Development of Understanding Zero

While the knowledge about the development of understanding positive integers is rapidly growing, the development of understanding zero is not well-known. Here we tested several components of preschoolers’ understanding zero: whether they can use empty sets in numerical tasks, whether they can use empty sets as soon as they understand the cardinality principle, whether they know what the word “zero” refers to, and whether they categorize zero as a number. The results show that preschoolers can handle empty sets in numerical tasks as soon as they understand the cardinality principle or even earlier, and some of them know that these sets are labeled as “zero.” However, they are unsure whether zero is a number. We argue that preschoolers might understand numbers as the properties of items or objects in a set. In this view, zero cannot be a number, because an empty set does not include any items, and the missing items cannot have any property, excluding also the number property. This model might explain why zero is handled correctly in numerical tasks, while it is not regarded to be a number.


Introduction
Children start to understand the use of symbolic exact numbers at around the age of 3 years (Wynn, 1990(Wynn, , 1992)).Although many details about the development of understanding natural numbers are already known, the development of understanding zero is hardly known, and it is not integrated into the numerical cognition models.The main aim of the present study is to extend the description of the development of understanding zero and to consider its theoretical consequences. 2

Lack of developmental models of understanding zero
Models about the development of numerical cognition mostly cannot specify how understanding zero is integrated into the more general numerical knowledge.In the infant literature there is an agreement that numerical information is probably processed by two representations.First, the approximate number system (ANS) can represent numbers in an imprecise way: infants are able to discriminate two values if the difference is larger than a specific ratio (Feigenson, Dehaene, & Spelke, 2004), which ratio (the Weber-fraction) improves with age (Piazza, 2010).Second, there is an object file system or visual indexes that can represent objects up to a limit of 3 or 4 items, which system can implicitly store the number of those objects (Feigenson et al., 2004).It is not clear, however, whether these systems can represent zero.
Regarding the ANS, Merritt and Brannon (2013) investigated whether four-year-old preschoolers can handle zero in a nonsymbolic comparison task.They were looking for the distance effect (increasing error rate with smaller distance), which effect is considered to be the sign of the ratio-based ANS processing (Moyer & Landauer, 1967).While in the whole sample the distance effect was observed in the comparison task, suggesting the role of the ANS in handling zero, other aspects of the results suggest that it was not the ANS that processed zero.First, many children could not solve the task: children, who performed relatively poorly in the positive number comparison task, performed randomly and did not show any distance effect with number pairs including zero.However, poor performance for positive numbers is surprising, because even much younger infants are sensitive to numerical information (Feigenson et al., 2004).Second, the performance with pairs including zero was worse than with pairs including one, however, the ANS would predict a better performance with the zeropairs.The authors hypothesize that preschoolers initially might avoid zero, because for example, zero is not included in the usual counting sequence (Merritt & Brannon, 2013).Because it is supposed that the same ANS works in non-humans and humans across various ages with basically the same properties (Piazza, 2010), other studies might help us to specify whether ANS could store zero.Unfortunately, it is not easy to tell whether ANS can represent zero, even when investigated in adults or non-humans (Dehaene, Bossini, & Giraux, 1993;Fischer & Rottmann, 2005;Merritt, Rugani, & Brannon, 2009;Varma & Schwartz, 2011).One problem is that presenting a set with zero involves methodological issues, e.g., nothing is shown, with radically different perceptual properties compared to non-zero stimuli.A related problem is that presenting nonsymbolic empty sets opens up different interpretational possibilities about what information could be used to differentiate empty set from non-empty sets.Overall, the results about whether ANS may represent zero are inconclusive.While some authors argue that ANS can store zero (Merritt & Brannon, 