account of human internal mental representation. Both movements arose as definitive rejections of behaviorist reductionism. Both paradigms fundamentally affirm that the human mind relies on rich, highly structured internal mental states, and that human behavior cannot be explained by environmental stimulus-response pairings alone.

Furthermore, both theories are built to account for the same classic semantic memory phenomena. Both frameworks successfully predict and model semantic priming—the empirical finding that exposing a participant to a concept (e.g., “doctor”) accelerates their subsequent cognitive processing of a related concept (e.g., “nurse”). Dual-coding explains this via associative spreading within the logogenic network, while single-code theory explains it via spreading activation across propositional nodes. Both frameworks accommodate human associative memory, categorizing ability, and the capacity to reason about absent objects, past events, and hypothetical futures. They simply differ on the format of the representational medium executing these operations.

5.2 Convergence in Computational and Hybrid Implementations

As cognitive science has matured, the rigid ideological boundaries separating modal and amodal purists have increasingly blurred. Modern cognitive scientists recognize that the human mind does not operate exclusively through free-floating predicate calculus, nor does it rely solely on raw sensory replays. Contemporary computational architectures often embrace hybrid designs that reconcile both perspectives.

In contemporary artificial intelligence, this convergence is clearly evident in multimodal deep learning models. Architectures like OpenAI’s CLIP (Contrastive Language-Image Pre-training) explicitly implement dual-stream networks. CLIP utilizes a visual transformer to process raw pixels into a visual vector space, and an independent text transformer to process linguistic strings into a text vector space. These two distinct streams are then mapped onto a shared, cross-modal mathematical latent space where the similarity between a picture and a descriptive sentence is calculated. This architecture is an engineering realization of Paivio’s dual subsystems linked via referential connections, implemented using modern matrix calculus.

Similarly, the theoretical paradigm of embodied cognition (or grounded cognition), pioneered by Lawrence Barsalou, directly bridges this divide. Barsalou’s theory of Perceptual Symbol Systems (PSS) challenges both traditional propositional theories and classic dual-coding theory by proposing that all abstract conceptual thought is executed via multimodal neural simulations. When an individual contemplates an abstract proposition, the brain runs perceptual-motor simulations grounded in sensory cortices, using modal representations to execute logical, propositional tasks. In this hybrid model, propositions are not disembodied, amodal strings; they are re-enactments of sensory states organized into compositional, productive structures. Through frameworks like PSS, modern cognitive science increasingly views sensory-rich modal representations and structured logical relations not as mutually exclusive rivals, but as complementary dimensions of an integrated cognitive engine.

6. Practical Applications and Applied Contexts

6.1 Educational Practice and Multimedia Learning Design

The practical application of Dual-Coding Theory has had a transformative impact on educational psychology, primarily through the influential work of Richard E. Mayer and his Cognitive Theory of Multimedia Learning (CTML). Grounded in Paivio’s dual-channel premise and Baddeley’s model of working memory, Mayer’s framework addresses a fundamental reality of the human classroom: human working memory is severely limited in its capacity to process novel information. If an educator presents information using only a single channel—for example, projecting slides dense with written text while lecturing verbally—the student’s verbal channel (the phonological loop and logogenic system) experiences rapid cognitive overload, precipitating severe information loss.

By applying dual-coding principles, instructional designers can bypass this cognitive bottleneck. When verbal explanations are delivered as spoken narration alongside dynamic, complementary visual animations or diagrams, the cognitive burden is distributed evenly across both the verbal logogenic channel and the nonverbal imaginal channel. This strategic distribution prevents either channel from becoming exhausted, enabling learners to build richer mental models. Educational research has repeatedly validated this approach, confirming that students who learn from properly balanced multimedia presentations exhibit significantly higher conceptual transfer and retention compared to those taught through unimodal instruction.

However, successful instructional design also relies on understanding where Single-Code principles govern conceptual mastery. If instructional visuals are decorative rather than explanatory, they introduce extraneous cognitive load. When teaching highly abstract logical systems—such as legal philosophy, software programming syntax, or formal propositional logic—relying excessively on visual metaphors can lead to severe misconceptions. In these domains, the clear, unambiguous hierarchical structuring of propositional rules, logical syllogisms, and explicit relational definitions is superior. The skilled instructional designer must constantly assess whether the target knowledge consists of concrete spatial mechanisms (best served by Dual-Coding) or abstract formal relationships (best served by structured Propositional scaffolding).

