The Number Conservation and Ordinality in Monkeys – Elizabeth Brannon and Herbert Terrace

The philosophical and scientific inquiry into whether mathematical competence is unique to human language has occupied cognitive science for over a century. For decades, the dominant theoretical paradigms in comparative psychology maintained that true numerical understanding—the capacity to conceptualize abstract cardinal values and their precise ordinal sequences—demanded symbolic representation. Non-human animals were widely presumed to operate strictly within the boundaries of associative conditioning, reacting to continuous sensory dimensions such as cumulative surface area, spatial density, or perceptual duration rather than discrete number. Where mathematical behavior appeared to manifest in non-human subjects, it was routinely dismissed as an artifact of sub-cognitive heuristics, low-level sensory biases, or unconscious cues provided by human experimenters.

This long-standing skepticism was fundamentally overturned by a groundbreaking series of experiments conceived and executed by Elizabeth M. Brannon and Herbert S. Terrace at Columbia University during the late 1990s and early 2000s. Operating at the intersection of psychophysics, operant conditioning, and cognitive ethology, Brannon and Terrace constructed an experimental methodology that dissociated discrete numerosity from confounding physical variables. Their empirical investigations demonstrated that rhesus macaques (Macaca mulatta) do not merely discriminate quantities; they comprehend numerical ordinality across both trained and novel sets. This established that the capacity to organize discrete visual numerosities along an internal, ascending mental continuum exists independently of language, formal education, and symbolic scaffolding.

The implications of Brannon and Terrace’s work extend far beyond comparative animal behavior. By demonstrating that non-human primates possess an endogenous representation of ordinal relations and exhibit formal hallmarks of number conservation, their research reshaped our understanding of the phylogenetic architecture underpinning human mathematical thought. Their findings provided critical empirical support for the existence of an evolutionarily conserved Approximate Number System (ANS), bridged the divide between Jean Piaget’s developmental constructivism and modern cognitive neurobiology, and laid the groundwork for contemporary neurophysiological investigations into the cortical representation of quantity within the primate brain.

1. Introduction to Numerical Cognition in Non-Human Primates: The Brannon and Terrace Landmark Studies

1.1 Historical Context of Animal Numerical Competence

The scientific study of animal numerical abilities was historically haunted by the cautionary tale of Clever Hans, an Orlov Trotter horse heralded in early twentieth-century Germany for his apparent ability to perform complex arithmetic. When his owner, Wilhelm von Osten, presented mathematical questions, Hans would tap his hoof to indicate the correct sum, difference, or product. However, rigorous experimental investigation conducted by psychologist Oskar Pfungst in 1907 revealed that Hans was not performing mental calculations. Instead, the animal was an extraordinarily perceptive observer of involuntary micro-cues: he detected microscopic postural shifts, changes in respiratory cadence, and subtle facial tensions exhibited by human onlookers as his hoof taps neared the correct solution. The Clever Hans phenomenon cast a long shadow over comparative cognition, establishing a pervasive scientific skepticism toward any claims of animal mathematical comprehension. For most of the twentieth century, any behavioral manifestation of numerical sensitivity in non-human subjects was routinely attributed to low-level associative conditioning or unexamined sensory artifacts.

Despite this climate of skepticism, mid-century ethologists and comparative psychologists began designing controlled paradigms to determine whether animals could respond to pure numerical stimuli. Early investigations focused primarily on relative quantity judgments—such as training birds or mammals to choose the larger of two grain piles—and the perceptual phenomenon of subitizing, the rapid, pre-attentive apprehension of small quantities (typically one to four items). Researchers like Otto Koehler used meticulously controlled photographic targets to test whether birds could match visual patterns based solely on element count. However, these early experiments faced significant methodological limitations. It was technically difficult to decouple the discrete number of visual elements from correlated continuous physical variables such as total cumulative surface area, spatial envelope, aggregate luminance, or edge density. As a result, critics argued that subjects were relying on sensory shortcuts rather than true numerical abstraction.

By the late twentieth century, the paradigm began shifting from behaviorist stimulus-response accounts toward cognitive formulations of animal mental representations. Groundbreaking work by David Premack with chimpanzees, Irene Pepperberg with African grey parrots, and Euan Macphail in comparative neurology suggested that animals might form mental representations that transcend immediate sensory inputs. At Columbia University, Herbert Terrace established an experimental laboratory dedicated to exploring the cognitive architectures of non-verbal organisms. Having previously challenged the linguistic capacities of chimpanzees in the famous Project Nim study, Terrace turned his focus toward non-linguistic structural cognition. Terrace hypothesized that while animals lack syntactic language, they might possess sophisticated cognitive faculties for organizing temporal, spatial, and numerical relationships. It was within this rigorous cognitive testing environment that Elizabeth Brannon arrived to address the challenge of non-human numerical representation.

1.2 The Seminal 1998 and 2000 Publications

The collaborative work of Elizabeth Brannon and Herbert Terrace culminated in two foundational papers that redefined the field of numerical cognition. In October 1998, they published a landmark paper in the journal Science titled “Ordering of the Numerosities 1 to 9 by Monkeys”. This brief yet transformative report provided the first empirical proof that non-human primates could master an ordinal rule across a limited set of numerosities (1, 2, 3, and 4) and spontaneously generalize that abstract relational rule to entirely novel numerosities (5, 6, 7, 8, and 9) on their very first trial presentations. The study bypassed the historic limitations of animal counting literature by demonstrating that rhesus monkeys do not simply learn static associative chains; they represent quantities as an open-ended, sequentially ordered mental dimension.

Following the immediate scientific impact of the 1998 report, Brannon and Terrace published an extensive, methodologically exhaustive monograph in the Journal of Experimental Psychology: Animal Behavior Processes in 2000, titled “Representation of the Ordinal Relations of Numerosities 1–9 by Rhesus Monkeys (Macaca mulatta).” This 2000 expansion provided an in-depth breakdown of the psychophysical profiles of their subjects, detailed error analyses, and expanded upon the stimulus-generation methodologies that eliminated continuous sensory confounds. Across both publications, the central thesis remained consistent: non-human primates can construct, store, and manipulate internal mental representations of numerical ordinality without any linguistic training, symbolic mediation, or direct social prompting.

A major factor in the success and enduring scientific validity of these studies was the experimental design formulated by Elizabeth Brannon. Recognizing that previous animal numerical studies had been undermined by uncontrolled perceptual variables, Brannon implemented an array of continuous-variable controls. She recognized that to prove monkeys were genuinely processing ordinal *number*, the experimental architecture had to render every other physical variable misleading or uninformative. By systematically manipulating cumulative surface area, element size, spatial density, shape complexity, and overall configuration, Brannon ensured that discrete number was the only invariant, predictive dimension available to the animals across thousands of discrete trials.

1.3 Core Research Questions and Hypotheses

The investigative program pursued by Brannon and Terrace was guided by three interrelated research questions, each addressing a fundamental controversy in cognitive psychology. The first question asked: Can non-verbal primates extract pure, abstract numerical information independent of continuous physical dimensions? Traditional behaviorist and perceptual theories posited that animals respond exclusively to physical energy distributions—such as the total amount of light reflected by an array, the gross retinal area covered by objects, or the overall contour length. Brannon and Terrace hypothesized that rhesus monkeys possess cognitive mechanisms capable of filtering out these continuous physical attributes, allowing them to isolate the discrete number of individual items as a distinct perceptual and cognitive invariant.

The second core research question investigated the structural organization of animal numerical representations: Do monkeys represent numbers as an ordered mental continuum rather than isolated, discrete perceptual categories? Under a simple categorical learning model, an animal might learn that a visual display of three dots is associated with a specific behavioral response, and a display of four dots with another, without understanding that three is inherently less than four, or that three occupies a specific position between two and four. Brannon and Terrace hypothesized that primate numerical cognition is organized along an analog, sequential mental continuum—a mental number line—where cardinal quantities possess intrinsic relational values that dictate their position relative to all other quantities.

The third, and most critical, hypothesis addressed cognitive extrapolation and rule transfer: Can knowledge of an ordinal rule (e.g., an ascending sequence) acquired within a restricted numerical range be spontaneously transferred to novel quantities outside that training set? If a monkey trained to select items in the sequence 1 → 2 → 3 → 4 is presented with novel pairs containing quantities from 5 to 9, how will it respond? If the animal merely memorized a closed, four-element motor or visual chain, performance on novel pairs should drop to chance (50%). Conversely, if the monkey acquired an abstract ordinal rule operating over an open-ended mental continuum, it should spontaneously order novel pairs (such as selecting 5 before 8, or 7 before 9) on its first exposure, without requiring differential reinforcement. Demonstrating this spontaneous transfer would provide definitive evidence of true non-symbolic numerical ordinality.

2. Theoretical Foundations: Cardinality, Ordinality, and Number Conservation

2.1 Differentiating Cardinality from Ordinality

In formal cognitive mathematics, numerical competence is bifurcated into two primary conceptual frameworks: cardinality and ordinality. Cardinality refers to the quantitative property of a set as an absolute value—the total count of discrete elements contained within a defined boundary. Understanding cardinality requires an organism to recognize that a collection of three apples, three stones, and three acoustic tones all share an invariant, abstract numerical identity: the cardinal value of “threeness.” Cardinal tasks in comparative psychology typically assess whether an animal can match sets based on quantitative equivalence (such as delayed matching-to-sample paradigms) or accurately judge which of two sets contains a larger absolute quantity. While cardinal competence demonstrates that an animal can categorize groups based on element count, it does not necessarily imply that the animal understands how those sets relate to one another along a formal sequential trajectory.