2013;Wynn & Chiang, 1998), these results can be debated, and according to some models it would be impossible to store zero by the ANS, because for example, the ANS might rely on perceptual properties, and missing stimuli (i.e., a non-symbolic empty set) cannot activate the appropriate features in the system (Dakin, Tibber, Greenwood, Kingdom, & Morgan, 2011).It is also possible that even if zero is stored by the ANS, it requires additional supporting processes.For example, it was found that monkeys could solve number matching task with empty set, and cells tuned to the empty sets were found both in the ventral intraparietal area and in the prefrontal cortex, as it has been observed for positive numbers (Ramirez-Cardenas, Moskaleva, & Nieder, 2016).However, 3 only the cells in the prefrontal cortex showed the distance effect for zero, a sign of ANS processing, and not the intraparietal cells, although the ANS is supposed to be localized in the intraparietal area (Dehaene, Piazza, Pinel, & Cohen, 2003).Thus, it is not clear whether ANS could handle zero either in general, or specifically in infants or in preschoolers.Some studies investigated more directly whether infants can represent the lack of objects.Sixand 8-month-old infants seemingly cannot maintain the representation of a lack of an object measured with violation-of-expectation paradigms: while they see the unexpected or "magical" disappearance of an object as an unexpected event, they do not see the "magical" appearance of an object as an unexpected event (Kaufman, Csibra, & Johnson, 2003;Wynn & Chiang, 1998).On one hand, we can suppose that representing the lack of object is handled by the representation handling numerical information.In this case, because the numerical handling of small (i.e., less than 4 or 5 items) sets of objects is attributed to the visual indexes or to the object files by default, and the ANS handles these small sets only in exceptional cases (Feigenson et al., 2004), the failure to detect a magical appearance supports the idea that it is the visual index or the object file system that cannot represent the lack of objects.On the other hand, it is also possible that representing lack of objects is handled by some other systems, representing other aspects of objects or locations, and these results cannot specify whether the ANS or the visual index can handle empty sets.To summarize, there is no consensus whether either the ANS or the visual index can represent zero, and there are hardly any data describing whether these or other systems can handle zero in infants.
The next important step in the development of number understanding is the understanding of the exact large symbolic numbers in preschoolers.At around the age of 3 or 4 years, children start to understand the conceptual principles of number use, termed understanding the cardinality principle (Lipton & Spelke, 2006;Sarnecka & Carey, 2008;Wynn, 1990), with which they are able to handle exact symbolic numbers.Number knowledge in preschoolers is usually measured with the "give a number" task in which children are asked to give a specific number of objects from a pile of objects (Wynn, 1990(Wynn, , 1992)).With this task one can determine what number understanding phase a child is in.First, preschoolers are in the pre-numeric phase, where although they know the counting list (the series of words "one-two-three", etc.), they do not know the meaning of these words and they fail in the task.Second, children become subsetknowers, when they can give 1, 2, 3, or 4 items, but not when asked for more, even when they know how the counting list continues.Finally, preschoolers become cardinality-principleknowers, when they can give any amount of items that is in their known counting list.This last phase is believed to show the real understanding of exact symbolic numbers (Lipton & Spelke, 2006;Sarnecka & Carey, 2008;Wynn, 1990Wynn, , 1992)).
There is no consensus what representational changes occur when a preschooler understands the cardinality principle.Most models suppose that in some way the systems available in infancy might play a role in the first steps (Carey, 2004(Carey, , 2009;;Piazza, 2010), although it is not known entirely how these or other systems contribute to the understanding of symbolic exact numbers.Partly related to this problem, it is not known whether preschoolers understand symbolic zero when they understand the cardinality principle.Importantly, the few former works investigating preschoolers' symbolic zero understanding (see the next subsection) cannot be integrated into this framework, because those works have not investigated whether those preschoolers understood the cardinality principle or not.

