Glance at three coffee cups on a table and you know there are three without counting them. Look at a crowd in a stadium and you can say "a few thousand" with some confidence, but not whether the exact figure is 4,312 or 4,318. Ask a child to work out 7 × 8 and you are calling on a different skill again: a memorized fact stored in words. These three experiences point to the central puzzle of numerical cognition, the scientific study of how minds represent and work with numbers.
Numerical cognition is a subdiscipline of cognitive science. It studies the cognitive, developmental and neural foundations of numbers and mathematics. It draws on cognitive psychology, developmental psychology, neuroscience and cognitive linguistics. The field sometimes touches philosophical questions about what mathematics is, but its main business is empirical: experiments with animals, infants, adults, brain-injured patients and people from cultures whose languages have few or no number words.
This article covers the main questions the field asks, the experiments that shaped it, what neuroscience has found, how number connects to space and everyday judgment, what we know about dyscalculia, and where the open debates lie.
The core questions
The field is organized around a few big questions:
- Do non-human animals have a sense of number? If so, how precise is it, and how does it compare with ours?
- How do infants come to understand number? Are babies born with some numerical ability, or is it built entirely through experience and teaching?
- How do people link symbols to quantities? A written "7" or the spoken word "seven" is an arbitrary symbol. How does the brain tie it to the idea of seven things?
- How do these basic abilities support complex mathematics? Calculus and algebra are recent cultural inventions. What older mental machinery do they rely on?
A recurring answer is that humans and many animals share at least two basic, non-verbal systems for quantity. Exact, symbolic number, the kind needed to tell 4,312 from 4,318, seems to depend on language and culture built on top of them.
Two foundational systems
Subitizing and the small-number system
Subitizing is the fast, accurate and effortless perception of small quantities, usually four items or fewer, without counting. The term was coined in 1949 by E. L. Kaufman and colleagues, from the Latin subitus, meaning "sudden." Earlier, in 1871, the economist and logician W. S. Jevons had described the feeling of immediately knowing how many items are present when the number is small.
The timing data are clear. In experiments where people report how many dots they see, response time rises by only about 40–100 milliseconds per extra item within the subitizing range. Beyond it, each extra item adds about 250–350 milliseconds. That jump suggests a change of process, from immediate perception to effortful counting or estimation. Larger sets can sometimes be grasped instantly if they form a familiar pattern, such as the pips on a die.
Two lines of evidence suggest subitizing really is its own mechanism. In 1976, Atkinson, Campbell and Francis showed observers afterimages of disks. Because afterimages move with the eyes, people cannot scan across them to count. Observers made no errors with one to four disks, even after 60 seconds, but made consistent errors with five to twelve. The second line comes from patients with simultanagnosia, part of Bálint's syndrome, which is linked to damage on both sides of the parietal lobes. Dehaene and Cohen (1994) reported that such patients cannot enumerate objects beyond the subitizing range but can still enumerate small sets. Subitizing has also been found for touch: a 2006 study by Riggs and colleagues found the same pattern when people judged how many fingertips were being stimulated, and congenitally blind adults show it too.
The neural story is less tidy. A PET study by Piazza and colleagues in 2002 found overlapping activity for subitizing and counting in the occipital extrastriate cortex and in superior parietal and intraparietal regions. Counting also recruited right inferior frontal areas and the anterior cingulate, which some researchers link to shifting attention. Later work, including Vetter and colleagues in 2008, found that attention affects subitizing too, which weakens the idea of a sharp split between the two processes.
Researchers often connect subitizing to what is called the parallel individuation system (or object-tracking system): a capacity to keep track of up to about three or four individual objects at once. This system is exact but limited in size.
The approximate number system
For larger quantities, humans and many animals use the approximate number system (ANS). It is a non-symbolic way of estimating how many items are in a set without counting or using language. Its signature is that it follows Weber's law: how easily two quantities can be told apart depends on their ratio, not their absolute difference. Telling 8 from 16 is easy. Telling 80 from 88 is hard, even though the gap is the same size as from 8 to 16.
Adult ANS precision is roughly 15%: an adult can typically distinguish a set of 100 items from one of 115, but finer differences become unreliable without counting. This ratio-based, fuzzy quality is what you are using when you judge that one checkout line looks longer than another.
The ANS and the parallel individuation system are usually described as working side by side: the small-number system handles sets up to about four, and the ANS handles larger ones.