Ordinality, by contrast, refers to the relational position of a set within a defined sequential progression. It is the comprehension of relative rank, sequence, and serial order: knowing not merely that a set contains four elements, but that four is intrinsically “more than” three and “less than” five, thereby occupying a non-arbitrary spatial or temporal coordinate along an ordered scale. Ordinal comprehension represents a qualitatively higher order of cognitive abstraction than simple cardinal discrimination. To master ordinality, an organism must mentally arrange disparate cardinal sets along a relational vector governed by transitive logic (if A < B and B < C, then A < C). While a pigeon or rat can be conditioned to peck a stimulus when presented with three dots, such performance can be achieved through isolated associative strength. Demonstrating ordinality requires showing that the internal representation of that stimulus carries an intrinsic relational tag that dictates its position relative to arbitrary, unencountered sets.

The theoretical debate regarding whether ordinality precedes or follows cardinality in phylogenetic and ontogenetic development remains a central topic in developmental and evolutionary psychology. Classic theorists, such as Jean Piaget, argued that true cardinal understanding cannot exist without an operational grasp of ordinal seriation, suggesting that children construct cardinality only after they understand serial positioning. Conversely, modern modular theorists, such as C. R. Gallistel and Rochel Gelman, have argued that primitive cardinal representations arise first via accumulator mechanisms, with ordinal relations subsequently derived from comparing these analog magnitudes. By investigating whether rhesus monkeys can generalize an ascending ordinal rule across novel quantities, Brannon and Terrace provided empirical evidence demonstrating that ordinal relational processing is an intrinsic feature of non-verbal primate mental architecture, operating hand-in-hand with basic magnitude estimation.

2.2 The Concept of Number Conservation

Number conservation is the cognitive capacity to recognize that the absolute quantity of discrete objects in a collection remains invariant despite changes in their physical, spatial, or morphological arrangement. First systematically formulated within developmental psychology by Jean Piaget, conservation serves as a litmus test for genuine mathematical understanding. If five stones are spaced closely together in a compact cluster, and another five identical stones are spread widely across a surface, an organism that relies purely on perceptual heuristics—such as the total spatial footprint or edge length—will judge the spread-out array to be quantitatively greater. An organism that conserves number, however, looks past these superficial spatial manipulations, maintaining an internal representation of the discrete cardinality of the set despite contradictory continuous visual cues.

Adapting conservation protocols for non-linguistic subjects presents profound methodological hurdles. Human children can be asked directly, “Are there more blocks here, or here?” With non-verbal animals, the researcher must infer the internal representational state entirely from physical motor selections within an operant environment. If an animal is rewarded for selecting displays based on number, the experimenter must prove that the animal is not simply responding to an unintended physical correlate of number. In the natural visual world, discrete numerosity almost always covaries with physical dimensions: five objects generally occupy more surface area, possess a larger cumulative boundary, reflect more total light, and exhibit greater spatial density than two objects of the same class. Therefore, establishing true number conservation in animals requires creating artificial visual environments where continuous physical dimensions actively conflict with discrete numerosity, forcing the subject to base its choices strictly on abstract numerical quantity.

The significance of number conservation testing in comparative cognition lies in its power to demonstrate true mathematical abstraction. If an animal’s performance drops to chance whenever spatial density or cumulative surface area is inverted, the animal is merely using continuous sensory proxies. If, however, the subject continues to systematically order and classify arrays regardless of extreme geometric distortions, variations in element size, and conflicting spatial distributions, the scientist has proven that the subject maintains an invariant, abstract representation of number. In the experimental paradigms of Brannon and Terrace, establishing number conservation was not an auxiliary control; it was the foundational prerequisite upon which all claims of monkey ordinality were evaluated.

2.3 The Approximate Number System (ANS) Paradigm

The theoretical scaffold underlying modern numerical cognition research is the Approximate Number System (ANS), an evolutionarily ancient cognitive mechanism that allows non-human animals and pre-verbal human infants to represent and manipulate non-symbolic numerical quantities. Unlike the exact, discrete representations enabled by human linguistic counting and symbolic mathematics, the ANS operates via internal, analog magnitudes. These magnitudes function much like mental representations of other continuous physical dimensions, such as brightness, weight, or duration. The primary hypothesis of the ANS is that numbers are mapped internally along a continuous, analog “mental number line,” where quantities are stored not as discrete symbols, but as noisy, overlapping Gaussian distributions of activation.

A defining signature of the ANS is its strict adherence to psychophysical laws, particularly the Weber-Fechner Law. Formulated by Ernst Heinrich Weber and Gustav Theodor Fechner in the nineteenth century, this law dictates that the perceived change in a stimulus is proportional to the initial magnitude of the stimulus itself. Applied to mental magnitude and quantity discrimination, the Weber-Fechner Law states that the ease with which two quantities can be differentiated is not a function of their absolute numerical difference, but rather the ratio between them. For instance, distinguishing between 1 and 2 items (a 1:2 ratio) involves an absolute difference of 1, exactly the same absolute difference required to distinguish between 8 and 9 items (an 8:9 ratio). However, under Weber’s Law, the 1 vs 2 discrimination is cognitively easier and faster than the 8 vs 9 discrimination, because the psychological distance between analog magnitudes scales logarithmically.

This ratio-dependence produces a critical psychophysical phenomenon known as scalar variability: the internal noise or imprecision associated with a mental representation increases in direct proportion to the magnitude of the number being represented. As the absolute value of a numerosity grows, its corresponding Gaussian activation distribution on the mental number line widens. Consequently, when an organism attempts to discriminate between two relatively large numbers that are close in value, the internal representations overlap substantially, leading to higher error rates and prolonged response latencies. In contrast, difference-dependent discrimination models would predict constant error rates for any identical mathematical step size regardless of absolute value. Demonstrating that an animal’s ordinal selections exhibit ratio-dependent Weberian characteristics serves as diagnostic evidence that the behavior is driven by the Approximate Number System rather than associative memorization.

3. The Piagetian Framework and Its Animal Cognition Analogues

3.1 Jean Piaget’s Structural Stages of Numerical Development

In his seminal 1952 work, The Child’s Conception of Number, Swiss psychologist Jean Piaget laid out a structural stage theory of cognitive development that profoundly influenced twentieth-century developmental psychology. Piaget asserted that true numerical comprehension is fundamentally impossible without the achievement of operational thought—a cognitive stage characterized by reversible mental operations, hierarchical classification, and conservation logic. According to Piaget, human children traverse a rigid developmental trajectory: passing from the sensorimotor stage through the pre-operational stage, and arriving at the concrete operational stage around seven years of age. It is only during this concrete operational period that a child can look past misleading perceptual transformations and demonstrate true number conservation.

Piaget substantiated this theory through classic conservation tasks involving liquids and discrete items. In the classic number conservation task, a child is shown two parallel rows containing an identical number of evenly spaced checkers. After the child agrees that both rows contain the same quantity, the experimenter spreads out the checkers in one row, increasing its physical length while leaving the actual count unchanged. Children in the pre-operational stage (typically ages four to six) routinely claim that the elongated row now contains “more” checkers, falling prey to the perceptual illusion created by spatial extent. Piaget argued that these children lack the mental operation of reversibility—the cognitive ability to mentally invert the transformation—and are therefore incapable of true numerical abstraction.

A foundational tenet of the classical Piagetian view was that non-human animals are permanently constrained to sensorimotor intelligence or, at best, a rudimentary form of pre-operational processing. Animals were assumed to lack the structural capacity for operational thought, reversibility, and formal conservation logic, leaving them permanently vulnerable to low-level perceptual confounds. However, by the late twentieth century, this rigid Piagetian orthodoxy faced substantial challenges from developmental researchers such as Karen Wynn, who demonstrated using looking-time methods that pre-verbal human infants possess sophisticated numerical expectations long before Piaget’s proposed timeline. These developmental insights laid the intellectual groundwork for comparative psychologists to ask whether non-human primates might also possess operational numerical capacities that had previously gone undetected due to flawed, linguistically dependent testing protocols.

3.2 Adapting Conservation Protocols for Non-Verbal Primates

To rigorously test whether a non-human animal can conserve number and extract ordinal relations, researchers had to redesign the classic Piagetian testing apparatus. The fundamental flaw of the traditional Piagetian interview was its complete reliance on linguistic competence. A human child must comprehend verbal syntax, parse experimenter questions, and articulate verbal judgments—demands that confound cognitive competence with communicative fluency. For non-human primates, all linguistic scaffolding had to be eliminated and replaced with non-verbal operant testing protocols. These protocols had to be intuitive, automated, and resistant to experimenter bias, social signaling, or inadvertent Clever Hans cues.

The primary methodological hurdle was establishing cross-dimensional equivalence without verbal instruction. How does one instruct a rhesus macaque to “order these sets from smallest to largest” when the animal shares no linguistic communication system with the human observer? The answer lay in the structural design of operant reinforcement contingencies. The experimenter must construct an interactive environment where the internal logic of the task can be acquired through operant trial-and-error, while ensuring that the rewarded dimension is strictly numerical. The animal must learn that its behavioral choices yield rewards (such as preferred juices or pellets) only when structured according to a consistent relational metric across thousands of variable trials.

