Introduction

Number and space are inherently related in everyday life in cultures with formal education systems. From calendars to measuring instruments we are used to seeing numbers mapped onto space. Since Galton’s initial systematic description of individuals reporting vivid associations between numbers and space1, subsequent studies have consistently demonstrated that a majority of Western educated adults exhibit a robust tendency to associate small numbers with the left side and large numbers with the right side of space. Among others, one of the most studied and reported evidence of this Spatial Numerical Association (SNA) phenomenon is the Spatial-Numerical Association of Response Codes (known as SNARC). First reported by Dehaene and colleagues in 1993, it specifically refers to being faster and more accurate to associate small numbers with the left hand/side of the space and large numbers with the right hand/side of the space2. The strength and the direction of the SNARC effect, however, have been shown to be malleable to reading habits and contexts: while left-to-right readers systematically show a left-to-right SNARC3,4,5,6, participants raised in cultures with writing systems that are organized from right-to-left or top-to-bottom tend to show no, inverted, or top-to-bottom SNARC7,8,9,10. While the precise contribution of reading and writing direction per se is still under debate11, also other cultural and contextual features seem to influence SNARC effects. For example, adding a clockface in the background and making time salient for the task12,13 can reverse SNARC (i.e., participants associate small numbers with the right and large numbers with the left, as depicted on the clock), and performing a short session of finger counting in a non-canonical order can also influence SNARC14. Additional evidence in favor of the important influence of cultural practices like reading on the direction of the number-to-space mapping comes from the study on populations lacking formal education. In a recent study, Pitt et al. (2021) examined a group of adult Tsimanes, an indigenous culture with an oral tradition from Bolivia, and a group of U.S. preschoolers, to investigate the directional bias when organizing cards depicting 1 to 5 dots (or 2, 4, 8, 16, 32 in a second experiment) based on their numerical value. Both populations exhibited a lack of consistent directional preference, with individuals equally likely to arrange the cards in either a left-to-right or a right-to-left order. These outcomes suggest that the direction of the number-to-space mapping is largely determined by cultural inputs, and that in the absence of cultural influence, mental mappings are “direction-agnostic,” as concluded by the authors15.

This conclusion, however, seems to stand in stark contrast with an independent growing body of evidence from both human infant and non-human animal studies indicating an early and strong culture-independent direction-specific (left-to-right) bias in associating number to space. Starting from birth, humans preferentially orient towards the left when experiencing a decrease and towards the right when experiencing an increase in numerosity16,17. Similarly, in their first year of life, infants prefer increasing sequences of dots moving from left to right (7 months old)18 and orient their attention to the left after being cued with a small numerosity and to the right after a large one (8 months old)19. Studies with non-human animals (for example, Clark’s nutcrackers (Nucifraga Columbiana)20, domestic chicks (Gallus gallus domesticus)20,21,22, and honeybees (Apis mellifera)23) also indicate an association between small quantities to the left side of space and large quantities to the right side. A simple interpretation of these findings is the existence of a universal biologically determined and possibly innate mechanism that predisposes to a left-to-right oriented SNA. How can we reconcile these findings with the previously reported absence of specific directional SNA in preschool children and in populations with no reading-writing systems?

One possibility is that, in humans, two dissociable kinds of SNAs co-exist: one that is biologically determined, universal, and potentially innate and has a left-to-right directional bias, and another that is learnt and that is shaped by the directionality of specific cultural input, such as reading and writing. According to this hypothesis, while the biologically determined SNA should manifest itself very early in life and persist throughout the lifespan, independent of both maturation and cultural influences, the culturally determined SNA should only emerge in the presence of strong and well-integrated cultural practices such as reading/writing24,25. In addition to our hypothesis, to account for the contrasting findings in the literature, we propose that the biologically determined SNA functions as a behavioral reflex and thus primarly manifests itself in tasks that do not require overt decision-making regarding SNA. In contrast, tasks that overtly and explicitly require ordering numbers in space are mostly influenced by culturally determined SNA. Therefore, the behavioral paradigms and/or tasks used to test the directional biases of SNAs might be key in differentiating between two different kinds of SNAs. Indeed, animal and infant studies necessarily rely on implicit procedures to examine associations between numbers and space, whereas studies on children and adults typically require explicit numerical judgments15. The idea that different kinds of SNA manifest themselves in different tasks is not new in the literature. Patro and colleagues initially proposed a taxonomy classification of different SNA kinds, specifically referring to preschooler studies26, later expanded by Cipora and colleagues to adult studies27. In particular, they distinguish SNA phenomena into two main categories, extension and directional, with the latter specifically referring to tasks where the directional processing of numerical orders is relevant, rather than the magnitude value per se. Within this category, the authors report the existence of implicit and explicit coding SNA based on the type of tasks required to study them (see27 for extended definitions). Similarly, in the present work, we will use the terms explicit and implicit types of SNA indicating the explicit/implicit reference to space in the instruction of the task. Our main aim is therefore to understand if different SNA emerge in implicit and explicit tasks and if the two can be dissociated. In particular, we expect that in implicit tasks, left-to-right SNA emerge independent from age and cultural factors, while in explicit tasks, requiring a conscious ordering of numbers in space, left-to-right SNA only emerge in those populations that have received a clear cultural left-to-right spatial ordering input.