Number in other animals
Evidence from many species suggests that a sense of quantity is old in evolutionary terms. Many species of birds, mammals, fish and insects have been shown to discriminate magnitudes.
Some notable studies:
- Rats. In bar-pressing experiments, rats trained to press a lever a set number of times produced press counts that clustered in a bell-shaped (Gaussian) distribution around the target, for targets such as 8 or 16. That is the kind of approximate, noisy output the ANS predicts. By comparing well-fed and hungry rats, researchers were able to separate the influence of time from the influence of number.
- Lions. In a field study published in 1994, McComb, Packer and Pusey played recorded roars of unfamiliar lionesses through hidden speakers on the African savannah. Whether resident lionesses approached or withdrew depended on how many intruders they heard compared with how many companions they had. A lone lioness retreated when she heard three unknown calls, while groups of sisters went to investigate. Because the lions were judging number from sound, the study is often cited as evidence that the sense of number works across the senses.
- Fish. Agrillo's 2012 work on guppies found that they could discriminate between one and four other fish, evidence for a small-number tracking system in fish. In another study, university students and guppies performed a numerical task in nearly the same way, with performance in both depending on the ratio between quantities.
- Primates. Lemurs and several other primate species have been shown to tell groups apart by number.
The most detailed animal evidence comes from neurophysiology in monkeys, discussed below.
Number in infancy
Babies discriminate quantities
Infant researchers rely on the fact that babies look longer at things that surprise them. In habituation studies, an infant is shown many displays with the same number of items until interest fades, then shown a new number. If the infant's looking time recovers, the infant has noticed the change.
Using this method, Feigenson, Dehaene and Spelke (2004) described how six-month-old infants habituated to displays of 16 dots noticed a change to 8. Six-month-olds could discriminate a 2:1 ratio (8 versus 16, or 16 versus 32) but not a 3:2 ratio. Ten-month-olds succeeded at both. Infants' numerical acuity, in other words, sharpens over the first year, and this is consistent with an ANS that obeys Weber's law and becomes more precise with age.
Infants also seem to connect number across the senses. Izard and colleagues (2009) played infants sequences of tones and then showed them arrays of dots, using pairs such as 4 versus 12, 8 versus 16 and 4 versus 8. The infants looked longer at visual displays whose number matched what they had heard. This suggests an abstract, approximate sense of number that is not tied to one sense. Precision, measured as the Weber fraction, improves through development.
Can babies add?
The most famous infant number study is Karen Wynn's "Addition and subtraction by human infants," published in Nature in 1992 (volume 358, pages 749–750). Wynn used a violation-of-expectation design with five-month-olds. In a "1 + 1" condition, an infant saw a doll placed on a stage, a screen came up, and a hand visibly placed a second doll behind the screen. When the screen dropped, the infant saw either two dolls (the expected result) or one (an impossible result). A "2 − 1" condition worked in reverse. Infants looked longer at the impossible outcomes. In a further experiment, a 1 + 1 operation ended with either two or three dolls, and infants looked longer at three. Wynn took this to mean they expected an exact answer, not just "something different."
The study was influential and also widely debated. Critics proposed that infants might be tracking individual objects, or responding to familiarity, rather than doing arithmetic. Later work by Koleen McCrink and Wynn found that infants seem to compute exact outcomes only for small numbers, and approximate outcomes for larger operations such as 5 + 5 or 10 − 5. That fits the two-system picture: exact tracking for small sets, approximate estimation for large ones.
Innate number concepts or built through language?
Do children start life with the concept of natural number? Two views have dominated:
- Gelman and Gallistel (1978) argued that children have an innate grasp of the principles of counting and the concept of natural number. On this view, learning to count is mainly a matter of mapping inborn concepts onto the number words of one's language.
- Susan Carey (2004, 2009) argued that infant systems only represent large numbers approximately and small numbers (up to about four) exactly through parallel individuation. On her account, the concept of exact natural numbers in general is a cultural and linguistic construction that children build as they learn the counting list.
These positions also challenge an older view. Jean Piaget held that children do not have a stable concept of number until about six or seven. Research since the 1970s on infants and preschoolers has pushed back on that timeline, though there is still disagreement about what infant abilities really amount to.
What happens in the brain
The parietal lobe and the intraparietal sulcus
Neuroimaging consistently points to the parietal lobes, especially the intraparietal sulcus (IPS) and the inferior parietal lobule (IPL), as central to number. Stanislas Dehaene and colleagues proposed that these regions do complementary jobs. The IPS supports estimation, comparison and on-the-fly quantity processing, often tested using subtraction. The IPL, connected with language areas, supports retrieval of memorized facts such as multiplication tables.
