Research on arithmetic and the brain

A survey of our research and main publications in the cognitive neuroscience of arithmetic


Back to the main Unicog page




In “The Number Sense”, Stanislas Dehaene proposed that the human capacity for arithmetic finds its ultimate roots in a basic cerebral system for perception and mental manipulation of approximate numbers, very ancient in evolution. According to this theory, we share this system with many animal species, and it appears very early in human development, independently of language. Of course, it is a primitive system, capable only of basic computations such as estimation, comparison, addition and subtraction of approximate numbers. On this shared basis, various human cultures invent increasingly elaborate cultural tools such as Arabic symbols, counting routines, algorithms for exact addition, multiplication etc.


Thus, the origins of human arithmetic lie in both a universal core system of approximate quantity, and on various cultural tools for exact arithmetic


Our laboratory uses several methods to test this theory: neuroimaging, neuronal modelling, mental chronometry, studies of patients with acalculia or dyscalculia, and studies of populations in the Amazon.


For accessible reviews of our work, you might consult the following publications:


Interview for the Edge website:


1. Neuroimaging



We use brain-imaging techniques such as fMRI to identify the human brain responses during arithmetic tasks. Our research has identified a brain area, the horizontal segment of the intraparietal sulcus, as playing an important role in the quantity representation.


The following paper provides a review of this line of research:



Recent specialized publications on neuroimaging of arithmetic include:


2. Neuronal modeling and cross-species comparisons



In 1993, Stanislas Dehaene and Jean-Pierre Changeux developed a neuronal network model of number processing, which made the prediction that the parietal cortex should contain “numerosity detectors”. These are neurons tuned to a specific number, and thus firing preferentially for instance to sets of 3 objects. The model explained why such a code led to classical behavioral findings such as the distance effect and Weber’s law. The model also predicted that these neurons should have a Gaussian tuning curve when plotted on a logarithmic axis of number size.


In 2002, these predictions were beautifully verified by Andreas Nieder and Earl Miller in the macaque monkey. Their work was published in a landmark series of papers in Science, Neuron, PNAS, and Journal of Cognitive Neuroscience. For accessible descriptions of their work, you may consult:


Following Nieder and Miller’s work, we used fMRI to demonstrate a similar sensitivity to numerosity in the human intraparietal sulcus, using an fMRI adaptation method:



The possibility of looking at number at the single-neuron level opens up many possibilities for further research. Particularly important is the prospect of being able to characterize the similarities and differences between the human and animal codes for number, and to investigate the changes induces by the acquisition of language and symbols. While this research is only in its earliest stages, we have proposed homologies between human and macaque cortical maps in the following papers:



In collaboration with Marc Hauser in Harvard, we have also studied behavioral paradigms for numerosity discrimination in the Tamarin monkey:




3. Mental chronometry


In spite of important advances in neuroimaging, chronometric methods remain essential to study the organization of mental representations in humans. In the past, we have published several papers that characterized the behavioral signature of the quantity representation. Following the seminal work of Moyer and Landauer (1967), we showed in particular that, even when numbers are presented as Arabic digits, a representation of their quantity is automatically activated and leads to distance effects in tasks such as number comparison or same-different judgements. In 1993, we also discovered and named the SNARC effect – a Spatial-Numerical Association of Response Codes which demonstrates that numbers are automatically associated with positions in space, thus supporting the metaphor of a spatially organized internal “number line”.


In 1998, Lionel Naccache, Etienne Koechlin and Stanislas Dehaene introduced subliminal priming combined with neuroimaging in the number domain. This work showed that Arabic numerals or spelled-out number words could be flashed for a very short duration and, in spite of their invisibility, could facilitate the processing of other numbers presented subsequently. This method has become a very useful tool to probe the organization of number representations, but also to study the differences between conscious and non-conscious processing – now a main topic of research in the lab.


4. Acalculia


Deficits of number processing can be extremely informative about the cerebral organization of number processes. Stanislas Dehaene and Laurent Cohen have collaborated on many years on single-case studies of adult patients with selective dysfunctions of number processing and calculation (acalculia). This work has revealed important dissociations between operations, particularly between exact arithmetic based on rote verbal learning (e.g. multiplication tables), and other operations such as subtraction or estimation which seem to involve internal manipulations of quantities.





5. Dyscalculia

A prediction of the number sense hypothesis is that there might exist children with very early deficits of number sense. Early injuries to the parietal lobe, due to perinatal injuries or to genetic causes, should lead to a disruption of number sense and therefore to a lifelong impairment in arithmetic. Thus, our lab has a growing interest in dyscalculia – impairments of number processing and calculation in children with other normal education and intelligence. Our work has largely focused on characterizing the dyscalculia that occurs in Turner’s syndrome, a genetic disease associated with loss of one X chromosome. In collaboration with Ann Streissguth’s team in Seattle, we have also shown that foetal alcohol syndrome (exposition of the foetus to alcohol during pregnancy) is often accompanied by striking deficits of number processing.


We have developed and are now testing a new tool for rehabilitation of dyscalculia: a computer game ("The Number Race") designed especially to help children practice their quantity manipulation skills and the links from quantity to Arabic numerals and number words. Anna Wilson and Stanislas Dehaene have designed an adaptive computer algorithm that can detect in which domain a child experiences difficulty, and present problems at the relevant level, difficult enough to be challenging, but easy enough to avoid discouragement. This software is open-source (under a GNU public licence) and can be downloaded here from our website.

For more information on dyscalculia, visit this page.



6. Amazon




Recently, a new exciting opportunity to test the number sense hypothesis has emerged with a collaboration with Pierre Pica, a linguist at CNRS (UMR 7023 “Formal Structures of Language”) and who collaborates with the linguistic section of the Brazilian National Museum (Rio de Janeiro). Pierre Pica regularly visits Brazil where he studies the indigenous populations of the Amazon. He has particularly focused on the Mundurukú, who live in an autonomous territory of the state of Para. The Mundurukú have a very limited vocabulary of number words, essentially limited to words for one, two, three, four and five. Studying this people provides a unique opportunity to investigate the capacity for arithmetic processing in the absence of a well-developed language of number words.


Of course, it is not possible to study the Mundurukú with neuroimaging or neuropsychological methods. However, Pierre Pica, Cathy Lemer, Véronique Izard and Stanislas Dehaene developed a series of behavioral tests which were run on a solar-powered computer with 55 native Mundurukú subjects. The results, which appeared on October 15th 2004 in the journal Science, indicate that, in spite of their reduced lexicon, the Mundurukú have a well-developed capacity for numerical approximation. They can estimate, add and subtract approximate numbers presented as sets of dots, even when the numbers range all the way up to 80 or more. However, their lack of a counting routine prevents them from being able to perform exact calculations even with small numbers. For instance, they were unable to report the precise outcome of 6 – 4. Thus, the results help tease apart which aspects of our arithmetic competence are universal, and which are dependent on cultural and linguistic tools.


Our original Science publication is



A press release describing this research can be found at


Several movies illustrating the difficulties that the Mundurukú have when attempting to count are available on


Explanations about these movies can be found onú.htm


 (all photos of the Amazon are © Pierre Pica and CNRS)














Back to the main Unicog page