Furthermore, applying these theories helps avoid common instructional errors such as the split-attention effect and the redundancy effect. The split-attention effect occurs when students are forced to divide their visual attention between two disparate visual sources—such as a diagram and a separate block of explanatory text placed at the bottom of the page. Because both inputs compete for the same visual processing channel, learning is impaired. The solution, derived from dual-coding, is to convert the written text into auditory narration, routing it through the verbal channel while the eyes remain fixed on the visual diagram. Alternatively, if text must remain visual, it should be integrated physically directly next to the corresponding parts of the diagram, minimizing visual search costs. Conversely, the redundancy effect occurs when an instructor reads verbatim text that is already displayed on a slide, overloading the verbal processor with duplicate streams of identical linguistic tokens. By adhering to dual-coding mechanics, educational design transforms from intuitive guesswork into a rigorous, scientifically grounded practice.

6.2 Artificial Intelligence and Machine Knowledge Representation

The ideological clash between Dual-Coding and Single-Code theories has deeply shaped the evolution of artificial intelligence. In early classical symbolic AI (GOFAI, or “Good Old-Fashioned Artificial Intelligence”), the field was entirely dominated by Single-Code philosophy. Inspired by the propositional frameworks of Anderson, Newell, and Simon, researchers attempted to build artificial general intelligence through massive networks of formal logic, expert systems, and symbolic ontologies. Projects like CYC spent decades cataloging millions of everyday facts into machine-readable propositional calculus, operating on the assumption that if an AI system possessed an exhaustive propositional knowledge graph of the world, human-level intelligence would naturally emerge.

While symbolic AI achieved success in narrow, highly formalized domains like chess and rule-based diagnostic systems, it ultimately stalled when confronted with the messy, continuous, ambiguous physical world. Symbolic systems struggled profoundly with computer vision, natural speech processing, and spatial navigation. They suffered directly from Harnad’s symbol grounding problem: their propositional symbols had no internal connection to sensory reality. An expert system could evaluate the truth-value of the proposition IS_DANGEROUS(FIRE), but it had no perceptual apparatus to recognize fire from a camera feed or understand the physical reality of thermal damage.

The modern revolution in artificial intelligence—driven by deep learning and multimodal foundational architectures—represents a dramatic shift toward a computational realization of Dual-Coding principles. Systems like CLIP, DALL-E, and large multimodal models (LMMs) move beyond single-modality processing. These models construct high-dimensional continuous vector embeddings that align textual descriptions with visual patterns in a shared latent space. In this architecture, the text encoder acts as a computational analog to the logogenic system, while the visual vision transformer operates as an analog to the imaginal system. The model achieves zero-shot image recognition and coherent image generation precisely because it cross-activates representations across these two distinct streams via shared mathematical projections—mirroring Paivio’s referential processing.

At the same time, the debate continues within the AI research community concerning the true nature of Large Language Models (LLMs) such as GPT-4. Scholars are locked in a debate: do these systems merely capture complex surface-level linguistic statistics—functioning as hyper-scaled amodal propositional processors—or do their billions of attention weights construct internal, emergent “world models” that simulate physical, spatial, and causal realities akin to mental imagery? As researchers strive to build autonomous robotic agents that must manipulate real-world physical objects based on verbal commands, reconciling propositional symbolic planning with continuous spatial-perceptual world models has emerged as the central technical challenge of modern machine intelligence.

6.3 Clinical Neuropsychology and Cognitive Rehabilitation

In the clinical domain, the practical distinction between modal and amodal representations carries life-altering consequences for patients suffering from neurological trauma, neurodegenerative diseases, and psychological disorders. Clinical neuropsychologists have extensively utilized Dual-Coding Theory to devise cognitive rehabilitation programs for patients suffering from aphasia resulting from left-hemisphere cerebrovascular accidents. When a stroke damages the left perisylvian language regions, the patient’s logogenic verbal network is severely impaired, leaving them struggling to access spoken words or comprehend complex syntax.