Brain-injury cases fit this division. Dehaene worked with Laurent Cohen on patients with parietal lesions and found that multiplication and subtraction can break down separately. An IPL lesion can leave subtraction intact while impairing multiplication, and an IPS lesion can do the reverse. Imaging studies agree: Campbell and Xue (2001) found that multiplication draws on a left-lateralized network including the inferior frontal cortex and superior and middle temporal gyri, along with the IPL and IPS. Barrouillet, Mignon and Thevenot (2008) found that subtraction relies more on the right IPS and posterior parietal lobe.
Other regions also play a part. Calculation engages frontal areas that overlap with language (including Broca's area), working memory and attention. Regions of the inferotemporal cortex process written number symbols such as Arabic digits (Piazza and Eger, 2016).
Number neurons in monkeys
Between 2002 and 2005, Nieder and colleagues recorded from single neurons in monkeys performing a delayed match-to-sample task. A monkey saw a display of dots, held its number in memory, and then indicated whether a second display had the same number. Neurons in the IPS and prefrontal cortex showed preferred numerosities. A given neuron fired most for, say, three items, a little less for two or four, and less still for numbers further away. The tuning got broader for larger numbers, following Weber's law. This was some of the clearest evidence that the brain has neurons tuned to number.
Similar patterns in humans
Human brain-imaging studies found similar tuning. In a 2004 fMRI study, Piazza and colleagues showed people dot sets based on 16 or 32, with deviants at ratios of 1.25, 1.5 and 2. The IPS responses formed Gaussian tuning curves, matching a roughly logarithmic approximate number scale like the one seen in monkeys. Timing differences suggested that number is processed first in the IPS and then passed to prefrontal cortex.
Children show the same system early. Cantlon and colleagues (2006) scanned adults and four-year-olds while they watched dot arrays and occasionally saw a change in number. Both groups activated the IPS for the deviant numbers. Children activated mainly the right IPS, with fewer activated voxels than adults. This suggests a non-symbolic number mechanism is in place by age four. A near-infrared spectroscopy study cited in the literature has reported right parietal specialization for number even in preverbal infants.
Dehaene and the "number sense"
Dehaene, born in 1965 in Roubaix, France, has done more than anyone to bring these findings together for a wide audience. His book The Number Sense, published in English by Oxford University Press in 1997 and based on his French book La Bosse des maths, argued that humans inherit an approximate sense of quantity shared with other animals, and that culture builds exact mathematics on top of it. He has directed the INSERM cognitive neuroimaging unit and in 2005 was elected to the chair of Experimental Cognitive Psychology at the Collège de France. With Jean-Pierre Changeux he built a computational model of number neurons. In 2014 he shared The Brain Prize with Giacomo Rizzolatti and Trevor Robbins.
Number, space and the mental number line
Number and space are closely tied in the mind. In 1880 Francis Galton described number-form synaesthetes: people who automatically experience numbers as laid out in a particular spatial pattern.
A more general link appears in the SNARC effect (Spatial-Numerical Association of Response Codes), described by Dehaene, Bossini and Giraux in 1993. When people decide whether a number is odd or even, they respond faster with the left hand to small numbers and faster with the right hand to large numbers, even though size is irrelevant to the task. This is often described as evidence for a mental number line running from left to right.
The direction of the effect depends on culture and context. Fischer, Mills and Shaki (2010) showed that it varies with things like reading direction. Núñez, Doan and Nikoulina (2011) questioned whether it reflects a built-in association at all, and suggested it may reflect strategies or learned conceptual metaphors. Imaging studies have found that parietal regions active for number overlap with regions involved in spatial processing (Hubbard and colleagues, 2005).
Number in everyday judgment
Numerical cognition also helps explain everyday judgments about prices and quantities. Thomas and Morwitz (2009) linked three well-known judgment heuristics to specific numerical effects:
- The left-digit effect (anchoring). People judge the gap between $4.00 and $2.99 as larger than the gap between $4.01 and $3.00, even though both are $1.01. The leftmost digit acts as an anchor, which helps explain why so many prices end in .99.
- The precision effect (representativeness). Precise-looking numbers can seem more credible or attractive. A price of $391,534 may be judged differently from a round $390,000.