In this work, we test space-number mapping, using both explicit and implicit tasks, in three different populations: Italian adults, Italian preschoolers, and Himba adults, a population with only oral culture (no reading/writing) from Northern Namibia, with limited mathematical knowledge and no formal schooling, tested in two separate field trips (see Methods section). While Italian and Himba adults are equivalent in the overall level of brain maturation, Italian preschoolers and Himba adults are equivalent in terms of their lack of, or extremely reduced, formal education. According to our hypothesis, we observe a similar SNA pattern across the three populations in the covert-implicit task, and a different pattern of SNA between Italian adults and the other groups in the overt-explicit task. Specifically, in the implicit task, we find a left-to-right SNA irrespective of age and culture (thus present in all three groups), while in the explicit task, a systematic left-to-right oriented SNA appears only in the adult literate group.

Results

Experiment 1. Explicit space number associations

To investigate how number and space are associated in an explicit task, we asked participants to manually order 10 cards, each depicting 1 to 10 dots, on the table in front of them, with the only constraint that they should arrange the cards so that they were well ordered according to their judgment (Fig. 1A). In all three groups, the majority of participants spontaneously chose a lateral linear spatial arrangement (left/right), while other spatial layouts chosen were non-lateral linear arrangements (diagonal or vertical), 2-D geometric configurations (grids or circles) and no geometric shape (random arrangements). Figure 1B reports the percentages of choice per group (see Table S1 in Supplementary for detailed contingency table and χ2 analysis). We then focused on the data from those participants who organized the cards laterally (left/right direction) and we analysed their mapping scores: for all groups the distribution of their mappings differed from chance, indicating that participants in all groups spontaneously understood the task and arranged the cards based on their numerosity and not randomly (Two-tailed two sample Kolmogorov-Smirnov goodness-of-fit hypothesis test; Italian Adults: N = 38, D = 0.957, pvalue <0.001, Himba Adults 2021: N = 60, D = 0.913, pvalue <0.001, Himba Adults 2022: N = 70, D = 0.913, pvalue <0.001, Italian Preschoolers: N = 28, D = 0.957, pvalue <0.001; Fig. 1C). However, only Italian adults systematically organized numerosities monotonically from left to right. By contrast, neither Himba adults nor Italian preschoolers showed such a systematic rightward bias, in that they were equally likely to order the cards from right to left and from left to right, as also confirmed by Bayesian analyses that reported moderate evidence against the alternative hypothesis (μ ≠ 0) for all populations, with the exception of Italian Preschoolers that reported anecdotal evidence (Table 1; Fig. 1C). This result replicates the one reported by Pitt et al. in 2021 and confirms that, for this explicit mapping task, in the absence of strong cultural bias, the mapping between number and space is arbitrary and inconsistent across individuals.

Fig. 1: Experiment 1—Ordering Card Task.
figure 1

A A picture taken on the field of one Himba participant during the Ordering Card Task. In this task, participants were asked to order 10 cards, depicting 1–10 dots, on the table in front of them, such that they would look in order. No other instruction was given. B Percentage of choice for each type of spatial arrangement in the four groups. Lateral line (red) refers to a lateral disposition, Random (brown) to an absence of identifiable configuration (e.g. random scatterplot), Sagittal/Diagonal (orange) to non-lateral lines, and Two Dimensional (yellow) to a configuration which was identifiable but not linear (i.e. grid, square, circle; Table S1). C Kendall Tau correlation coefficient between ideal left-to-right order and participants’ disposition was calculated: negative values indicate leftward bias, positive values indicate rightward bias, and values close to 0 indicate not-ordered mappings (see Methods section for a detailed explanation of the analysis). For all groups, the mapping distributions (red) are different from the chance distribution (gray; Two-tailed two-sample Kolmogorov-Smirnov goodness-of-fit hypothesis test; Italian Adults: N = 38, p < 0.001, Himba Adults 2021: N = 60, p < 0.001, Himba Adults 2022: N = 70, p < 0.001, Italian preschooler: N = 28, p < 0.001). Black dots and whiskers show the mean mapping scores and the standard error of the mean. Only Italian Adults showed a consistent left-to-right preference at group level (Two-tailed one-sample Wilcoxon signed rank test; Italian Adults: N = 38, p < 0.001, Himba Adults 2021: N = 60, p = 0.257, Himba Adults 2022: N = 70, p = 0.805, Italian Preschooler: N = 28, p = 0.188; Table 1).

Full size image
Table 1 Statistical analysis against chance level for mapping scores in Experiment 1
Full size table

Experiment 2. Implicit space number associations

To investigate an implicit number to space mapping we employed a computerized Numerosity Comparison Task using a Go/No-Go paradigm. After the presentation of a first set of dots (reference stimulus), participants were required to press a central button when a second set of dots (test stimulus) was smaller (Decreasing task) or larger (Increasing task) than the reference stimulus. Critically, while the reference stimulus was presented in the center, the test stimulus was presented either on the left or on the right side of the screen. We defined targets as Congruent or Incongruent based on their quantitative relation with the reference and their spatial location: targets that were smaller than the reference and appeared on the left, as well as targets that were larger and appeared on the right, were defined as congruent, whereas smaller right targets and larger left targets were defined as incongruent. Congruency was thus defined with respect to the canonical left-to-right orientation (Fig. 2B). Using the Inverse Efficiency Score as a composite summary measure of reaction times and accuracy, we found that in contrast to the explicit task, where Himba adults and Italian preschoolers behaved differently from Italian adults, all groups exhibited a consistent congruency effect in the implicit task, but only for the Decreasing Task instruction. Specifically, performance was significantly better when smaller numerosities were presented on the left than on the right side of the screen (Fig. 2C).