To operationalize number conservation, the visual stimuli presented on these touchscreens had to be systematically engineered to create active perceptual conflicts. If an animal is asked to differentiate between a smaller set and a larger set, the visual arrays must be constructed so that the continuous physical dimensions systematically contradict the discrete count. The experimenter must generate arrays where two items are physically larger, wider, brighter, and more spread out than a competing array of four items. If the subject continues to accurately select the sets in their true numerical order under these conflicting conditions, the experimenter has achieved a non-verbal analogue of the Piagetian conservation task. The animal demonstrates that its internal choices are governed by the abstract invariant of cardinality rather than misleading sensory attributes.

3.3 Operationalizing Non-Human Conservation and Seriation

In Piagetian theory, seriation is defined as the systematic cognitive ordering of stimuli along an identifiable quantifiable gradient, such as arranging sticks from shortest to longest or arrays from least to most populous. Piaget considered seriation and conservation to be intertwined, reciprocal operations: an organism cannot truly seriate elements based on number unless it can conserve the cardinality of each element across spatial transformations. In the animal laboratory, operationalizing seriation required developing behavioral criteria capable of distinguishing between superficial perceptual grouping and genuine cognitive seriation based on abstract ordinal rank.

To prove that an animal is engaging in true seriation rather than perceptual grouping, three conditions must be met:
First, the animal must demonstrate transitive competence across multiple, non-adjacent items within a sequence. If an animal is trained on items A, B, C, and D, it must accurately judge relationships between novel pairings like B and D without explicit associative training on that specific dyad.
Second, the seriation behavior must be invariant to geometric, morphological, and spatial transformations of the stimuli, confirming number conservation.
Third, the internal rule governing the serial sequence must demonstrate transfer to entirely novel elements that fall outside the initial training range.

Bridging the gap between concrete operational seriation and non-human behavioral output required Elizabeth Brannon and Herbert Terrace to construct an experimental paradigm that wedded Piagetian structural questions to the rigorous methodologies of psychophysical operant conditioning. By defining clear behavioral metrics—such as response latencies, transitivity matrices, and transfer accuracy on first-trial exposures—they established a quantitative framework that moved the study of animal numerical cognition beyond qualitative observation, providing an empirical bridge between Piagetian developmental theory and comparative neurobiology.

4. Methodological Architecture: The Simultaneous Chaining Paradigm

4.1 Design and Setup of the Operant Touchscreen Apparatus

The empirical breakthrough achieved by Brannon and Terrace rested upon a meticulously designed operant apparatus known as the simultaneous chaining paradigm. Testing was conducted inside sound-attenuating, computer-controlled operant chambers specifically tailored for rhesus macaques (Macaca mulatta). The subjects sat inside a primate testing apparatus facing a high-resolution, touch-sensitive cathode-ray tube (CRT) monitor. By automating every aspect of stimulus presentation, data acquisition, and reward delivery through specialized computer hardware, the experimental architecture eliminated any human presence within the testing environment, removing the possibility of Clever Hans-style experimenter cuing.

The subjects participating in these landmark studies were two male rhesus macaques, historically identified in the literature as Rosencrantz and Macduff. Both animals had undergone extensive habituation to the experimental laboratory environment and were maintained under strict ethical protocols approved by the Columbia University Institutional Animal Care and Use Committee (IACUC). Their daily fluid intake was carefully managed to motivate performance for positive reinforcement, while ensuring health, physiological hydration, and welfare standards were met. The monkeys engaged in testing sessions voluntarily, receiving automated fluid rewards (such as sweetened fruit juice, water, or banana slurry) delivered directly through an electronic peristaltic pump fitted with a stainless-steel lick-tube positioned near the animal’s mouth.

Every behavioral interaction was tracked with millisecond-level precision. When a monkey touched a stimulus on the screen, the system registered the spatial coordinates, the response latency (reaction time), and the exact sequence of touches within the trial. Correct responses were paired with distinct audiovisual feedback, such as an immediate high-pitched auditory chime coupled with a metered droplet of fluid. Errors were immediately met with negative feedback: all visual stimuli vanished from the screen, the testing chamber went dark or displayed a solid red error screen, an aversive low-frequency buzzer sounded, and a non-rewarded timeout period (ranging from 2 to 10 seconds) was enforced before the subject could initiate the next trial.

4.2 The Simultaneous Chaining Mechanics

The simultaneous chaining paradigm, originally developed by Herbert Terrace for the study of serial list learning, differs fundamentally from traditional operant paradigms such as successive discrimination or simple matching-to-sample. In a typical successive discrimination task, an animal is shown stimuli one at a time and must decide whether to respond or withhold a response (a go/no-go design). In simultaneous chaining, all stimuli comprising a given sequence are presented concurrently on the visual display at the exact same moment. The animal is required to execute a coordinated sequence of motor responses, touching the stimuli one by one in an exact, predefined logical order (such as A → B → C → D).

The operational mechanics of the simultaneous chaining task impose a substantial cognitive load on the subject. Because all items appear at once across the display, the animal cannot simply react reflexively to an isolated target. Instead, before initiating its very first touch, the animal must visually scan the array, identify all relevant items, filter out distractors, determine the relative hierarchical rank of each stimulus, and chart a behavioral plan of execution. The task requires cognitive organization: the subject must hold the overarching sequence in working memory while executing individual motor selections, inhibiting the impulse to touch more perceptually salient items out of turn.

The operational rules governing simultaneous chaining are intentionally unforgiving. A trial is scored as correct if, and only if, the animal touches every single stimulus in its exact predetermined order. If a trial requires touching four stimuli in the order 1 → 2 → 3 → 4, touching stimulus 1 followed immediately by stimulus 3 results in an immediate trial termination, the firing of the error timeout, and the complete loss of any reward for that trial. The subject receives no partial credit. This all-or-nothing contingency forces the animal to extract the underlying structural logic of the sequence, discouraging careless exploratory touches and incentivizing careful behavioral planning before initiating a motor response.

4.3 Randomization and Spatial Dissociation

To guarantee that the animals were truly learning an abstract numerical sequence rather than an automated motor habit, Brannon and Terrace decoupled the numerical identity of the stimuli from their spatial coordinates. In earlier animal sequencing experiments, stimuli were often presented in fixed spatial configurations or followed predictable spatial trajectories (e.g., always moving from left to right, or clockwise around a circle). Under such flawed conditions, an animal does not need to learn anything about the conceptual relationships between the items; it simply learns a motor routine—a series of muscle memories tracing a static geometric path across the interface.

Brannon and Terrace eliminated these spatial strategies by randomizing stimulus positions across a multi-cell virtual grid on every single trial. The testing interface was configured as a virtual 4 × 4 or 3 × 3 matrix, offering 9 to 16 discrete, evenly spaced spatial locations. On any given trial, the computer algorithm selected spatial coordinates at random without replacement. Consequently, on trial n, the stimulus representing numerosity 1 might appear in the top-left corner, numerosity 2 in the bottom-right, numerosity 3 in the center, and numerosity 4 in the top-right. On trial n+1, the spatial locations of those exact same numerical values were completely scrambled across the screen.

This relentless spatial randomization ensured that the subjects could not rely on motor-pattern memorization, spatial kinesthetics, or visual scanning heuristics. A sequence of touches that yielded a reward on one trial would lead to an immediate failure on the next if executed in the same physical direction. The only invariant feature linking successful trials together across thousands of iterations was the abstract numerical rank of the stimuli. To succeed, the monkeys were compelled to locate the items visually, determine their numerical value, compare those values to their internal ordinal representation, and execute their physical responses based strictly on the ascending ordinal rule.

5. Rigorous Stimulus Controls: Disentangling Discrete Numerosity from Continuous Perceptual Variables

5.1 The Confounding Nature of Non-Numerical Visual Cues

The primary methodological challenge in all non-symbolic numerical cognition research is the ubiquitous covariation between discrete numerosity and continuous physical dimensions. In the natural physical environment, discrete number rarely presents itself in isolation; it is deeply intertwined with visual energy distributions. When the number of elements within an array increases, several non-numerical physical properties naturally covary with that increase. If a researcher displays three dots on a screen and compares them to six dots, the set of six dots will naturally possess twice the cumulative surface area, twice the total perimeter (contour length), and reflect twice the aggregate luminance, provided the individual dots are uniform in size.

Beyond surface area and perimeter, spatial envelope and spatial density present powerful confounding perceptual cues. The spatial envelope, or convex hull, refers to the area of the smallest convex polygon that can be drawn around a collection of objects—essentially the total footprint of the array. The spatial density refers to how closely packed the elements are relative to one another within that envelope. In unconstrained visual displays, larger numerosities tend to either occupy a larger overall envelope or exhibit greater local visual crowding. If an animal is rewarded for selecting displays with more items, the animal might simply select the display with the highest spatial frequency, the highest density of edges, or the largest outer boundary, completely bypassing the cognitive extraction of discrete number.