Contradicting results about understanding zero in preschoolers
To our knowledge, there are only two studies investigating quantitatively the symbolic zero concept in preschoolers (Bialystok & Codd, 2000;Wellman & Miller, 1986).1 (While the Merritt & Brannon, 2013 paper collects preschooler data about processing zero, the comparison task is not symbolic.Also, while Davidson, 1992 uses numerical tasks with zero in preschoolers, those tasks alternate training tasks and questions, so it is not clear what is the effect of the training in the session.)As we mentioned, none of these two studies is related to the current developmental models.Importantly, the conclusions of these two works contradict each other, whether handling zero is more difficult or not for preschoolers compared to handling positive integers.In this subsection, first, we briefly summarize the two studies, then we discuss the potential causes of their differing conclusions.Wellman and Miller (1986) presented children between 3 and 7 years of age with the following tasks: (a) count items in sets (including empty sets), (b) name the smallest number they knew, (c) name Indo-Arabic symbols, and (d) compare numbers between 0 and 5 in Indo-Arabic notation.Their main finding was that the understanding and the use of zero were delayed compared to the use of positive numbers.In a detailed analysis the authors concluded that there are three typical behavioral patterns, or in their term, phases.In the first phase, the children could name the 0 symbol, although they did not understand the meaning of it.In the second phase, the children could count backward to zero, understanding that zero means nothing.Finally, in the third phase, the children knew that zero was the smallest number, and they could compare numbers even if one of the numbers was zero.The progress of the development was slow: 4-year-old children did not reach the first phase, and only 6-year-old children were able to reach the final phase.
However, a later study found evidence that understanding zero is easy for preschoolers.Bialystock and Codd (2000) investigated the understanding and spontaneous notation of positive integers, zero, and fractions in preschool children.They found that understanding zero is not harder than understanding positive integers, a conclusion that is not in line with the previously described study by Wellman and Miller (1986).In this work, children between 3 and 7 years of age were asked to give different amounts of cookies to puppets, and to make notes about these amounts.The children had to recall that amount 20 minutes later, and again 2 weeks later.In both instances children were allowed to use the notes they had made.According to the results, children were able to solve the zero giving task.However, it is important to note that the instruction was not formed in the usual mathematical way, e.g., children were not told that "Give Big Bird zero cookies", instead they were instructed to "Give Big Bird no cookies for lunch."Children could make a note of the number zero as efficiently as making a note of the positive integers.Similarly, they could recall the correct number after 20 minutes be it either zero or positive integers.However, two weeks later unlike 5-year-old children, 3-and 4-yearold children were not able to recall the number zero as successfully as recalling positive integers.(The same result was found by Hughes, 1986, revealing that children can use notes for denoting zero, although quantitative results were not published for that study.) The contradiction between the conclusions of the two studies might originate from methodological differences and from interpretational issues.Obviously, the two studies used completely different tasks, and it is possible that zero can be handled more easily in some tasks than in others.Yet there are some less trivial sources of differences.First, while Wellman and Miller (1986) suggest three phases of the development of understanding zero, actually, their data seem to reveal four phases.(Although the authors mentioned that the data might include some inconsistency, they insisted on the interpretation with three phases.)There is an extra phase between the second and the third phases: after successfully counting back to zero, preschoolers could compare numbers with zero, whereas they did not know that zero was the smallest number (see Table 1 in Wellman & Miller, 1986).This phase is actually paradoxical: while children know that zero is smaller than one, they think that one is the smallest number.
What causes this dissociation in their knowledge about zero?One possibility is that children do not think that zero is a number.This possible misconception is observable even in adults: in a study 15% of preservice elementary school teachers responded that zero is not a number (Wheeler & Feghali, 1983).Another possible explanation is that zero is not part of the counting list (which usually starts with "one"), and that is why children handle it differently (Merritt & Brannon, 2013).Both explanations would suggest that the meta-knowledge about the number status of zero could be independent of handling correctly the zero value.Consequently, one might think that children can understand zero sufficiently when they can compare zero correctly, but they do not yet know that zero should be categorized as a number.If this is the case, children could understand zero earlier than it was originally proposed by Wellman and Miller.We get back to this problem and to a more detailed list of possible explanations in the General discussion.
A second methodological problem that could cause different conclusions in the two studies is that the linguistic formulation of the tasks regarding zero might influence the performance.While the mathematical viewpoint suggests that zero is a number just like any other integers, and therefore zero should be used linguistically in the way other numbers are used, natural language mostly uses different linguistic forms to offer a statement about zero.For example, we usually do not say that "the car travels with zero kilometers per hour", instead, we say that "the car has stopped."Similarly, we do not say that "give zero cookies to Peppa Pig", but we say that "do not give any cookies to Peppa Pig" or "give no cookies to Peppa Pig."Thus, it is possible that using the mathematical language is harder for children than using the natural language, because the mathematical version is used less frequently.This could be hypothetically confirmed by the previous data: the Wellman and Miller (1986) study used the mathematical language and found difficulties in understanding zero, while the Bialystock and Codd (2000) study used the natural language and found no difficulties with handling zero.However, based only on these data one cannot be sure whether the language form really influenced the performance, because of the many other differences between the two studies.