Fig. 2: Experiment 2—Numerosity Comparison Task.
figure 2

A Stimuli used in the Numerosity Comparison Task to represent 4, 12, and 36 dots. B A schematic representation of one exemplar trial in the Numerosity Comparison Task. In the Decreasing Task instruction, participants were instructed to press a central key as fast as possible (using their dominant hand) only when the test numerosity was smaller than the prime numerosity. In the Increasing Task instruction, instructions were reversed. In this trial, the test numerosity is smaller and appears on the right side of the screen, corresponding to an incongruent condition. All participants but the Himba tested in 2021 performed both tasks, in counterbalanced order. In the group of Himba tested in 2021, half the participants performed the task with the Increasing instruction and the other half with the Decreasing instruction. See Methods section. C In the implicit task (B) we measured participants’ performance in terms of a combined measure of speed and accuracy (Inverse Efficiency Score, IES), and we calculated the Congruency Effect for each Task instruction (see Methods section for the detailed analysis, and Data distribution in Fig. S2). In the graph individual points, mean values, standard errors of the mean, and significant pvalues are reported for each group and Task instruction (Two-tailed one sample t test and two-tailed Wilcoxon signed rank test; Decreasing condition (red): Italian Adults: N = 47, p < 0.001, Himba Adults 2021: N = 33, p =0.004, Himba Adults 2022: N = 78, p = 0.026, Italian Preschoolers: N = 38, p = 0.008; Increasing condition (gray): Italian Adults: N = 47, p = 0.505, Himba Adults 2021: N = 26, p = 0.805, Himba Adults 2022: N = 78, p = 0.119, Italian Preschooler: N = 38, p = 0.421; see Tables 3, 4). Significance levels are defined as follows: * = pvalue <0.05, ** = pvalue <0.01, **** = pvalue <0.0001).

Full size image

These conclusions were statistically supported by mixed ANOVA (Group x Task instruction) on the congruency effect, which showed a main effect of Task instruction, but no main effect of Group nor a Group x Task instruction interaction (Table 2).

Table 2 Analysis of variance on the congruency effect by Group, Task instruction, and Congruency condition
Full size table

In all groups, the congruency effect was significantly greater than zero only for the Decreasing Task instruction (Table 3), and not for the Increasing Task instruction as confirmed by Bayesian analyses that reported moderate evidence against the alternative hypothesis (μ ≠ 0) for all groups but the Himba adults 2022 for whom the evidence was anecdotal (see Table 4, Fig. 2C). Figure 2C displays means and standard errors for each Task instruction and Group separately to better appreciate the consistent pattern across the different groups.

Table 3 Statistical analysis against chance level for Congruency Effect for each population in the Decreasing Task instruction of Experiment 2
Full size table
Table 4 Statistical analysis against chance level for Congruency Effect for each population in the Increasing Task instruction of Experiment 2
Full size table

In order to verify the strength of our conclusions we also ran a second set of analyses on the raw inverse efficiency data without pre-computing the congruency effect, using mixed-effect models. For each subject and each condition, we tested the Inverse Efficiency Scores against three predictors: Congruency condition (Congruent vs. Incongruent), Task instruction (Decreasing vs. Increasing), and Group (Himba Adults 2022, Himba Adults 2021, Italian Adults, Italian Preschoolers), and their interactions. The results indicated a significant main effect of Congruency condition (\({{{{\rm{\beta }}}}}_{{{{\rm{Congruency\; condition}}}}}\) =–0.107, SE = 0.445, z = 2.403, p = 0.016) together with a trend for a Congruency condition and Task instruction interaction (\({{{{\rm{\beta }}}}}_{{{{\rm{Congruency\; condition}}}}*{{{\rm{Task\; Instructions}}}}}\,\)= 0.111, SE = 0.060, z = 1.846, p = 0.065). No other interactions appeared. This analysis confirms and strengthens the previous results indicating that in all groups, participants were more efficient in detecting decreasing numerosities when they appeared on the left side and increasing numerosity when they appeared on the right side, and that this effect tended to be more prominent when the task was to detect a decreasing sequence. Group factor also was significant as a consequence of the notable difference in accuracy and reaction times between the Italian adults and the other groups (see Supplementary Materials for detailed models’ output). We also performed mixed-effect analyses on the reaction time and error rate separately (see “Methods” section for more detailed descriptions of the models). Error rate results consistently highlighted a significant main effect of Congruency condition (\({{{{\rm{\beta }}}}}_{{{{\rm{Congruency\; condition}}}}}\) = 0.587, SE = 0.278, z = 2.107, p = .035), whereas this effect was not present in RTs (\({{{{\rm{\beta }}}}}_{{{{\rm{Congruency\; condition}}}}}\) = –0.0615, SE = 0.0391, t = –1.575, p = 0.115). No Congruency condition per Group interaction was found significant in any model for any variable of interest. In the RTs model the only significant variable was Group, as adults were way faster compared to all other groups (see Supplementary Materials for detailed models’ output).