- The ease-of-computation effect (availability). Differences that are easy to calculate, such as $5.00 − $4.00, can be judged larger than harder ones like $4.97 − $3.96, even when the hard difference is larger.
Language, culture and number
Studies of cultures whose languages have few number words test how much exact number depends on language.
The Pirahã
The Pirahã of the Brazilian Amazon have become a central case. The linguist Daniel Everett at first reported that Pirahã had words for "one" (hói) and "two" (hoí), told apart only by tone. In a 2005 analysis he argued instead that the language has no true number words, and that these terms mean something more like "small quantity" and "larger quantity." Everett reported that after eight months of lessons, no Pirahã adult learned to count to ten or add 1 + 1. He attributed this to culture and lack of need, not to any inability, and did not claim the Pirahã cannot learn to count.
In a 2004 Science paper, psychologist Peter Gordon tested Pirahã speakers on non-verbal matching tasks. He reported that accuracy dropped once quantities went above three and fell further as numbers grew; in his words, "They couldn't keep track at all." Susan Carey called it "a spectacular finding," while Lisa Feigenson of Johns Hopkins cautioned that "the jury is still out" on how much language shapes thought. He presented this as support for a strong form of the idea that language shapes thought. Pirahã adults have also been described as unable to mark an exact tally for a pile of fewer than ten nuts.
A 2008 follow-up by Frank, Everett, Fedorenko and Gibson worked with four Pirahã speakers. When spools were added one at a time, all four used hói for one and hoí for two, but for larger numbers they mixed hoí with a word for "many." When spools were taken away from ten, one speaker was already using hói when six remained. The authors concluded the words are much more likely to be relative terms, like "fewer" and "more," than exact words for "one" and "two." They also found that Pirahã speakers could match exact quantities when the objects stayed in view, but struggled when the quantities were hidden before matching. That points to the role of number words as a memory tool for holding exact quantities.
The Munduruku
Another Amazonian group, the Munduruku, has number words only up to about five. In a 2004 Science study, Pierre Pica, Cathy Lemer, Véronique Izard and Stanislas Dehaene found that Munduruku participants could compare and add large approximate quantities far beyond their naming range, but had difficulty with exact arithmetic on numbers larger than four or five. The authors argued for a distinction between a non-verbal system for approximating number, shared across cultures, and a language-based counting system needed for exact number and arithmetic. Dehaene studied the Munduruku together with Pica and Elizabeth Spelke.
Counting and ways of life
Ethnographic accounts point the same way. The anthropologist Napoleon Chagnon concluded that the Yanomami had little need for counting. Hunters, for example, kept track of their arrows as individual, recognizable objects, using the same skills they used to recognize family members, not by number. Steven Pinker (2008) has argued that no known hunter-gatherer culture has a full counting system, and has linked the spread of numeracy to agriculture, trade and the need to manage large numbers of interchangeable items. Results from different cultures do not all agree: Pica and colleagues (2004) and Butterworth and colleagues (2008) both studied groups without full sets of number words, with mixed results that complicate a simple "no words, no exact number" conclusion.
Dyscalculia: when number sense struggles
Dyscalculia is a learning disorder involving persistent difficulty with arithmetic: understanding quantities, manipulating numbers, doing calculations and learning number facts. It is sometimes called "math dyslexia." It is considered lifelong, and in the United Kingdom it is classified as a specific learning difficulty. The American Psychiatric Association's diagnostic manual describes it under the heading of a specific learning disorder with impairment in mathematics.
The term dates back at least to 1949. Slovak researcher Ladislav Košč brought it wider recognition in 1974, defining it as "a structural disorder of mathematical abilities." Estimates of prevalence range from about 3% to 6% of the population. Many studies find similar rates in males and females, though results vary.
Two main explanations compete:
- The magnitude-representation (number module) hypothesis holds that the core problem is an imprecise intuitive number sense, a weak ANS.
- The access-deficit hypothesis holds that people with dyscalculia understand quantities but struggle to link them to symbols such as digits and number words.
Evidence for both is mixed. Dyscalculia has been linked to differences in the intraparietal sulcus, the same region central to number processing in imaging studies. It often occurs alongside dyslexia and ADHD, and has been associated with conditions including Turner syndrome. There is no agreed set of diagnostic criteria. Clinicians usually combine achievement tests, tests of related skills such as working memory, and teacher reports. Brain imaging can tell groups apart in research, but it is too costly for routine diagnosis.
Does number sense predict math achievement?