If these continuous variables are not controlled, an experimenter cannot legitimately claim to have demonstrated numerical competence. An animal that reliably chooses six dots over three dots might not be demonstrating numerical understanding; it may merely be an organism executing simple phototaxis, responding to the display that radiates greater overall visual luminance or activates low-level edge-detectors in the primary visual cortex (V1). Elizabeth Brannon recognized that to provide definitive proof of primate number conservation and ordinality, the experimental architecture had to construct an empirical firewall between discrete number and every single continuous perceptual variable capable of mediating performance.

5.2 The Multi-Attribute Control Matrix

To neutralize every continuous perceptual confound, Brannon and Terrace developed a multi-attribute control matrix for stimulus generation. Rather than utilizing static, repeating visual images, they generated thousands of unique, non-repeating stimulus displays where continuous variables were systematically, algorithmically manipulated to oppose, match, or dissociate from discrete numerosity. By forcing the continuous cues to either vary randomly or actively contradict the numerical sequence, the researchers ensured that relying on any non-numerical dimension would drop the animal’s performance down to chance levels.

The cornerstone of this control matrix was the equating of cumulative surface area across differing numerosities. For instance, in a trial requiring the ordering of 1, 2, 3, and 4, the total combined surface area of the single element in display 1 could be programmed to equal the aggregate surface area of the four elements in display 4. Under this control condition, individual element size was inversely proportional to numerosity: the single item in display 1 was large, while the four individual items in display 4 were small. If the monkeys were using cumulative surface area as an ascending cue, they would fail, as the area remained entirely static across the sequence.

To prevent the monkeys from using element size as an inverse cue (e.g., “always select the largest individual element first”), Brannon and Terrace implemented inverse area controls. In these configurations, cumulative surface area was inversely correlated with number: display 1 was given a massive cumulative surface area, while display 4 was given a tiny cumulative surface area, meaning both total area and individual element size decreased as numerosity increased. Furthermore, the researchers varied element shape (using circles, squares, triangles, polygons, and irregular organic forms), color (hue, saturation, and lightness), and internal spatial density within each trial. Aggregate luminance was balanced using high-contrast color combinations, and perimeter-to-area ratios were varied by combining simple rounded shapes with highly articulated, serrated geometric configurations.

5.3 The Mathematical Construction of Non-Confounded Stimulus Sets

The construction of these stimulus arrays relied on custom algorithms designed to pseudo-randomly generate unique geometric configurations for every individual trial. Rather than drawing from a small library of pre-rendered graphics, the software dynamically calculated element dimensions, spatial coordinates, and color assignments prior to rendering the array on the touchscreen monitor. This algorithmic approach prevented the monkeys from memorizing static visual templates, forcing them to process the relational properties of each newly presented set on the fly.

The experimental stimuli were bifurcated into two distinct classifications: Standard Stimulus Sets and Controlled Stimulus Sets. In the Standard Sets, visual elements were relatively uniform in size, allowing continuous variables to covary naturally with numerosity. These sets established baseline acquisition parameters. In the Controlled Sets, continuous physical dimensions were held constant or systematically inverted across adjacent and non-adjacent sequence items. If an animal was relying on an associative heuristic linked to any physical dimension—such as cumulative brightness, contour length, or spatial envelope—the transition from Standard to Controlled stimulus sets would produce a catastrophic collapse in ordering accuracy.

To verify that rote visual template memorization was mathematically impossible, Brannon and Terrace tracked stimulus exposure rates. Because the algorithmic generator combined multiple geometric shapes, varying color palettes, fluctuating scale factors, and randomized spatial grid placements, the probability of an animal encountering the exact same visual display twice across tens of thousands of trials was virtually zero. The only persistent invariant uniting a display of two jagged blue stars with a display of two tiny red circles was the abstract cardinal count: the invariant property of “twoness.”

6. Acquisition Phase: Training Rhesus Monkeys on Ascending Numerical Sequences (1 to 4)

6.1 Baseline Training with Numerosities 1, 2, 3, and 4

The training regimen for subjects Rosencrantz and Macduff began with an incremental acquisition phase designed to introduce the mechanics of the simultaneous chaining task using the numerosities 1, 2, 3, and 4. The experimental protocol did not expose the monkeys to the full four-item sequence immediately. Instead, training utilized a stepwise scaffolding procedure. The monkeys were initially introduced to two-item sequence pairs (e.g., 1 vs 2). Once performance met a rigorous criterion (typically ≥ 80% accuracy over consecutive blocks of 100 trials), the sequence was expanded to three items (1 → 2 → 3), and finally to the complete four-item sequence (1 → 2 → 3 → 4).

The reinforcement contingencies during this acquisition phase were strict. A trial commenced when the monkey touched a neutral start stimulus in the center of the display. Instantly, the numerical stimuli appeared simultaneously at randomized grid coordinates. If the monkey touched stimulus 1, it remained on the screen or provided a brief auditory confirmation; if the monkey then touched stimulus 2, followed by 3, and concluded with 4, the computer registered a successful trial, dispensed a fluid reward, and initiated a brief inter-trial interval. If the monkey touched any stimulus out of sequence at any point in the chain, the trial immediately aborted, visual stimuli were extinguished, an error tone sounded, and an error timeout was enforced.

The individual learning trajectories revealed substantial learning demands. Both Rosencrantz and Macduff required tens of thousands of trials spanning several months to master the full four-item sequence across diverse, algorithmically generated stimulus sets. Mastery demanded that the monkeys simultaneously solve two distinct cognitive challenges: suppressing the prepotent motor impulse to rapidly tap whatever stimulus caught their eye, and attending to discrete element count while ignoring fluctuating sensory properties. Despite the high task difficulty, both subjects successfully achieved the rigorous performance criteria, demonstrating robust ordering accuracy on the 1 → 2 → 3 → 4 sequence even when challenged with extreme continuous-variable controls.

6.2 Evidence of Rule Extraction During Training

A critical question during the acquisition phase was whether the monkeys were learning six independent pairwise associations (1 before 2, 2 before 3, 3 before 4, 1 before 3, 1 before 4, 2 before 4) or whether they were extracting an overarching, integrated ordinal rule. To evaluate this, Brannon and Terrace conducted detailed error analyses and latency distributions across intermediate elements of the sequence. If an animal learns independent, rote pairs, error rates on intermediate transitions (e.g., touching 3 before 2) should remain relatively uniform or reflect localized associative strength. Conversely, if an integrated ordinal rule is extracted, errors should systematically cluster around numerically adjacent items, exhibiting psychophysical distance effects.

The behavioral data strongly supported the rule extraction hypothesis. When errors occurred, the monkeys rarely committed catastrophic sequence violations (such as reaching for 4 on the first touch). Instead, sequence errors were concentrated on adjacent items within the chain—such as touching 3 instead of 2, followed by an immediate correction toward the remaining items. Furthermore, response latency analyses provided clear windows into internal cognitive planning. The reaction time required to execute the first touch in a trial was significantly longer than the latencies for subsequent touches in the chain. This prolonged initial latency indicated that the monkeys were pausing to formulate a full behavioral plan, visually surveying the entire array and organizing their trajectory according to an ordinal hierarchy before initiating physical contact.

The latency patterns also demonstrated that the monkeys were not simply executing an automated motor program. Even after thousands of trials, the time required to initiate a touch to stimulus 2 after touching stimulus 1 varied systematically based on the spatial distance and numerical relationships of the remaining items. The monkeys did not treat each stimulus encounter as an isolated learning event. They operated within a structured cognitive space, evaluating the stimuli in relation to an internal ordinal framework that mapped discrete element counts to a unified, sequential behavioral output.

6.3 Control Sequences: Ascending Versus Descending Training Groups

To determine whether primate numerical cognition possesses an inherent directional bias, Brannon and Terrace examined performance across directional conditions. In human culture, numerical sequences are predominantly conceptualized in an ascending direction (from smaller to larger), a convention reinforced by language, counting rhymes, and reading direction. But did non-human primates exhibit a natural cognitive predisposition toward ascending orders (1 → 2 → 3 → 4), or could they acquire a descending sequence (4 → 3 → 2 → 1) with equal ease?

To investigate this, experimental monkeys were trained either on an ascending rule (1 → 2 → 3 → 4) or a descending rule (4 → 3 → 2 → 1) under identical operant chaining protocols. The comparative acquisition data yielded significant asymmetries. Monkeys trained on the ascending sequence reached behavioral mastery in substantially fewer trials and exhibited significantly lower error rates than subjects assigned to the descending sequence. The ascending training group exhibited rapid rule abstraction, whereas the descending group struggled extensively with intermediate sequence transitions and exhibited higher rates of perseverative errors.

These findings pointed toward an endogenous cognitive architecture biased toward ascending magnitude representation in the primate lineage. An ascending progression mirrors natural ecological accumulation: quantities in the wild typically build up from a singular unit to a larger collective through the progressive addition of discrete elements. The ease with which the monkeys mastered the ascending sequence suggested that mapping smaller numerosities to earlier positions and larger numerosities to later positions aligns naturally with the functional architecture of the non-human primate brain, rather than being an arbitrary artifact of cultural conditioning.