Taken together, these results indicate that when participants are required to process numerosities, independently of age and education they show a congruency effect, being more efficient in processing increasing numerosity when presented on the right and decreasing numerosities when presented on the left side of space. This effect appears stronger for the decreasing task and more prominent in error rate.

Explicit and Implicit tasks are not correlated

As a final analysis, to further probe the different nature of implicit and explicit behaviors, and to further demonstrate the independence of one behavior from the other, we tested whether the tendency to map numbers onto space in one direction of Himba adults and Italian preschoolers in one task predicted tendency to map number to space in the other task (Italian Adults have been excluded from the analysis due to the extreme skewness of their data, see Fig. 1C). We correlated the explicit mapping scores with the implicit congruency effect for each task instruction and for each participant (Fig. 3). No correlation was found between the two measures, as also confirmed by Bayesian analyses that reported moderate evidence against the alternative hypothesis (τ ≠ 0) (Two-tailed Kendall Tau correlation; for Decreasing Task instruction, N = 110, τ = –0.073, p = 0.286, 95% CI = [–0.209 0.064], log10(BF) = –0.633; for Increasing Task instruction, N = 110, τ = 0.017, p = 0.799, 95% CI = [0.122 0.157], log10(BF) = –0.889). These results further suggest that the explicit and implicit performances are unlikely to be related, as they might be the result of two qualitatively different forms of SNAs.

Fig. 3: Correlation between the two tasks.
figure 3

The correlation between Mapping Scores (x axis) and Congruency Effect (y axis) is not significant for either of the two task instructions (Two-tailed Kendall Tau correlation; Decreasing Task instruction (red), N = 110, p = 0.285; Increasing Task instruction (gray), N = 110, p = 0.799). The graph displays the linear regression with 95% confidence intervals, and individual points for each participant who completed both tasks (circles refer to Himba Adults 2021, triangles to Himba Adults 2022, and squares to Italian Preschoolers). The analysis included only the participants for which it was possible to calculate the Kendall tau scores (i.e. those who ordered the cards on a lateral line, see Fig. 1C), moreover, Italian adults were excluded from the analysis due to the skewness in their data (98% of the scores were perfectly equal to 1, see Fig. 1C).

Full size image

Discussion

The origins of the Spatial Numerical Associations effects are currently a major topic of debate. One view holds that all SNA phenomena are determined by cultural factors, in the absence of which the number to space mappings are “direction agnostic”15. This view, however, clashes with growing evidence that non-human animals and pre-verbal humans do display a clearly left-to-right directional SNA. Thus, the specific role of culture in the emergence of SNA phenomena remains to be understood. In the past, conflicting findings have been reported regarding SNA during the critical age when children begin schooling. Notably, while some studies suggest that children at this age do not consistently exhibit a left-to-right oriented space-number mapping or SNARC effects before starting primary school28,29,30, others demonstrate that SNA can emerge when small adjustments are made to the task features. For instance, using non-symbolic instead of symbolic stimuli31 or indirectly accessing magnitude information through a color discrimination task32 has been shown to bring out these effects, highlighting how task design can significantly emphasize or hinder SNA effects within the same age. We think that our results on Himba adults, pre-school kindergarten, and Italian educated adults can help to shed light on this debate. We found that an implicit task can reveal a clear left-to-right oriented SNA in participants who, when given an explicit task, either show no SNA or exhibit an SNA with an opposite orientation. Specifically, a task requiring explicit ordering of numbers in space elicited a systematically left-to-right oriented SNA only in Italian educated adults, while in a task where space was implicitly elaborated, we found evidence for a consistent left-to-right oriented SNA in all three tested populations. The experiment using the explicit task not only replicated a previous study similarly conducted in another Indigenous population15, but it extended those findings. Indeed, while in Pitt and colleagues’ experiment participants were instructed to order the numerosity in line and specifically on the lateral axis, in our experiment there were no instructions about the spatial disposition to use. Our results show that in all the populations the majority of the participants spontaneously chose a lateral linear arrangement. This result seems to suggest that the lateral linear organization of numerosities might not be strictly related to cultural factors. Further experiments are needed to investigate which biological or environmental factors might support the emergence of this spontaneous choice of a lateral arrangement of numerosities and if and to what extent this explicit behavior is consistent within the participants over different testing sessions.

The main scope of our study was to investigate the influence of cultural factors and of task demands on SNA. Therefore, we conducted two different experiments, one with an explicit spatial component and one with an implicit one, to investigate the influence of cultural factors and conscious decision-making on SNA. While the cultural background influenced performance only in the explicit spatial mapping task, it had no influence on the response patterns in the implicit spatial mapping task. Indeed, Himba adults and Italian preschoolers both mapped numerosities into space when explicitly asked, but without a systematic directional bias at the group level. A left-to-right directional bias was instead present in all groups when they were implicitly tested. As suggested by other researchers, this could mean that consistent directional SNA requires specific cultural practices or experiences to emerge33, but this is true only when the request is explicit and participants need to make conscious decisions on the mapping. In the context of implicit tasks instead, we believe that humans, as well as many animals, may be endowed with a biologically grounded mechanism by which the perception of changes in numerosity (and possibly other quantities) triggers specific lateralized spatial attentional shifts, thus showing an implicit form of SNA that might not be consistent with the explicit form of SNA even within the same individual. Indeed, the directionality of the number-space associations in the two tasks did not correlate across participants, further supporting the idea that the two tasks investigate different phenomena.