Aims of the study
To create appropriate models, it is essential to have reliable data first.Because former studies (Bialystok & Codd, 2000;Wellman & Miller, 1986) drew contradicting conclusions about the 6 processing of the zero value in preschoolers, and because based only on former data one cannot specify the sources of those contradictions, the main aim of this work is to provide additional, more systematically collected data about the symbolic understanding of zero in preschoolers.Importantly, our aims do not build upon the theoretical models, such as the ANS, the visual index, and so on (presented in the first part of the Introduction), but they intend to clarify the main phenomena and describe the development more precisely and extensively than previous works did (presented in the second part of the Introduction).More specifically, first, a more comprehensive range of tasks (comparison, addition and subtraction, with verbal and nonverbal versions) is used to investigate whether children can handle zero in numerical tasks as efficiently as they can handle positive integers (Aim 1).Second, partly as a component of the first aim, we want to test the potential effect of language use (contrasting mathematical vs. natural language) on the performance in those tasks (Aim 2).Additionally, to put these findings into the context of current models about understanding the cardinality principle, the present work investigates whether subset-knowers and cardinality-principle-knowers have a different level of understanding of zero if understanding zero is available at such an early age (Aim 3).Finally, we want to investigate whether the extra phase we emphasized in the data on Wellman and Miller (1986) is reliably observable, i.e., whether at some point children can compare zero appropriately, while they still think that 0 is not a number (Aim 4).

Participants
Forty 3-and 4-year-old Hungarian preschoolers participated in the study, 20 boys and 20 girls, with mean age 4 years 0 months, range between 3 years 2 months and 5 years 1 month.Two preschools were involved in the data collection, one of them in the capital and the other one in a country town.Both preschools mostly receive children from middle-class families.None of the preschoolers received formal training about handling zero in the preschools formerly.

Tasks
The tasks covered three main areas (Table 1).( 1) Give-N task categorized children whether they are cardinality-principle-knowers or subset-knowers, and additionally whether they understood that zero is associated with empty sets or not.(2) Comparison, addition and subtraction tasks measured whether preschoolers could use zero in operations as efficiently as they use positive integers.(3) Meta-knowledge about numbers was measured to investigate whether children understood that zero was a number.See the order of the tasks at the end of the Methods section.Data collection for a single child took approximately 30 minutes.

Measuring number knowledge
Give-N Give N balls to an agent.

Comparison
Choose the larger set.
Addition Add two values.