Rehabilitation therapists systematically leverage the structural independence of the nonverbal imaginal system—often preserved in the undamaged right hemisphere—to construct compensatory cognitive pathways. Through therapies like Visual Action Therapy (VAT) and the use of visual imagery mnemonics, patients are trained to encode essential everyday concepts, needs, and objects using vivid, interactive mental imagery and spatial gestures. By pairing visual imagenes with residual, damaged logogens, referential cross-activation can gradually rehabilitate lexical retrieval, or provide a functional communication system that bypasses the damaged linguistic channel entirely.

Conversely, insights from Single-Code Theory are critical when treating patients with frontal lobe injuries, traumatic brain injury (TBI), or executive dysfunction who struggle with abstract categorical reasoning, rule-shifting, and deductive planning. Patients who become trapped in concrete thinking—unable to generalize past the immediate visual stimuli in front of them—benefit from structured propositional retraining. Cognitive rehabilitation protocols systematically drill abstract logic trees, hierarchical categorization tasks, and predicate-argument relation identification, rebuilding the patient’s capacity to formulate domain-general rules independent of concrete visual distractions.

The imagery debate is also vital for understanding conditions such as aphantasia—a neurological and cognitive condition where individuals lack the capacity to voluntarily generate visual mental imagery. Despite possessing entirely normal visual perception and scoring normally on standard intelligence tests, individuals with congenital aphantasia report complete darkness in their “mind’s eye.” Neuropsychological assessments reveal that aphantasics frequently perform surprisingly well on classic spatial memory and mental rotation tasks, although they solve these problems using different computational strategies. Rather than relying on continuous analog visuospatial manipulation, aphantasics appear to depend almost exclusively on amodal spatial coordinates, propositional rules, and structural verbal logic. This demonstrates that human cognition can successfully navigate spatial challenges using alternative representational codes.

Finally, in the realm of psychiatry and clinical psychotherapy, Dual-Coding Theory provides a foundational framework for understanding and treating Post-Traumatic Stress Disorder (PTSD). Traumatic memories are frequently stored not as coherent, sequential verbal narratives, but as intrusive, sensory-rich nonverbal imagenes—visceral auditory sounds, terrifying visual scenes, and intense somatic feelings that trigger instantaneous autonomic panic responses. Dual-coding explains that these traumatic memories have failed to become integrated into the verbal logogenic network; they exist as uncoupled, raw, sensory imagenes that intrude into consciousness without narrative context.

Therapeutic interventions such as Prolonged Exposure (PE) therapy, Cognitive Processing Therapy (CPT), and Eye Movement Desensitization and Reprocessing (EMDR) work systematically to bridge this gap. By guiding the patient to verbally articulate the traumatic sensory experience—translating the overwhelming, nonverbal imagenes into sequential linguistic logogens—the trauma is gradually embedded into an autobiographical narrative context. This referential cross-activation allows the brain’s prefrontal cognitive control networks to regulate the hyperactive limbic responses generated by isolated imaginal memory traces, transforming an overwhelming flash-forward into a manageable, contextualized memory of the past.

7. Frequently Asked Questions

7.1 Is mental imagery real or merely an epiphenomenon?

The consensus in contemporary cognitive neuroscience is that mental imagery is a functionally real, causally efficacious component of the human cognitive architecture. It is not an epiphenomenon. The historical argument mounted by Zenon Pylyshyn—which claimed that the subjective experience of mental pictures is an illusion masking amodal propositional logic—has been systematically tested by modern neuroimaging and causal brain-stimulation methodologies. Experiments using functional Magnetic Resonance Imaging (fMRI) have shown that visual mental imagery tasks consistently activate retinotopically mapped primary and secondary visual cortices (Brodmann Areas 17, 18, and 19). When an individual visualizes a small geometric circle, the central portion of the visual cortex lights up; when they imagine a large circle, the neural activation expands outward into the peripheral representations of the visual field, precisely mirroring the neural dynamics of real visual perception.

Furthermore, studies utilizing Repetitive Transcranial Magnetic Stimulation (rTMS) have established a clear causal connection between this perceptual neural activation and behavioral task performance. When researchers apply focal magnetic pulses to temporarily disrupt the functional activity of the primary visual cortex, participants exhibit immediate, significant increases in reaction time and error rates on mental rotation, high-resolution imagery inspection, and spatial scanning tasks. If mental imagery were a detached epiphenomenal read-out, disrupting the primary visual cortex would have no impact on the execution of these tasks. While Pylyshyn’s critique remains foundational in warning cognitive scientists against naive homunculus models, the evidence confirms that the brain relies on its perceptual hardware to execute spatial and imaginal thinking.