What is the origin of this biologically grounded, culture-independent mechanism that links numbers to space? Two main hypotheses have been postulated. One is the brain’s asymmetric frequency tuning (BAFT) hypothesis, where the hemispheric specialization for different spatial frequency (SF) bands is used to explain SNA. According to this theory, as SFs are present when fewer bigger elements are presented, an attentional bias towards the left side could occur as the right hemisphere predominantly selects low SFs (and vice versa for larger smaller elements with high SFs selected by the left hemisphere)34. However, in our opinion, this hypothesis cannot fully explain SNA phenomena. For example, in the case of the study reporting SNA in newborns by Di Giorgio and colleagues, while the BAFT hypothesis may explain the infant’s behavior in the test phase, it does not account for the infants’ behavior during the habituation phase. In the habituation phase when presented with two identical stimuli on the left and on the right side of the screen (i.e. sets of 4 dots or sets of 36 dots, the same used in Experiment 2, see Fig. 2A), newborns looked an equal time to both the two stimuli17. According to the BAFT hypothesis, however, they should have looked longer to the stimulus on the right when there were 4 dots and to the left when there were 36 dots given the respective low and high values of SFs in the stimuli.

An alternative hypothesis is that the biological predisposition for a left-to-right oriented SNA could originate in the lateralized organisation of the bilaterian nervous system with the left side of the brain attending to stimuli with positive valence and the right side to stimuli with negative valence35. It has been argued that changes in numerosities towards larger or smaller sets of stimuli are in appetitive contexts associated with differential hemispheric activation and consequently with contralateral hemispace biased attention (to the left for a change from large to small numbers and to the right for a change from small to large numbers; see fig. 7 therein)36. An advantage of this hypothesis is that it could explain SNAs as the results of overall increasing and decreasing change situations, being not limited only to some specific visual features as SFs bands.

We also found an asymmetry in the implicit task, showing a significant Congruency Effect for smaller stimuli presented on the left side, but a reduced one for larger stimuli presented on the right. Similar asymmetries have been reported also in other studies37,38,39. For example, in 2012, Shaki and colleagues found that effects dependent on instructions (such as press for larger or press for smaller) occur in non-numerical magnitude tasks, like judging size and height38. We propose that the non-symbolic nature of our stimuli, as opposed to the symbolic stimuli often used in SNARC studies, may explain this asymmetrical response pattern. The Approximate Number System (ANS, or number sense) is the ability to perceive quantities in a nonverbal and non-symbolic manner40, making it highly relevant to the stimuli used in Experiment 2. Research indicates that the neural basis of the ANS first develops in the right parietal cortex41,42,43,44, extending later to the left parietal cortex45,46,47 a shift likely linked to language acquisition. While the right hemisphere is initially dominant for quantity perception, the left hemisphere later contributes semantic numerical knowledge. In our task, which involves non-symbolic numerosities, the right hemisphere might likely play a more relevant role, enhancing performance in the left visual field (i.e. congruent condition in the decreasing task). However, in the increasing task, we did not observe a significant congruency effect for the incongruent condition (i.e. left visual field). A possible explanation for the observed asymmetry between the increasing and decreasing tasks could be the interaction between increased right hemisphere activation in response to non-symbolic numerosities and the left-to-right SNA. Future studies are needed to further examine the possible role of lateralized hemispheric activation in SNA behaviors.

Although we found evidence for left-to-right SNA in the implicit task in all the populations we tested, further research is needed to overcome some limits of the present study. For example, while we carefully selected Himba adults who reported little to no literacy competence and Italian preschoolers who were reported to be unable to read and write by their parents, we cannot firmely rule out that some cultural context may have influenced their observed behavior48. We also cannot exclude that participants used strategies based on non-numerical features, such as total colored area and convex hull to solve the tasks as these variable were not controlled in the study. However, this potential confound does not affect the conclusions, as the observed differences in behavior across the two tasks occurred despite using similar stimuli. Future studies will be needed to better disentangle the precise role of numerical features versus other non-numerical dimensions in the emergence of this phenomenon. Another limitation of the current study could be found in the lateralized presentation of the test stimuli. Indeed, in 2018, Shaki and Fischer discussed how explicit magnitude processing and explicit spatial-directional processing might bias the SNA effects per se, as “they may have artificially imposed spatial-numerical associations” (see49). According to their description in our numerosity comparison task, both the magnitude processing and the spatial component are explicitly activated, as we ask participants to respond to laterally presented stimuli based on their magnitude value (i.e. press for less or press for more). However, we believe that since the stimuli are presented alternately on the left or right side of the screen (rather than simultaneously on both sides), any potential spatial attentional shift would equally affect both congruent and incongruent conditions, as the magnitude/side pairs are meticulously counterbalanced. Nevertheless, we found a congruent effect coherent only with the left-to-right SNA. Moreover, through the use of a central response key, instead of two lateralized response keys, we further reduced the possible effect of explicit spatial mapping happening at the hand-motor level. It is worth noticing that, given the unique challenges of working with the Himba and preschoolers (i.e. participants who are unfamiliar with computers and tasks of this nature, unlike Western adults), we opted for a paradigm that required instructions as simple as possible (i.e. press one key for less/more). Additionally, classical parity judgment tasks were also unsuitable, as neither Italian preschoolers nor the Himba possess sufficient symbolic numerical knowledge to perform them. Finally, as an additional limit of our study, we note that our implicit task was relatively brief, which may have introduced noise in estimating individual participants’ biases. However, it is also possible that hemispheric biases, even if biologically determined, may exhibit minimal variability across individuals (e.g., the left hemisphere’s dominance in language processing among right-handers, despite individual exceptions50), and this could be a source of variability as well. Now that preliminary evidence supports a left-to-right oriented SNA, the next step is to investigate how consistent this effect is across other tasks. Indeed, in the current study, the explicit and implicit tasks differed in several ways (e.g., the implicit task included feedback, while the explicit task did not; one task also had more trials than the other). Therefore, future research is needed to address whether these differences may have played a role.