Subtraction
Subtract one value from another.
Meta-knowledge about zero Smallest number Name the smallest number.
Is it a number?Specify whether different things are numbers or not.Table 1.Summary of the tasks.
Give-N task.In the give-N task the children could see a pile of balls, and they had to give a specific number of balls to a toy bird.The task measures (a) whether the child understands the principle of cardinality, and (b) whether the child understands what the words "nothing" and/or "zero" refer to.Importantly, zero-knowledge is measured with the same task and with the same criteria as positive integer knowledge, thus, this method could serve as an appropriate starting point to categorize children whether they know what "zero" refers to.The task is similar to the give-a-number task described by Wynn (1990) and by Bialystock and Codd (2000), the consensually accepted tool to measure the preschoolers' cardinality-principle and number knowledge.At the end of the trial the experimenter explicitly asked the child whether she was ready with the task.This is essential in the zero-trials when the children do not have to give any items.In the task two versions of zero were applied.In the "natural" version of zero we utilized the form that is used in everyday language: "do not give any balls to the bird."In the "mathematical" version we used the form that reflects the mathematical viewpoint: "give zero balls to the bird."In Hungarian, the nouns after number words always use the singular form, e.g., "give zero ball" or "give two ball", thus, the plural or singular form of the noun could not help the children to solve the task.(See the summary of the questions in all tasks in Hungarian and their English translations in the Appendix.)The following numbers were tested in the given order: 2, "natural" 0, 5, 3, "mathematical" 0, 4. The current order of the numbers was used to avoid that children could rely on the order of the increasing number words in the task.The whole number series was repeated twice.It has been argued that children solving the tasks with numbers 4 or larger than 4 show an understanding of the cardinality principle (Condry & Spelke, 2008;Sarnecka & Carey, 2008;Wynn, 1990Wynn, , 1992)).A child was categorized as knowing a number if both trials of that number were solved correctly, and if known numbers were not given as response for higher unknown numbers.2Following these results the children were categorized as cardinality-principle-knowers if they could give 4 and 5 in both trials, otherwise they were categorized as subset-knowers.Independently of the previous categorization, the children were categorized as "zero"-knowers if they did not give anything in the "mathematical" 0 ("zero") tasks in both trials.

8
Comparison.The aim of the task was to test whether the children knew the position of zero among other numbers, and to investigate whether the form of the task (verbal vs. non-verbal) influences their performance.In the comparison task the children either could see two sets of objects or could hear two numbers, and had to choose the larger one.To test whether the children have problems with the verbal description of the task we used both a verbal and a nonverbal object version.In the object version two sets of balls were placed on the two opposite sides of the table, and the question was "On which side can you see more?"The number of the objects in a set was not named by the experimenter.In the case of the zero value the appropriate side of the table remained empty.In the verbal task no objects were used, and the question was "Which one is more, the x or the y?" (The Hungarian translation of "which one" does not include the word "one" in the question, thus, this part of the question could not confuse the children.)In the verbal condition the number 0 could be labeled either as "zero" (mathematical version) or "nothing" (natural version).If children understood the principle of cardinality measured by the Give-N task, the 3-2, 4-1, 2-4, 3-zero, 2-nothing, 1-5, zero-4, 1-3, 2-zero, nothing-4, 2-1, zero-1 number pairs (3 pairs with "zero", 3 pairs with "nothing", and 6 pairs with positive values only) were used in this order.Otherwise the 3-2, 2-1, 1-2, 1-zero, 2-nothing, 1-3, zero-2, 1-2, 2-zero, nothing-1 number pairs (3 pairs with "zero", 2 pairs with "nothing", and 5 pairs with positive values only) were applied in this order.The use of the two series for the two groups ensures that (a) the children see tasks only with a number range corresponding to their capability, while (b) it is possible to measure a relatively wide number range in the cardinality-principle-knowers.The two series do not prevent us from investigating the main question, whether zero is processed as correctly as positive numbers, because no direct comparison of the two groups in the positive number performance is required.Because in the object condition the "zero" and the "nothing" versions mean the same stimulus (missing objects), only one version was used.