7.2 How does Cognitive Load Theory relate to Dual-Coding and Single-Code models?

John Sweller’s Cognitive Load Theory (CLT) is rooted in the architecture of working memory, heavily incorporating the core principles of Dual-Coding Theory. CLT asserts that human working memory possesses a very limited capacity when handling novel information, but long-term memory maintains virtually unlimited storage in the form of complex cognitive schemas. To explain how information moves effectively from transient working registers into permanent long-term storage, CLT relies on Alan Baddeley’s multi-component working memory framework—specifically the division between the phonological loop (which corresponds directly to Paivio’s verbal logogenic channel) and the visuospatial sketchpad (which mirrors Paivio’s nonverbal imaginal channel).

Cognitive Load Theory posits that cognitive overload occurs when either of these modality-specific working memory channels is burdened with excessive extraneous load. By applying dual-coding principles, instructional materials can be structured to split the instructional burden across both channels—presenting visual information via the visuospatial sketchpad while presenting verbal context through the auditory phonological loop. However, Single-Code theory becomes essential when modeling the ultimate destination of this information: schema acquisition within long-term memory. Once information has been processed through these dual modal channels, it is organized into hierarchical, amodal conceptual schemas. These schemas integrate propositional rules with multimodal sensory traces, allowing the mind to access consolidated chunks of information with minimal working memory load during subsequent problem-solving tasks.

7.3 Can neuroimaging definitively prove one theory over the other?

While neuroimaging has delivered substantial empirical evidence favoring modal, sensory-based representations during active imagery tasks, neuroimaging alone cannot decisively resolve the foundational debate between Dual-Coding and Single-Code theories. The inherent methodological challenge lies in the interpretive gap between neural activation and representational format. When an fMRI scanner reveals that the primary visual cortex or the parahippocampal place area activates during an imagery task, this demonstrates that these anatomical regions are engaged. However, it does not definitively reveal the mathematical or computational code running within those neural circuits.

A committed single-code theorist can argue that the observed visual cortex activation is merely top-down sensory reconstruction driven by underlying amodal, propositional computations originating in the prefrontal cortex and anterior temporal lobes. From this perspective, the visual cortex serves as a transient, downstream display monitor rather than the primary representational format of the knowledge itself. Furthermore, neuroimaging paradigms risk circular reasoning: researchers often interpret neural activations through the lens of whichever psychological theory they already accept. Modern neuroscientists increasingly recognize that resolving this debate requires more than static brain scans; it demands sophisticated computational models, such as multivariate pattern analysis (MVPA) and deep neural network decoding, capable of directly examining the representational geometries embedded within neural population codes.

7.4 Which theory better explains abstract concept comprehension?

Single-Code Theory holds a structural advantage when modeling the comprehension and manipulation of purely abstract, non-physical concepts. Concepts such as “justice,” “counterfactual,” “syllogism,” “infinitive,” and “probability” have no direct, physical, or sensory referents. Single-code theory models these concepts through predicate calculus, treating them as abstract relational structures defined by their logical operations and connections to other symbolic concepts within a semantic web. This formal approach accounts for human reasoning in advanced mathematics, theoretical physics, and formal philosophical logic without needing to force physical sensory characteristics onto abstract ideas.

Dual-Coding Theory attempts to explain abstract concept comprehension by claiming that abstract logogens are grounded through situational, episodic contexts and idiosyncratic imagery—such as an individual visualizing a wooden gavel or a balance scale when contemplating “justice.” Alternatively, Paivio points to George Lakoff and Mark Johnson’s Conceptual Metaphor Theory, arguing that human minds comprehend abstract ideas by projecting concrete, embodied spatial schemas onto abstract domains (e.g., conceptualizing time through spatial metaphors, such as “looking forward” to the future or “leaving the past behind”). While this metaphor-grounding hypothesis explains how concrete experiences enrich human everyday language, it encounters limitations when applied to advanced, non-metaphorical theoretical logic. Consequently, cognitive scientists widely conclude that a comprehensive model of human thought requires both formats: Dual-Coding captures our grounded, sensory-metaphorical understanding, while Single-Code mechanisms handle our capacity for formal, amodal, symbolic computation.