In conclusion, this study provides insights into the relationship between SNA, formal education, and cultural influences. The findings highlight the importance of task instructions in revealing different forms of SNA and provide a demonstration of dissociability between covert/implicit and overt/explicit behaviors within the same population lacking formal cultural biases driving a tendency to organize information in space. However, the debate about the precise foundations of these two dissociable types of SNA is still open. We suggest that on one side we are all endowed with a biologically predisposed SNA mechanism, rooted in general brain hemispheric asymmetry35, that is present from birth and remains constant in its left-to-right orientation over time, regardless of cultural influences. On the other hand, in explicit tasks requiring an explicit number to space mapping, culturally influenced mechanisms support the final overt behavior. This work did not aim to investigate the nature of these latter mechanisms and how they are influenced by culture. Now that an initial dissociation between different forms of SNAs has been found, further studies are needed to understand which mechanisms support each form. Indeed, it is not yet clear how the culturally determined form of SNA emerges as a parallel independent process from the biologically predisposed one or whether it is its evolution. One possibility is that the direction of cultural biases in humans is purely conventional, thus not directly issuing from the inborn SNA mechanism. Indeed left-to-right and right-to-left writing/reading systems both exist, even though overall the right-to-left systems are much less frequent than the left-to-right ones (according to the ISO codes list, there have been only 40 languages written from right-to-left throughout all history, among more than 200 see https://unicode.org/iso15924/codelists.html). In summary, the biologically predisposed form of SNA may rely on different mechanisms from the ones that support forms of SNA that humans developed culturally, allowing for the co-existence of different types of SNA, even within the same individuals. Future research will clarify the intricate interplay between cultural practices, brain lateralization, and the development of SNA, in order to shed more light on this intriguing phenomenon.

Methods

The study has been conducted following the Declaration of Helsinki and has been approved by the local University of Trento Research Ethics Committee (for Italian preschoolers and adults’ study), the ethics committee of Inserm (for Himba studies; opinion number 21-855) and the University of Namibia Decentralized Ethics Committee (DEC) (Ethical Clearance Reference Number: SAHS37/23). The Italian adults gave their written consent after being informed about the study’s purpose and procedures. The Himba adults, who were unable to read or write, provided oral consent following the ethical guidelines provided by the ethical committees of Inserm and of University of Namibia, and in agreement with the Declaration of Helsinki. For the Italian preschoolers, we obtained informed consent from their legal guardians. Italian adults received a small monetary reimbursement for their participation, while Himba adults were given a set of four gift items including maize meal, sugar, vaseline, and soap (the approximate value of the items was equivalent to 5 euros).

Participants

  • 191 Himba adults, an indigenous group with an oral culture from northern Namibia, with little or no formal education, were recruited in small villages. All were monolingual native speakers of Otjihimba, a dialect of the Otjiherero language. Data has been collected in two different missions, one in October/November 2021 (90 Himba people tested in 4 villages) and the other one year later in November 2022 (101 Himba people tested in 2 other villages). Here, we will refer to the first group as Himba adults 2021 and to the second one as Himba adults 2022. Among these, we report data for 130 participants (60 from the first mission, 44 females, mean age = 33 ± 16 years, mean grade in school=0.5 ± 1.3; 70 from the second mission, 37 females, mean age = 33 ± 11 years, mean grade= 0.1 ± 0.3 years) who self-reported as being unable to read/write (note that a few participants have attended one or two years of school but were still unable to read/write). The same translator and experimenter conducted the two missions, and the experimental conditions were relatively similar. To our knowledge, none of the participants had prior experience with experimental research. It is important to mention that there is some uncertainty regarding the age of some participants, as it is culturally rare to track count of age. When necessary, we relied on the translator to provide an estimated assessment of the participants’ ages. All experiments were conducted on an outdoor table, in a shaded area.

  • 45 Italian preschoolers, recruited in kindergartens in the area of Rovereto (TN), took part in the study. Among them, 2 were discarded as they did not complete all the tasks. 43 children (21 females; mean age = 5.1 ± 0.3 years) were included in the final sample, and they were all reported by their parents as being unable to read/write.

  • 47 Italian young adults (36 female; mean age = 21.7 ± 2.3 years), recruited through the social media group of the University of Trento, took part in the study.

Stimuli

  • Ordering Card Task: Stimuli were ten white cards (4 ×4 cm). On each card black dots from 1 to 10 were printed. The dots were 0.5 cm in diameter, and they were displayed in a random configuration (Fig. 1A).