Addition and subtraction.In the arithmetic tasks children had to add or subtract two numbers, and tell the results.To test whether the children had problems with the verbal description of the task, the tasks could be either in verbal form or shown with objects.In the verbal form either the "natural" or the "mathematical" form of zero could be used.In the object version, while the task was also explained in verbal form (in case of zero both in the "natural" and in the "mathematical" forms), the operands of the task were shown with arrays of balls, one operand on one side and the other operand on the other side of the table.In case of zero value, the appropriate side of the table remained empty.If the child understood the principle of cardinality measured by Give-N task, the 1+1, 3+1, 2+0, 1+2, 0+3, 2+nothing, nothing+3; 2-1, 4-2, 3-0, 3-1, 2-0, 2-2, 3-nothing, 2-nothing tasks (4 tasks with "nothing", 4 tasks with 0, and 7 tasks with positive values only) were used in this order, otherwise the 1+1, 2+1, 2+0, 1+2, 0+1, 2+nothing, nothing+1; 2-1, 2-0, 1-0, 1-1, 2-nothing, 1-nothing tasks (4 tasks with "nothing", 4 tasks with 0, and 5 tasks with positive values only) were applied in this order.The motivation for the use of the two series was the same as in the comparison task.The task was embedded in a small story.In the addition task the following story was used: "The bird had x balls.The dog had y balls.How many balls do they have altogether when they play together?"In the subtraction task the following story was applied: "The bird had x balls.Then, the bird gave y balls to the dog as a present.How many balls was the bird left with?"In the "natural" linguistic version we applied forms that were mostly used in everyday speech, e.g., "The dog didn't have any balls."(In Hungarian it is not possible to use a version that is close to the English "Give no balls" version, because in Hungarian the predicate is negated, and the results are similar to a version as "Do not give none balls to the dog.")In the "mathematical" linguistic version we used a form that reflected the mathematical viewpoint, e.g., "The dog had zero balls." 3 Smallest number.The children were asked what the smallest number was.The aim of the task was to find out whether children regard zero as the smallest number.The task was similar to the task utilized in the study of Wellman and Miller (1986).The task was applied as the first task of the session, so as not to influence the reply with the result of some of the tasks that might teach about the number zero.(See further details in the Order of the tasks part below.)However, because several other tasks could have had an effect on this knowledge, we repeated this task at the end of the session to see whether the knowledge about zero could be changed by the knowledge the children acquire in the testing process.
Is it a number?The children had to categorize whether numbers and other things were numbers or not.The children were verbally asked "is the … a number?"The aim of this task was to study explicitly whether children regarded zero as a number.To explore whether children understood this categorization task, additional numerical and non-numerical words were used to validate the task.The following words were used in the following order: three, two, nothing, kitten, pop (sound), zero.Similar to the Smallest number task, this task was presented at the beginning of the session, and it was also repeated at the end of the session.
Order of the tasks.Because some tasks could have provided some information about the meaning and use of the number zero to the children, the order of the tasks was mainly influenced by the potential of the task to modify the replies of the children in later tasks.The tasks about the status of the zero should be the first tasks.Additionally, tasks that include verbal and object version at the same time could teach the meaning of zero, and should be placed at the end of the session.Thus, the following order of the tasks was applied: smallest number, is it a number, give-N task, comparison, addition and subtraction verbal version, addition and subtraction object version, smallest number repeated, is it a number repeated.