7. The Critical Test: Spontaneous Generalization to Novel Numerosities (5 to 9)

7.1 Experimental Protocol for Novel Number Pair Testing

The acquisition of the 1 → 2 → 3 → 4 sequence, while methodologically impressive, could not on its own definitively resolve the central theoretical question. Critics could reasonably argue that an exceptionally intelligent monkey might memorize every permutation of four items through sheer associative persistence, without possessing an abstract concept of ordinality that extends beyond the number 4. The true empirical crucible—the experimentum crucis of Brannon and Terrace’s research program—required testing whether the monkeys could spontaneously extrapolate their ordinal knowledge to entirely novel numerosities that had never been reinforced during training.

To execute this test, Brannon and Terrace introduced five completely novel numerosities: 5, 6, 7, 8, and 9. The testing protocol shifted from a four-item simultaneous chain to a pairwise presentation paradigm. This pairwise design allowed the researchers to present all possible combinatorial dyads derived from the numbers 1 through 9. These pairs fell into three distinct experimental classes:
1. Familiar Pairs: Combos consisting exclusively of familiar, trained numerosities (e.g., 1 vs 3, 2 vs 4).
2. Familiar-Novel Pairs: Combos pairing a familiar numerosity with an entirely novel numerosity (e.g., 2 vs 5, 3 vs 8, 4 vs 9).
3. Entirely Novel Pairs: Combos pairing two completely novel numerosities that the monkeys had never before encountered in the context of the experiment (e.g., 5 vs 6, 5 vs 9, 6 vs 8, 7 vs 9).

To prevent the monkeys from acquiring the correct ordering of the novel pairs through operant trial-and-error during the test phase, Brannon and Terrace implemented a non-differential reinforcement schedule on novel trials. In standard training trials, only the correct sequence resulted in reward. However, on critical novel pair test trials, the monkeys received a fluid reward regardless of which item they selected first, provided they touched both items on the screen. Because both choices yielded an identical positive outcome, the monkeys could not learn the correct ordering of novel numbers via reinforcement feedback. The subjects’ performance on these novel pairs reflected their raw, unconditioned representational capacity. If the monkeys had merely memorized a closed, four-element list (1 → 2 → 3 → 4), their first-trial performance on entirely novel pairs (such as 6 vs 8) should have landed precisely at chance: 50%.

7.2 Unprecedented Results: Spontaneous Ordinal Sorting

The experimental results, published in Science in 1998, provided a decisive result in animal cognition. When presented with completely novel numerical pairs, both Rosencrantz and Macduff spontaneously ordered the stimuli in ascending numerical sequence well above chance levels on their very first trial exposures. There was no drop to 50% accuracy, and no prolonged learning curve. On trials featuring entirely novel pairs (5, 6, 7, 8, and 9), the monkeys correctly selected the smaller numerosity before the larger numerosity on roughly 75% of first-trial encounters, an outcome that easily cleared standard thresholds for statistical significance.

The monkeys’ performance on familiar-novel pairs (such as choosing 3 before 7, or 4 before 5) was equally robust, consistently approaching or exceeding 80% accuracy. Most critically, performance on entirely novel dyads—where both numbers lay completely outside the trained range of 1 to 4—demonstrated that the monkeys were not simply anchoring novel items to a familiar baseline. When presented with 5 versus 9, 6 versus 8, or 7 versus 9, the monkeys reliably touched the smaller numerosity first. The spontaneous nature of this performance, captured on pristine first-trial presentations under non-differential reinforcement, proved that the monkeys were not learning the sequence during testing; they were projecting an existing ordinal rule onto novel quantities.

This finding represented an unprecedented empirical demonstration in the history of comparative psychology. Previous animal sequencing studies had demonstrated that animals could learn to seriate arbitrary physical stimuli (such as letters, colors, or geometric shapes) using simultaneous chaining. However, when those earlier animals were trained on an arbitrary sequence like A → B → C → D and then presented with novel items E and F, they had no way of knowing how E and F related to the sequence; their ordering accuracy on E vs F was strictly chance. Brannon and Terrace’s monkeys succeeded because their training had not been anchored to arbitrary perceptual tokens. By training the monkeys on pure numerosity, the researchers had engaged an open-ended, continuous cognitive dimension—a mental number line—enabling the animals to effortlessly place novel quantities in their appropriate ordinal positions.

7.3 Confirmation of Non-Symbolic Numerical Ordinality

The spontaneous generalization of the ascending rule to novel numerosities confirmed that rhesus monkeys possess true non-symbolic numerical ordinality. The data ruled out the counter-hypothesis that performance had been mediated by memorized photographic templates or associative stimulus-stimulus chains. If an animal memorizes a static chain of visual images, introducing new images breaks the associative chain. The monkeys’ ability to integrate unencountered quantities into an ordered sequence demonstrated that their internal representations possess an inherent, relational vector.

This breakthrough provided empirical proof that non-human primates possess an endogenous, abstract ordinal scale that extends beyond their explicitly trained behavioral repertoire. The monkeys did not view the numbers 1, 2, 3, and 4 as an isolated, self-contained list; they understood them as points along a continuous mathematical continuum that naturally extends to 5, 6, 7, 8, 9, and beyond. Cardinal quantities are intrinsically bound to their ordinal positions within the primate mind: understanding what “five” is inherently carries the relational understanding that it is larger than “four” and smaller than “six.”

Furthermore, the instantaneous nature of this generalization, occurring in the total absence of differential reinforcement, settled an enduring philosophical debate regarding the nature of animal reasoning. It proved that complex, logical relational behavior could occur in a non-verbal organism without trial-and-error learning. The monkeys were capable of cognitive inference: they perceived the cardinal value of the novel sets, accessed their internal mental scale, applied the previously acquired ordinal rule (“select smaller before larger”), and planned their motor actions accordingly. The Brannon and Terrace experiments provided definitive evidence that the structural logic of mathematics has roots that run deep into our pre-linguistic evolutionary heritage.

8. Psychophysical Signatures: The Numerical Distance and Magnitude Effects

8.1 The Numerical Distance Effect

Beyond demonstrating above-chance ordering accuracy on novel pairs, Brannon and Terrace subjected their behavioral data to psychophysical modeling, uncovering the quantitative signatures that characterize the internal mental representations of their primate subjects. Chief among these signatures was the Numerical Distance Effect, a robust psychophysical phenomenon first described in human numerical cognition by Robert Moyer and Thomas Landauer in 1967. The numerical distance effect dictates that the cognitive discriminability of two numbers is an inverse function of the numerical distance separating them: as the mathematical distance between two values increases, discrimination accuracy increases and response latencies systematically decrease.

When Rosencrantz and Macduff were presented with numerical dyads, their behavioral performance mirrored the classic human distance effect. When the numerical distance between two stimuli was large (such as a distance of 4, seen in pairs like 1 vs 5 or 5 vs 9), the monkeys exhibited high accuracy rates (approaching 90%) and rapid response latencies. Conversely, when the numerical distance was minimal (a distance of 1, seen in adjacent pairs like 5 vs 6 or 7 vs 8), their accuracy dropped toward 65–70%, and their response latencies grew longer. The monkeys required significantly more time to evaluate and resolve adjacent numerical pairings than non-adjacent pairings.

Mathematical modeling of the monkeys’ error rates as a function of absolute numerical distance yielded linear and logarithmic regression curves identical to those observed when human subjects are asked to judge which of two Arabic digits is larger. This striking convergence between non-verbal rhesus macaques and literate adult humans demonstrated that beneath our symbolic mathematics lies an analog, non-symbolic processing engine. The presence of the distance effect in monkeys proved that the animals were not relying on an arbitrary lookup table; they were comparing analog quantities along a continuous mental dimension where nearby values overlap and require greater cognitive effort to resolve.

8.2 The Numerical Magnitude (Size) Effect

The second diagnostic psychophysical signature documented by Brannon and Terrace was the Numerical Magnitude Effect (also frequently referred to as the Numerical Size Effect). The magnitude effect dictates that for any fixed numerical distance, the cognitive discriminability of two quantities decreases as their absolute numerical magnitude increases. In simpler terms, distinguishing between two numbers separated by a step size of 1 becomes progressively more difficult as the numbers themselves grow larger.

In the Brannon and Terrace data, this effect manifested clearly when comparing performance across identical numerical intervals. Consider a constant numerical distance of 1: when the monkeys were presented with the small pair 1 vs 2, their ordering accuracy was near-perfect, and their latencies were minimal. However, when presented with the pair 3 vs 4, accuracy dropped slightly and latencies increased. When pushed out into the novel quantities with the pair 8 vs 9—which shares the exact same mathematical distance of 1 as the 1 vs 2 pair—ordering accuracy experienced its steepest decline, hovering just above statistical significance, accompanied by the longest decision latencies recorded in the study.

The emergence of the numerical magnitude effect provided compelling evidence that the primate mental number line is not organized linearly, but rather along a compressed, logarithmic or scalar scale. In an uncompressed, perfectly linear mental representation with constant noise, discriminating 8 from 9 should produce the exact same error rate and reaction time as discriminating 1 from 2, since the absolute difference in both cases is 1. The systematic deterioration of performance as absolute values increase proves that larger quantities are subject to scalar variability: the internal representational noise scales up with the magnitude of the stimulus, leading to greater mental overlap and higher error rates for larger sets.