  • Numerosity Comparison Task: For this computerized task, the stimuli were visual arrays of 4, 12, or 36 black squares (1.3 × 1.3 degree of visual angle each, distance from the screen was approximately 50 cm) presented on a white background (17 ×17 degree) (stimuli similar to Di Giorgio et al., 2019, Fig. 2A).

Experimental protocols

The first data collection for this study took place with the Himba population in 2021 and has been subsequently replicated with Himba population, Italian Adults, and Italian preschoolers in 2022.

The three populations ran the exact same three experiments, albeit in slightly different settings: Italian adults were tested in a semi-dark room of the University laboratory, Italian preschoolers in a quiet room in their kindergarten, and Himba adults were tested outside, sitting at a shaded table located nearby their local villages. While all participants received oral instructions from the experimenter (helped by a translator in the case of Himba people), Italian adults also read the instructions on the computer screen (for the computerized experiment). All the participants performed first the Ordering Card Task and then the Numerosity Comparison Task. Before the experimental tasks, we interviewed both Himba and Italian adults on their literacy (i.e. their ability to read and write) and their schooling level. The same information about reading/writing ability was collected from Italian preschoolers’ parents through a written questionnaire.

Ordering Card Task

For the Ordering Card Task, participants were asked to manually order 10 cards depicting 1 to 10 dots on the table in front of them, with the only constraint that they would be considered to be in order. No other instruction was given. Participants were presented with the 10 cards piled up in a random order and headed down to the table in front of them.

Numerosity Comparison Task

Stimuli were presented on a laptop computer screen and participants sat at approximately 50 cm from it. Trials started with a fixation cross at the center of the screen that lasted for 1 s, followed by a set of dots presented centrally for 500 ms (hereafter reference stimulus). Then, a black screen appeared for 200 ms and was followed by a second set of dots (hereafter test stimulus), which was presented on the left or on the right side of the screen. The test stimulus remained on the screen until the subject pressed the response-key or for a maximum of 3 s. Participants were instructed to press a central key as fast as possible (with their dominant hand) only when the test stimulus was less numerous compared to the previous stimulus (hereafter Decreasing Task instruction), or only when the test stimulus was more numerous (hereafter Increasing Task instruction). Feedback was always provided (a green happy smiley for correct responses and a red sad smiley for incorrect responses were presented for 1 s after the participants’ response).

The experiments started with 18 trials to familiarize the participants with the task, followed by 48 experimental trials. Of those, 30 were targets (i.e. trials that required a response; 15 presented on the left and 15 on the right, 10 trials for each numerosity comparison) and 18 were distractors (i.e. trials that did not require a response: 9 presented on the left and 9 on the right).

We analyzed the responses to the target trials only. All the experimental groups did the task in both conditions (with the order counterbalanced across participants), with the exception of Himba 2021 who performed the task in a between-subjects design. Both stimuli presentation and data collection were performed with Psychopy software51.

Statistical analysis

Data analysis and plot generation have been performed with R software52, R version 4.1.3 (2022-03-10) in RStudio environment53. For all statistical analyses alpha = .05 significance level was chosen. The effect size was measured according to Cohen’s formula d = \(\frac{{{{{\rm{M}}}}}_{1}-\,{{{{\rm{M}}}}}_{2}\,}{{{{{\rm{SD}}}}}_{{{{\rm{pooled}}}}}}\) for parametric t tests and with the recommended d = \(\frac{{{{{\rm{Z}}}}}_{{{{\rm{statistic}}}}}}{\sqrt{{{{\rm{N}}}}}}\) formula for non-parametric tests; partial eta squared is reported for ANOVA effect size54. Whenever a non-significant effect was determined (i.e. pvalue > .05), Bayesian analyses were performed to allow meaningful interpretation of the null results using the JASP software (version 0.18.3.0)55. Interpretation of the Log10(Bayesian Factor) is provided according to the scale proposed by Jeffreys in 196156.

Ordering Card Task

For the Free Ordering task, we started by performing a χ2 analysis, on the frequency with which each type of spatial configuration (lateral line, sagittal line, two-dimensional, and random shape) was chosen by the participants in each group (Fig. 1B, Table S1). Then, for those participants who used a lateral linear configuration, we performed the quantitative analysis, which, following Pitt et al.15, proceeded in two steps.

First, we tested the systematicity of the lateral line mapping. To do so, for each participant we correlated the chosen order to that of the ideal left-to-right order using two-tailed Kendall’s Tau Rank Correlation Coefficient, yielding a score between -1 and 1. The absolute value of the score indexes the systematicity of the mapping (i.e. how orderly: a score of ±1 corresponds to a mapping where numerosity increases monotonically in one direction across all ten positions). Intermediate scores reflect imperfectly ordered mappings (Fig. S1). To determine whether participants performed the task by ordering the cards with a certain systematicity, we compared the distribution of mapping scores they produced to the distribution of mapping scores that would be expected by chance (i.e. as a result of random arrangements of the 10 stimuli on the line). The chance distribution was generated by performing all the possible permutations of the 10 elements sequences without repetitions (10! = 3,628,800 permutations) and it was compared to the real data distributions using the two-tailed two sample Kolmogorov-Smirnov goodness-of-fit test.