8. Conclusion: Toward a Unified Model of Cognitive Representation

8.1 Synthesis of Decades of Theoretical Debate

The historic debate between Allan Paivio’s Dual-Coding Theory and the Single-Code Propositional Theory represents one of the most intellectually fertile controversies in the history of cognitive science. What began in the late 1960s as a polarized ideological battle between analog mental pictures and computational symbolic code has forced cognitive psychology, philosophy of mind, and neurobiology to develop increasingly sophisticated models of internal mental representation. The debate dismantled naive mental realism, pushed behaviorism permanently to the margins, and drove the creation of quantitative experimental paradigms capable of probing the internal architecture of human memory.

Looking back across more than five decades of theoretical discourse and empirical investigation, it is clear that the reality of human cognition does not conform strictly to either extreme. Paivio’s Dual-Coding Theory was fundamentally correct in its assertion that the human mind retains modality-specific representational architectures. The brain does not collapse its hard-won perceptual-spatial evolution into uniform equations of abstract text; long-term memory retains modal structures, and mental imagery functions as a genuine computational medium that shapes reasoning, learning, and recall. The persistent robustness of the concreteness effect, the picture superiority effect, cross-modal interference limits, and the retinotopic activation of early visual cortices all validate the core premises of Dual-Coding.

Concurrently, the Single-Code theorists were equally justified in their mathematical critiques. Zenon Pylyshyn, Jerry Fodor, and John Anderson correctly identified the profound logical traps of the “picture-in-the-head” fallacy, forcing the scientific community to confront the symbol grounding problem and recognize that human thought requires formal, relational, compositional mechanisms to execute abstract reasoning, evaluate truth-values, and navigate syntax. Introspective experience cannot be uncritically accepted as architectural proof; the mind possesses high-level, amodal, propositional capacities that transcend immediate sensory particulars.

The contemporary trajectory of cognitive science, neuroscience, and artificial intelligence points away from a rigid dichotomy and toward unified, multi-level computational frameworks. Models such as Lawrence Barsalou’s Perceptual Symbol Systems, hybrid cognitive architectures like ACT-R (which now integrates perceptual-motor buffers alongside central production systems), and multimodal foundational artificial intelligence models show that the mind is a hybrid information-processing system. Human thought is simultaneously grounded in modal, sensory-motor simulations and capable of operating through abstract, compositional, propositional structures. By moving past the historical opposition between Dual-Coding and Single-Code theories, modern science continues to unravel the true computational architecture of the human mind: a system capable of transforming raw sensory experience into enduring, abstract wisdom.

References

• Anderson, J. R. (1983). The architecture of cognition. Harvard University Press.
• Anderson, J. R., & Bower, G. H. (1973). Human associative memory. Winston & Sons.
• Baddeley, A. D. (1986). Working memory. Oxford University Press.
• Barsalou, L. W. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22(4), 577–660. https://doi.org/10.1017/s0140525x99002149
• 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
• Kosslyn, S. M. (1980). Image and mind. Harvard University Press.
• Kosslyn, S. M., Ganis, G., & Thompson, W. L. (2001). Neural foundations of imagery. Nature Reviews Neuroscience, 2(9), 635–642. https://doi.org/10.1038/35090055
• Lakoff, G., & Johnson, M. (1980). Metaphors we live by. University of Chicago Press.
• Mayer, R. E. (2009). Multimedia learning (2nd ed.). Cambridge University Press. https://doi.org/10.1017/CBO9780511811678
• Paivio, A. (1971). Imagery and verbal processes. Holt, Rinehart and Winston.
• Paivio, A. (1986). Mental representations: A dual coding approach. Oxford University Press.
• Paivio, A. (1991). Dual coding theory: Retrospect and current status. Canadian Journal of Psychology, 45(3), 255–287. https://doi.org/10.1037/h0084295
• 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
• Rademaker, R. L., & Pearson, J. (2012). Training visual imagery: Improvements of imagination, perception, and basic cognitive functions. Journal of Cognitive Neuroscience, 24(7), 1580–1589. https://doi.org/10.1162/jocn_a_00232
• 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
• Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285. https://doi.org/10.1207/s15516709cog1202_4