A major practical question is whether individual differences in the ANS matter for learning mathematics. Halberda and colleagues (2008) reported that people's non-verbal number acuity correlated with math achievement. Later studies reported that preschool ANS precision predicted later school math performance, and that impaired ANS acuity can distinguish children with dyscalculia from other low-achieving children.
The picture is not settled. Longitudinal studies do not consistently find that non-symbolic abilities predict later symbolic ones. In some cases early symbolic skills, such as knowing number words and digits, predicted later non-symbolic acuity instead, which reverses the expected direction. In adults, non-symbolic number ability does not always explain differences in math achievement. Whether training the ANS could improve school mathematics therefore remains an open research question, not an established fact.
Open questions
Several debates remain live:
- How many core systems are there? The two-system picture (small exact plus large approximate) is widely used, but the boundaries between subitizing, object tracking and attention are still argued over.
- What is innate? Researchers disagree about whether infants have number concepts or simply sensitivity to quantity-related cues, and about how much of exact number depends on language.
- Is number perceived directly? Some researchers argue that people perceive number directly. Others propose that judgments rely partly on related cues such as total area or density. (The approximate-number literature notes this concern, which is one reason experiments carefully control dot size and spacing.)
- How do symbols attach to quantities? The step from an approximate sense of "about seven" to the exact symbol 7 is perhaps the central developmental puzzle, and it bears directly on education and dyscalculia.
- How culturally variable is the number–space link? Findings on the SNARC effect show that even apparently basic associations are shaped by culture.
Research in the area is published in many psychology and neuroscience journals, and the field has its own open-access outlet, the Journal of Numerical Cognition.
Frequently asked questions
What is numerical cognition in simple terms? It is the scientific study of how people and animals perceive, represent and work with numbers, from instantly seeing that there are three apples to doing long division.
What is the difference between subitizing and counting? Subitizing is the instant, accurate recognition of small quantities, usually up to about four. Counting is slower and step by step, and is needed for larger sets unless they form a familiar pattern. Reaction-time studies, afterimage experiments and patients with simultanagnosia all suggest the two rely on partly different processes.
Do animals really understand numbers? Many animals, including rats, lions, fish, birds and monkeys, can discriminate quantities, usually approximately and following Weber's law. Monkeys have neurons tuned to specific numbers. This is not the same as symbolic arithmetic, which appears to require language and culture.
Can babies do math? Infants can tell large quantities apart by ratio, and that ability sharpens during the first year. Wynn's 1992 study suggested five-month-olds have expectations about simple additions and subtractions of small sets, though the meaning of those results is still debated.
Where in the brain are numbers processed? The intraparietal sulcus is central to quantity processing. The inferior parietal lobule, frontal areas and language regions help with memorized facts and calculation, and inferotemporal areas help recognize written digits.
Is dyscalculia the same as being bad at math? No. Dyscalculia is a specific, persistent learning disorder affecting an estimated 3–6% of people, not ordinary difficulty with a subject. It has been linked to differences in parietal brain regions and often occurs alongside dyslexia or ADHD.
Do you need words to have numbers? Studies of the Pirahã and Munduruku suggest that approximate number does not need words, but exact representation of larger numbers, and especially exact arithmetic, depends heavily on a counting system supplied by language and culture.
Sources and further reading
- Wikipedia, "Numerical cognition": https://en.wikipedia.org/wiki/Numerical_cognition
- Wikipedia, "Approximate number system": https://en.wikipedia.org/wiki/Approximate_number_system
- Wikipedia, "Subitizing": https://en.wikipedia.org/wiki/Subitizing
- Wikipedia, "Dyscalculia": https://en.wikipedia.org/wiki/Dyscalculia
- Wikipedia, "Pirahã language": https://en.wikipedia.org/wiki/Pirah%C3%A3_language
- Wikipedia, "Stanislas Dehaene": https://en.wikipedia.org/wiki/Stanislas_Dehaene
- Wikipedia, "Karen Wynn" (on the 1992 Nature study): https://en.wikipedia.org/wiki/Karen_Wynn
- Nature news (19 August 2004), on Peter Gordon's Pirahã study: https://www.nature.com/news/2004/040816/full/news040816-10.html
- Pica, P., Lemer, C., Izard, V. & Dehaene, S. (2004), "Exact and approximate arithmetic in an Amazonian indigene group," Science 306 (PubMed record): https://pubmed.ncbi.nlm.nih.gov/15486303/