8.3 Theoretical Fit with the Weber-Fechner Law

The simultaneous presence of both the numerical distance effect and the numerical magnitude effect established that the monkeys’ ordinal performance was governed by the Weber-Fechner Law. In modern cognitive psychophysics, internal numerical representations are modeled as overlapping Gaussian tuning curves plotted along an analog magnitude continuum. Each specific numerosity activates a central peak of mental representation, with the spread (standard deviation) of the distribution reflecting the degree of internal cognitive noise. Under Weberian scalar scaling, the ratio of the standard deviation to the mean representation remains constant across the entire number continuum.

By analyzing the monkeys’ choice accuracy across all pairwise combinations of numerosities 1 through 9, Brannon and Terrace fitted psychophysical performance curves to calculate the monkeys’ Weber fraction (denoted as w). The Weber fraction serves as an index of internal numerical acuity: it defines the minimum percentage difference required between two quantities for an organism to reliably discriminate between them. The calculated Weber fraction for Rosencrantz and Macduff was remarkably consistent, hovering around w ≈ 0.3 to 0.4. This value means that for a monkey to reliably discriminate a pair, the larger number must exceed the smaller number by approximately 30 to 40 percent.

This mathematical proof had profound implications. It confirmed that the cognitive architecture deployed by rhesus monkeys to order novel numerosities is identical in both structure and psychophysical properties to the Approximate Number System deployed by human adults when performing non-symbolic visual dot estimations under speeded conditions. The monkeys’ behavioral data did not reflect a crude, heuristic approximation; it followed the exact mathematical laws that govern sensory and perceptual processing across biological systems. The work proved that numerical ordinality is an authentic perceptual property, directly extracted, scaled, and manipulated by the primate brain.

9. Competing Explanations and Robustness: Ruling Out Perceptual and Associative Artifacts

9.1 The Subitizing Alternative and the Small-Number Constraint

Following the publication of Brannon and Terrace’s findings, critics and cognitive theorists explored alternative explanations to determine whether the monkeys’ performance could be explained without invoking an integrated mental number line. The most prominent counter-hypothesis centered on the dual-system model of numerical cognition, which posits a fundamental divide between small and large quantities. Under this view, processing quantities from 1 to 4 is mediated by an Object Tracking System (OTS)—an attentional mechanism linked to visual working memory that tracks individual items via discrete mental “pointers” (often called visual FINSTs, or Finger of Instantiation). In contrast, quantities greater than 4 are processed by the Approximate Number System (ANS).

Proponents of this dual-system model argued that the monkeys’ acquisition of the initial 1 → 2 → 3 → 4 sequence was not an exercise in numerical magnitude estimation, but rather an exercise in visual object tracking or perceptual subitizing. If the OTS handles up to four discrete items, the monkeys might have processed the initial training set using an entirely different cognitive architecture than the one required for numbers 5 through 9. Under this critique, the monkeys’ spontaneous transfer to numbers 5 through 9 was viewed with skepticism: could an animal really shift effortlessly from an attentional tracking system directly into an analog magnitude system without missing a beat?

Brannon and Terrace addressed this challenge by pointing out the psychophysical continuity of their data across the 1–4 and 5–9 boundaries. If the monkeys had shifted from an Object Tracking System to an Approximate Number System between 4 and 5, their behavioral profiles should have exhibited a sharp discontinuity—a sudden break in reaction time curves, an abrupt drop in accuracy, or a disruption of the Weberian distance effect at the boundary. Instead, the psychophysical data displayed seamless continuity: response latencies and error rates followed a single, uninterrupted logarithmic function from 1 all the way through 9. This continuity provided empirical evidence that, for rhesus macaques in this simultaneous chaining task, a unified analog magnitude system operates across the entire span of small and large numerosities.

9.2 The Spatial Frequency and Visual Density Counter-Arguments

A second major alternative hypothesis suggested that the monkeys’ performance might be driven by low-level, non-numerical visual features that slipped through the experimental controls. Visual neurophysiologists pointed out that arrays containing different numbers of geometric elements naturally generate distinct spatial power spectra and visual density profiles. When an image contains many small, distributed elements, it produces high spatial frequency energy; an image containing fewer, larger elements produces lower spatial frequency energy. Skeptics argued that the monkeys might simply be sorting images based on their Fourier power spectra or Gabor-filter metrics, using low-level early visual cortex activations rather than abstract numerosity.

To evaluate this spatial frequency hypothesis, extensive post-hoc computational analyses and follow-up behavioral tests were conducted. Researchers modeled the visual displays using digital Gabor filters designed to simulate the receptive fields of simple and complex cells in primate visual areas V1 and V2. If the monkeys had been ordering the displays based on low-level spatial frequency energy or local edge density, these computational models should have successfully predicted the animals’ behavioral choices across both standard and controlled trials.

The computational models failed to predict monkey behavior. Across the multi-attribute control matrix designed by Brannon and Terrace, displays with identical numbers of elements varied dramatically in their spatial frequency spectra due to randomized element shapes, varying internal spacings, and inverse area manipulations. A display containing four jagged, elongated polygons yielded a vastly different Fourier profile than a display containing four compact, smooth circles, yet the monkeys sorted both with equivalent accuracy based solely on element count. The failure of low-level visual computational models proved that monkey ordering behavior could not be explained by early sensory filters. The animals were constructing a high-level, abstract cognitive representation that transcended raw visual spatial frequencies.

9.3 Associative Learning and Chain Traversal Models

The final line of alternative explanation emerged from traditional behaviorist learning theory, which sought to account for the results using associative reinforcement models, such as Value Transfer Theory. Under value transfer models, stimuli that are repeatedly presented alongside rewarded items acquire secondary associative value through temporal and spatial contiguity. A theorist could argue that during the extensive baseline training on 1 → 2 → 3 → 4, stimulus 1 acquired the highest net positive associative value (since it was always touched first and was consistently present during reward delivery), while stimulus 4 acquired the lowest net value, creating a descending gradient of associative strength across the familiar items.

However, when these associative learning models were mathematically applied to the critical novel test phase, they broke down entirely. Value Transfer Theory and related reinforcement models depend fundamentally on prior reinforcement history. A stimulus that has never been presented and never reinforced possesses an associative value of zero. When Rosencrantz and Macduff were presented with novel numerosities like 5, 6, 7, 8, and 9, every single one of these stimuli had an identical associative reinforcement value: exactly zero. Pure associative models predict that when an animal is forced to choose between two stimuli that both possess an associative strength of zero (such as 6 vs 8), its choice must be random, yielding 50% accuracy.

Because the monkeys ordered completely novel, zero-value pairs at roughly 75% accuracy on their very first trial exposures, associative reinforcement models were invalidated. The monkeys’ choices were not driven by accumulated associative strength, secondary reinforcement gradients, or backward chaining mechanics. The only theoretical architecture capable of explaining above-chance ordering on zero-value novel stimuli is a model in which the animals spontaneously extract the discrete cardinal value of the displays and evaluate their relative coordinates along an internal, open-ended mental number line. The associative alternative was definitively set aside in favor of a cognitive, representational account.

10. Comparative Neurological Substrates: Neural Basis of Ordinality and Numerosity

10.1 The Intraparietal Sulcus (IPS) as the Cortical Locus of Quantity

The behavioral discoveries made by Elizabeth Brannon and Herbert Terrace provided a clear psychophysical framework that soon catalyzed neurophysiological investigations into the primate brain. If rhesus monkeys possess an endogenous, analog representation of numerical ordinality, there must be a physical neurobiological substrate responsible for encoding these abstract numerical values. Neurophysiologists, most notably Andreas Nieder and Earl Miller at the Massachusetts Institute of Technology (MIT), set out to locate and record from the specific cortical circuits responsible for non-symbolic quantity processing in rhesus macaques.

Their electrophysiological recordings identified the Intraparietal Sulcus (IPS)—and specifically the Ventral Intraparietal Area (VIP) deep within the posterior parietal cortex—as the primary cortical locus for abstract numerical representation. By inserting microelectrodes into the parietal cortex of monkeys performing visual quantity tasks, Nieder and colleagues discovered individual neurons that responded selectively to specific numerosities. One neuron might fire maximally when the monkey viewed a display of three items, firing significantly less for two or four items, and remaining virtually silent for displays of one or five items. Another neuron would show peak tuning for five items, falling off in a Gaussian profile for adjacent quantities.

Crucially, these parietal neurons exhibited tuning profiles that matched the psychophysical signatures identified by Brannon and Terrace. First, the neural tuning curves were tuned to *abstract discrete numerosity*, firing consistently regardless of whether the display contained circles, triangles, or irregular shapes, and remaining invariant across wide shifts in cumulative surface area and density. Second, when plotted on a logarithmic scale, these neural tuning curves maintained constant widths, providing a direct neurophysiological basis for the Weber-Fechner Law and the numerical magnitude effect. The parietal cortex was revealed to house an authentic population code for the primate mental number line.

10.2 The Role of the Prefrontal Cortex (PFC) in Rule Encoding and Sequence Planning

While the posterior parietal cortex serves as the sensory-cognitive locus for extracting and tuning cardinal values, the execution of the simultaneous chaining task demands higher-order cognitive control: planning sequences, inhibiting prepotent touches, and applying an abstract ordinal rule. This executive orchestration is carried out by the lateral prefrontal cortex (PFC), a cortical region that shares dense, reciprocal neuroanatomical projections with the intraparietal sulcus.