Second, we tested the direction of the mapping. The sign of the Kendall’s Tau correlation coefficient indexes the direction of the mapping: positive scores indicate an overall rightward mapping, while negative scores indicate a leftward one. To determine whether a preferential direction of mapping emerged at the population level, we tested whether the group average score differed significantly from 0, using the non-parametric two-tailed Wilcoxon signed-rank test to account for the non-normal distribution of the scores.

Numerosity Comparison Task

We started the analysis by excluding participants who responded at or below chance (62.5% accuracy), resulting in the removal of 5 Himba 2021, 6 Himba 2022, and 5 pre-schoolers.

In order to condense speed and accuracy in a single measure, for each subject and condition we computed the Inverse Efficiency Score (IES), defined as \(\frac{{{{\rm{R}}}}{{{{\rm{T}}}}}_{{{{\rm{mean}}}}}}{{{{\rm{Accuracy}}}}}\)57 (i.e. lower scores mean a better performance). The one-tailed Shapiro-Wilk test was performed to test for the normal distribution of data, and data resulted not normally distributed. However, due to the necessity of testing for both repeated measure factors and interaction effects, we decided to perform an ANOVA test as it has been demonstrated to be robust against Type 1 Error also in case of violation of normality assumption58. We first analysed the Himba Adults 2021 data with a two-way mixed ANOVA setting Congruency condition as a within-subjects variable and Task instruction as a between-subjects variable (Table S1). As we found a significant interaction between Congruency condition and Task instruction, in the subsequent data collection and data analysis Task instruction has been considered as a relevant independent variable to insert into the model. We performed a three-way mixed ANOVA, with Group (Himba Adults 2022 vs Italian Adults vs Italian Preschooler) as a between-subjects variable, Congruency condition (Congruent vs Incongruent), and Task instruction (Decreasing vs Increasing) as within-subjects variable (Table S2). Planned t tests were performed to compare differences between congruent and incongruent conditions for each Task for each Group. Both two-tailed two-sample paired Student’s t tests (Table S4) and one-tailed with less alternative two-sample paired Student’s t tests were performed, as for priori defined hypothesis of congruent condition eliciting better performance (i.e. lower scores) than incongruent condition (Table S5).

To account for the average difference in the inverse efficiency score across groups (see Table S1), mostly due to children being slower and Italian adults being more accurate compared with the other groups, we also computed a congruency effect for each subject and Task instruction by normalizing the difference in the IES for incongruent and congruent conditions by their sum: \((\frac{{{{\rm{IE}}}}{{{{\rm{S}}}}}_{{{{\rm{incongruent}}}}}-{{{\rm{IE}}}}{{{{\rm{S}}}}}_{{{{\rm{congruent}}}}}}{{{{\rm{IE}}}}{{{{\rm{S}}}}}_{{{{\rm{incongruent}}}}}+{{{\rm{IE}}}}{{{{\rm{S}}}}}_{{{{\rm{congruent}}}}}}).\) One-tailed Shapiro-Wilk test was performed to test for normal distribution of data (Tables S6, Fig. S2) and a 3x2 ANOVA was implemented to test for differences across Group and Task instruction (Himba adults tested in 2021 were excluded from the model as they did not perform the task in a within-subjects design fashion, Table 2). Then to investigate whether this congruency effect was significant we tested it using two-tailed one-sample Student’s t tests against zero (alternative hypothesis µ ≠ 0) for normally distributed data and a two-tailed Wilcoxon signed-rank tests against 0 (alternative hypothesis µ ≠ 0) for not normally distributed data (Fig. 2C, Tables 3, 4),

Mixed Model analyses were carried out to analyse Reaction Times, Errors, and Inverse Efficiency Scores using the R -package “lme4”59. Given the data distributions, we modeled Reaction Times (RT) and Inverse Efficiency Scores (IES) with a Gamma distribution and Errors with a binomial distribution. Here we reported the description of the models used for the three variables (see Supplementary Materials for the detailed model outputs and models comparison).

  • Inverse Efficiency Score model

    For IES we tested the fixed effect of Congruency condition, Task instruction, Group, and their full interaction. We also added the random effect of participants and the random slopes for the variable of interest (i.e. Congruency condition). It was not possible to add the random slope for type of testing as the IES is an aggregate value that collapses over different reference-test stimulus pairings. We also used the “bobyqa” optimizer from the “glmerControl” R package.

    Model formula: IES~congruency * task * group +(1+congruency |part_id)

  • Errors model

    Italian Adults were not included in the model as they made no mistakes. For Errors, we tested the fixed effect of Congruency condition, Task instruction, Group, and their full interaction. We also added the random effect of participants and type of test (i.e. the reference-test stimulus pairing). It was not possible to add random slopes for the variable of interest (i.e. Congruency condition) as it consistently led the model to fail to converge. We also used the “bobyqa” optimizer from the “glmerControl” R package.

    Model formula: Errors ~ congruency * task * group + (1 | part_id) + (1 | type_test)

  • Reaction Times model

    For RT we tested the fixed effect of Congruency condition, Task instruction, Group, and their full interaction. We also added the random subject slopes for the variable of interest (i.e. Congruency condition) and the random effect of type of test (i.e. the reference-test stimulus pairing). We also used the “bobyqa” optimizer from the “glmerControl” R package.

    Model formula: RT~ congruency * task * group + (1 + congruency | part_id) + (1 | type_test)

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.