Single-unit electrophysiological recordings in monkeys performing numerical ordering tasks have shown that prefrontal neurons encode both cardinal magnitude and the overarching ordinal rule governing the behavior. When a monkey is trained on an ascending sequence task, neurons within the dorsolateral prefrontal cortex (dlPFC) fire to encode the relative sequential phase of the behavior (e.g., “select smallest first,” “select next in sequence”). Prefrontal neurons fire well before a physical movement is initiated, reflecting the internal cognitive planning that Brannon and Terrace first inferred from their initial response latency analyses.

This functional division of labor illustrates a two-tier cortical network for non-verbal numerical cognition. The posterior parietal cortex processes the visual scene, strips away continuous perceptual confounds, and generates an analog population code representing the cardinal quantity of each display. The prefrontal cortex accesses these parietal representations, aligns them along an ordinal vector according to the active task rule (ascending vs descending), and issues motor commands to the primary motor cortex to execute the sequence. The simultaneous chaining paradigm deployed by Brannon and Terrace engaged this distributed frontoparietal network, confirming that non-human primates utilize advanced cortical circuitry to execute numerical operations.

10.3 Evolutionary Homology Between Primate and Human Brain Structures

The neurobiological findings in rhesus macaques led directly to translational functional neuroimaging studies in human subjects, revealing an evolutionary homology between the non-human primate brain and the human brain. Functional Magnetic Resonance Imaging (fMRI) studies led by cognitive neuroscientists such as Stanislas Dehaene have repeatedly demonstrated that when human adults view non-symbolic dot arrays, perform mental arithmetic, or compare Arabic numerals, the human intraparietal sulcus lights up with robust metabolic activity. The exact sub-regions of the human IPS activated during numerical tasks correspond topographically to the VIP and lateral intraparietal areas (LIP) identified in macaque electrophysiology.

This deep phylogenetic continuity is central to Stanislas Dehaene’s Triple-Code Model of numerical cognition. The Triple-Code Model posits that human numerical architecture relies on three distinct representational codes:
1. A visual symbolic code (Arabic numerals, processed in visual occipitotemporal regions).
2. A verbal auditory code (number words, processed in linguistic perisylvian networks).
3. An analog magnitude code (the abstract mental number line, localized within the bilateral intraparietal sulci).

The third code—the analog magnitude system—is not a unique, de novo invention of the human species. It is an inherited, evolutionarily conserved neural system shared with non-human primates. Non-human primates possess the third code in full functional maturity: they process, conserve, and ordinally arrange analog quantities using the exact same parietal circuits that humans repurpose during development. When human children learn formal symbolic mathematics, they do not build a numerical system from scratch. Instead, cultural education repurposes—or “neuronal recycles”—the evolutionarily ancient frontoparietal magnitude circuits that evolved in ancestral primates to support survival behaviors like foraging, social tracking, and spatial navigation.

11. Broader Implications for the Evolution of Mathematical Thought and Pre-Linguistic Systems

11.1 Phylogenetic Continuity of Numerical Competence

For centuries, Western philosophical tradition was dominated by the Cartesian doctrine that language is an absolute prerequisite for abstract thought. René Descartes famously argued that animals are biological automata, lacking reason, self-awareness, and the capacity for conceptual abstraction, because they lack verbal language. In the realm of mathematics, this view persisted well into the twentieth century, with many philosophers and linguists arguing that mathematical thought is an offshoot of linguistic syntax. Number, under this view, was treated as a cultural artifact invented through formal linguistic symbols.

The empirical research of Elizabeth Brannon and Herbert Terrace dealt a blow to this Cartesian dogma. By proving that rhesus monkeys can conserve number, isolate discrete cardinality from continuous variables, and spontaneously seriate novel quantities, their work proved that mathematical logic is biologically primary, not linguistically derived. The mental number line does not require words, syntactic markers, or Arabic numerals to exist. Language provides a symbolic mechanism for precision, allowing exact calculations of massive quantities, but the underlying conceptual framework—the understanding of more, less, order, and relation—is evolutionarily primitive, predating the divergence of Old World monkeys and hominids by at least 25 to 30 million years.

The evolutionary drivers responsible for the emergence of this ancient numerical capacity are rooted in fundamental ecological demands. For a wild primate, the capacity to process numerical quantity and ordinal relationships confers significant adaptive advantages:
First, in foraging optimization, an animal that can quickly assess and order fruit clusters based on discrete count—while ignoring misleading leaf coverage—maximizes its caloric intake.
Second, in intergroup territorial conflict, wild primates routinely evaluate numerical superiority. Classic field experiments by Richard Wrangham and colleagues have shown that chimpanzees and wild baboons decide whether to engage in aggressive territorial battles or retreat based on a mental assessment of the number of vocalizing rivals versus the number of allies in their own coalition.
Ordinal competence allows a primate to rapidly calculate relative social dominance and group size, directly enhancing reproductive fitness and survival.

11.2 The Ontogeny-Phylogeny Parallel: Infant Cognition and Primate Ordinality

The insights derived from rhesus monkeys yielded immediate implications for human developmental psychology, revealing parallels between the phylogenetic evolution of numerical competence and its ontogenetic emergence in human infants. For decades, developmental researchers relying on Piagetian interviews assumed that human infants were mathematically blank slates, incapable of processing quantity until early childhood. However, inspired by the animal psychophysical methods established by Terrace, Brannon, and their contemporaries, developmental cognitive scientists adapted non-verbal looking-time protocols (such as habituation-dishabituation and violation-of-expectation paradigms) to probe the mental lives of pre-verbal infants.

Elizabeth Brannon subsequently extended her comparative research program directly to human infants, conducting cross-species investigations that tested 6- to 12-month-old babies on visual ordinal tasks analogous to those given to Rosencrantz and Macduff. The findings revealed that human infants possess an active, non-symbolic Approximate Number System that exhibits the exact same psychophysical signatures observed in adult rhesus macaques. Pre-verbal infants notice changes in visual numerical quantities, display a distinct numerical distance effect, and demonstrate ordinal expectations, looking significantly longer when an ascending sequence of dot arrays (e.g., 2 → 4 → 8) is unexpectedly reversed into a descending sequence.

These cross-species paradigms demonstrated that the non-verbal monkey is a powerful cognitive model for the pre-verbal human infant. Before an infant utters its first word or learns to count on its fingers, its parietal cortex represents discrete quantity, conserves number across spatial manipulations, and organizes magnitudes along an ordinal trajectory. Ontogeny recapitulates phylogeny not in physical morphology, but in the shared, foundational reliance on an evolutionarily conserved Approximate Number System that provides the cognitive bedrock upon which formal cultural mathematics is subsequently erected.

11.3 The Bridge from Ordinality to True Symbolic Mathematics

How does a cognitive system transition from the noisy, analog magnitude estimations of the Approximate Number System to the exact, discrete calculations of formal symbolic mathematics? This question represents one of the major research frontiers in cognitive science. Non-symbolic ordinality provides the conceptual scaffolding required to build that bridge. To understand what an abstract symbol (such as the Arabic numeral “5”) truly represents, a child cannot merely learn its spoken name or arbitrary visual shape. The child must map that symbol onto two foundational cognitive structures: its cardinal magnitude (what five items feel like) and its ordinal rank (that 5 falls after 4 and before 6).

The work of Brannon and Terrace demonstrated that non-human primates possess the complete non-symbolic scaffolding onto which symbolic numerals can be mapped. Building directly on the 1998 and 2000 studies, subsequent primate researchers, including Elizabeth Brannon, Tetsuro Matsuzawa, and Jessica Cantlon, demonstrated that non-human primates can be taught to associate discrete numerosities with arbitrary Arabic numerals. Chimpanzees like Ai at Kyoto University learned to touch Arabic numerals 1 through 9 on a touchscreen in their correct ascending sequence, matching the symbols to visual arrays containing corresponding numbers of dots.

However, comparative research also revealed the limits of animal symbolic mastery. While monkeys and apes can learn to map a limited set of symbols onto their internal analog magnitudes, their symbolic representations retain the scalar variability and noise of the Approximate Number System. Humans take a unique conceptual leap through language: by combining discrete linguistic counting routines with symbolic recursive syntax, human children construct an exact integer concept where every number is precisely n + 1. Non-human primates provide the biological proof that the analog foundation of this system—the mental number line, ordinal seriation, and number conservation—evolved millions of years before humans created the cultural symbols that turned analog magnitude into exact mathematics.

12. Subsequent Developments, Criticisms, and the Modern Legacy of Brannon and Terrace

12.1 Direct Replications and Extensions in Other Species

The landmark discoveries of Brannon and Terrace triggered a global wave of comparative research, with scientists seeking to determine whether non-symbolic numerical ordinality was unique to Old World primates or represented a widely distributed cognitive adaptation across the animal kingdom. The simultaneous chaining and pairwise ordinal paradigms were rapidly replicated and adapted across diverse primate taxa. Researchers confirmed that olive baboons (Papio anubis), chimpanzees (Pan troglodytes), and New World capuchin monkeys (Cebus apella) all exhibited the capacity to spontaneously order familiar and novel numerosities, accompanied by the classic signatures of the numerical distance and magnitude effects.

Beyond the primate order, comparative psychologists extended ordinality testing across disparate vertebrate and invertebrate classes. In avian species, researchers working with pigeons (Columba livia) demonstrated that while pigeons require significantly more training trials than macaques, they too can learn to order visual numerosities 1 to 3 and show weak generalization to 4. In corvids, testing demonstrated that New Caledonian crows and carrion crows possess numerical acuity and neural tuning curves in their endbrain (the nidopallium nidopalliale) that rival those of rhesus macaques, despite lacking a laminated mammalian cerebral cortex.

Most remarkably, researchers extended non-verbal ordinal paradigms to invertebrates, specifically the European honeybee (Apis mellifera). In experiments led by Adrian Dyer and Scarlett Howard, honeybees were trained to enter a Y-maze and select visual displays based on ascending or descending numerical rules. The bees not only learned to order small quantities (1 vs 2, 2 vs 3), but they also demonstrated spontaneous transfer to novel quantities (4 vs 5) and exhibited an implicit understanding of the concept of zero as a quantitative value positioned at the lower end of the numerical continuum. These findings across insects, birds, and primates demonstrated that numerical ordinality is an example of convergent cognitive evolution, solving universal ecological challenges across diverse biological brains.

12.2 Methodological Critiques and Continuous Stimulus Debates

Despite its widespread acclaim, the experimental research program of Brannon and Terrace was not exempt from scientific critique. The central point of ongoing contention has focused on the persistent difficulty of completely eliminating continuous sensory confounds in visual arrays. Methodologists such as Tiziana Gebuis, Reynvoet, and colleagues have argued that even the multi-attribute control matrices designed by Brannon and Terrace may leave residual sensory cues that animals can exploit. They demonstrated that mathematically, it is impossible to simultaneously equate cumulative surface area, individual element size, perimeter, convex hull, spatial density, and aggregate luminance within the exact same visual display; controlling for one variable inevitably forces another to covary with number.

This methodological debate spurred significant refinements in comparative experimental design. In response to these critiques, Jessica Cantlon, Elizabeth Brannon, and their colleagues developed standardized, publicly accessible algorithmic stimulus-generation software packages (such as the MATLAB-based program *MakeStim*). These tools systematically vary all continuous perceptual dimensions orthogonally across massive trial blocks, ensuring that while individual trials may contain a residual physical cue, no single continuous dimension can predict the correct response across the entire experimental testing session.

The ongoing debate transformed the field’s understanding of how organisms perceive quantity. Rather than viewing number and continuous sensory variables as mutually exclusive opposites, contemporary cognitive scientists increasingly view quantity processing as an integrated, multi-sensory gestalt. While animals naturally attend to continuous cues like surface area and density, the work of Brannon, Terrace, and their successors demonstrated that primates can, when required, strip those continuous layers away, isolating the underlying discrete numerical core to execute abstract ordinal judgments.

12.3 Enduring Contributions to Cognitive Science and Animal Minds

The historical significance of Elizabeth Brannon and Herbert Terrace’s research cannot be overstated. Before their 1998 and 2000 publications, animal numerical competence was largely viewed as an esoteric, fringe subject within comparative psychology, continually besieged by the ghost of Clever Hans and behaviorist reductions. Brannon and Terrace brought the study of animal mathematics into the mainstream of cognitive neuroscience and evolutionary psychology. They proved that rigorous psychophysics, automated operant technology, and high-level cognitive theories could be combined to test the abstract contents of an animal’s mind with quantitative precision.

Their establishment of the simultaneous chaining paradigm as a gold standard for studying serial cognition transformed comparative methodology. They shifted the paradigm away from simple associative conditioning models, providing an experimental blueprint for exploring how non-verbal organisms represent time, space, and order. Their work directly inspired the neurophysiological studies of Andreas Nieder, Earl Miller, and Stanislas Dehaene, which uncovered the parietal and prefrontal circuits that generate mathematical thought, ultimately tracing the biological origin of human arithmetic back to ancient non-verbal mechanisms.

Elizabeth Brannon and Herbert Terrace redefined our understanding of the animal mind. By proving that a rhesus monkey can conserve number across visual transformations, hold an abstract ordinal rule in working memory, and spontaneously project that rule across an open-ended mental continuum, they demolished the theoretical barrier separating human mathematical reasoning from animal cognition. Their work stands as an enduring testament to the profound evolutionary continuity that unites the human intellect with the rest of the natural world.

Conclusion

The landmark investigations conducted by Elizabeth Brannon and Herbert Terrace fundamentally transformed cognitive science by demonstrating that numerical conservation and ordinality exist independently of human language and culture. Through the simultaneous chaining paradigm, rigorous stimulus-generation algorithms, and multi-attribute controls, they proved that rhesus macaques can isolate discrete cardinal numbers from confounding physical variables and spontaneously seriate novel quantities along an internal, ascending mental number line. The emergence of diagnostic psychophysical signatures—specifically the numerical distance and magnitude effects, both adhering to the Weber-Fechner Law—provided empirical proof that non-verbal primates utilize an analog magnitude representation identical in structure and function to the human Approximate Number System.

By providing a decisive non-verbal analogue to Piagetian number conservation and seriation, Brannon and Terrace dismantled the long-standing Cartesian assumption that abstract mathematical reasoning requires symbolic syntax. Their empirical findings revealed that the cortical machinery of mathematics—housed within the reciprocal circuits of the intraparietal sulcus and the prefrontal cortex—is an evolutionarily ancient adaptation shared across the primate lineage. Human symbolic arithmetic, far from being an isolated cultural invention, is built upon this evolutionary foundation, repurposing pre-linguistic cognitive architectures that evolved millions of years ago to track, conserve, and order quantity in the natural world.

Ultimately, the legacy of Brannon and Terrace’s work lies in its validation of cognitive continuity between non-human animals and humans. By demonstrating that monkeys possess an intrinsic understanding of numerical order, their research enriched our understanding of animal minds and provided a foundation for modern cognitive neuroscience, developmental psychology, and evolutionary biology. The discovery that non-verbal primates spontaneously represent the abstract relations of numbers 1 through 9 remains one of the most compelling demonstrations in comparative psychology, proving that the roots of mathematical thought are deeply embedded in our shared evolutionary heritage.

References

• Brannon, E. M., & Terrace, H. S. (1998). Ordering of the numerosities 1 to 9 by monkeys. Science, 282(5389), 746–749. https://www.science.org/doi/10.1126/science.282.5389.746
• Brannon, E. M., & Terrace, H. S. (2000). Representation of the ordinal relations of numerosities 1–9 by rhesus monkeys (Macaca mulatta). Journal of Experimental Psychology: Animal Behavior Processes, 26(1), 31–49. https://doi.org/10.1037/0097-7403.26.1.31
• Cantlon, J. F., & Brannon, E. M. (2006). Shared system for ordered numeric representations in monkeys and humans. PLoS Biology, 4(5), e125. https://doi.org/10.1371/journal.pbio.0040125
• Cantlon, J. F., & Brannon, E. M. (2007). How much does number matter to a monkey (Macaca mulatta)? Journal of Experimental Psychology: Animal Behavior Processes, 33(1), 32–41. https://doi.org/10.1037/0097-7403.33.1.32
• Dehaene, S. (2011). The Number Sense: How the Mind Creates Mathematics (Revised and Expanded ed.). Oxford University Press.
• Dehaene, S., Piazza, M., Pinel, P., & Cohen, L. (2003). Three parietal circuits for number processing. Cognitive Neuropsychology, 20(3-6), 487–506. https://doi.org/10.1080/02643290244000239
• Gallistel, C. R., & Gelman, R. (1992). Preverbal and verbal counting and computation. Cognition, 44(1-2), 43–74. https://doi.org/10.1016/0010-0277(92)90050-R
• Gebuis, T., & Reynvoet, B. (2012). The role of visual sensory cues in non-symbolic number estimation. Cognition, 122(3), 394–409. https://doi.org/10.1016/j.cognition.2011.11.003
• Howard, S. R., Avarguès-Weber, A., Garcia, J. E., Greentree, A. D., & Dyer, A. G. (2018). Numerical ordering of zero in honey bees. Science, 360(6393), 1124–1126. https://www.science.org/doi/10.1126/science.aar4975
• Moyer, R. S., & Landauer, T. K. (1967). Time required for judgements of numerical inequality. Nature, 215(5109), 1519–1520. https://doi.org/10.1038/2151519a0
• Nieder, A., & Miller, E. K. (2002). Coding of cognitive magnitude: Representation of number of dots in primate prefrontal cortex. Science, 297(5587), 1708–1711. https://www.science.org/doi/10.1126/science.1072493
• Nieder, A., & Miller, E. K. (2004). A parieto-frontal network for abstract number representation. Nature Reviews Neuroscience, 5(11), 888–898. https://doi.org/10.1038/nrn1496
• Pfungst, O. (1911). Clever Hans (The Horse of Mr. von Osten): A Contribution to Experimental Animal and Human Psychology (C. L. Rahn, Trans.). Henry Holt and Company.
• Piaget, J. (1952). The Child’s Conception of Number. Routledge & Kegan Paul.
• Terrace, H. S. (1984). Simultaneous chaining: The emergence of a new approach to animal cognition. Journal of Experimental Psychology: Animal Behavior Processes, 10(2), 200–214. https://doi.org/10.1037/0097-7403.10.2.200
• Wynn, K. (1992). Addition and subtraction by human infants. Nature, 358(6389), 749–750. https://doi.org/10.1038/358749a0