Groups based on number knowledge
First, with the Give-N task we specified the number knowledge level of the children (subsetknowers vs. cardinality-principle-knowers).This categorization was used in the succeeding tasks to contrast children based on their number knowledge.Based on the trials with numbers between 2 and 5, 20 children understood the cardinality principle (i.e., they could solve the give-N tasks perfectly), and 20 children did not reach this phase, being subset-knowers.Two of these children could not even solve the give 2 balls task, and because number 1 was not 3 One might ask whether the non-numerical versions of these tasks really measure numerical abilities or whether they measure some other abilities.For example, it is possible that children use a non-numerical concept of nothing to solve the tasks.However, it is important to see that for a preschooler both the concept of nothing and the numerical concept of zero could be appropriate to solve numerical tasks with zero.At their age preschoolers can solve only a few numerical tasks: comparison, addition, subtraction (Levine, Jordan, & Huttenlocher, 1992), and in all of those tasks both the concept of nothing and the concept of zero would give the same correct result.Therefore, a correct or erroneous numerical task cannot differentiate in a simple way whether the concept of nothing or concept of zero was used, and consequently, the question whether any versions promote different strategies is testable with these methods.measured in the task, they could be either "one"-knowers or pre-number-knowers.Because we could not specify if these two children were subset-knowers or pre-number-knowers, we excluded them from further analysis.Among the remaining 18 subset-knowers there were 7 "two"-knowers, 10 "three"-knowers and 1 "four"-knower.
Second, we specified whether children understood zero ("zero"-knowers vs. "zero"-notknowers).Again, this categorization was used in the analysis of the following tasks.Based on the trials with zero we found that while practically all children understood the natural version of give zero task ("do not give any balls"), the mathematical version ("give zero balls") proved to be more difficult (Figure 1).A child was categorized as "zero"-knower, if the mathematical version ("zero") of the task was solved correctly in both trials.We could not rely on the natural version ("nothing") of the task, because this task seemed to be trivial for the preschoolers, resulting in a ceiling effect (Figure 1).None of the "zero"-knowers solved the task by adding zero accidentally for unknown numbers, because zero was never given when a positive number was asked by the experimenter.The difficulty with the mathematical version could be rooted either in not knowing the word "zero", or in the unusual and unnatural form of the task (i.e., give something which is nothing).However, some occasional note by the children showed that at least some of them did not know what the word "zero" refers to.Some examples of those notes are: "What does zero mean?", "I cannot count up to zero", "That would be too much for me", "Zero is hundred".Additionally, these data reveal that also subset-knowers might be able to handle zero correctly in the give-N task, and some of them even understood the word "zero."All of the subset-knowers solving the mathematical version of the give zero task were "three"knowers.A 2 (subset-knower group vs. cardinality-principle-knower group as a betweensubjects factor) × 2 ("natural" vs. "mathematical" version of zero task as a within-subjects factor) ANOVA on the proportion of correct responses confirmed statistically the previously described effects: a main effect of number knowledge group, F(1,36)=7.59,p=0.009, ηp 2 =0.174, and main effect of linguistic version, F(1,36)=46.62, p<0.001, ηp 2 =0.564 were found to be significant, and a tendency in the interaction, F(1,36)=3.66,p=0.064, ηp 2 =0.092 was found.
Figure 1.Proportion of correct responses for the natural and mathematical version of "give 0 objects" in subset-knowers and children understanding cardinality.Understanding the principle of cardinality and understanding the give-zero task are computationally independent of each other, i.e., in principle a child could know what "zero" refers to with or without understanding the cardinality principle, and a cardinality-principleknower might or might not know what "zero" refers to.These two orthogonal dimensions created four groups of children that was used in the subsequent analyses (Figure 2, this is practically the same information as in Figure 1, but shows the four groups more directly): 4 subset-knower and "zero"-knower (4.08 years mean age 0.26 SD), 14 subset-knower and "zero"-not-knower (3.8 years mean age, 0.4 SD), 12 cardinality-principle-knower and "zero"knower (4.33 years mean age, 0.55 SD), and 8 cardinality-principle-knower and "zero"-notknower (4.06 year mean age, 0.37 SD).A child is a "zero"-knower if she can solve correctly the mathematical versions of the give-zero tasks.

Operations with zero: comparison, addition and subtraction
These tasks tested (1) whether children could use empty sets in simple numerical operations, and more specifically, whether the value of zero is more difficult to handle than handling positive integers.(2) To check the effect of language (a) we contrasted verbal and nonverbal versions, which could reveal whether the children have either linguistic or conceptual problems with the zero value, and (b) we contrasted the natural linguistic version (e.g., "add nothing to three"), and the mathematical linguistic version (e.g., "add zero to three").( 3) In all of these tasks we also investigate the differences between cardinality-principle-knowers and subsetknowers.
Because of a data collection problem 2 out of 4 subset-knower and "zero"-knower children's data were not available.Because the remaining data of the 2 children might be misleading due to extremely small sample size, this whole group was excluded from the analysis.
First, in the object versions of the tasks, overall, handling zero was not more difficult, than handling positive values (left in Figure 3, and left in Table 2).(In the figures and in the appropriate ANOVAs the critical information is the main effect of number type (positive numbers vs. zero vs. nothing) reflecting an overall effect of zero, or the interaction between number type and number knowledge groups reflecting a group specific effect of zero.However, the main effect of groups is not relevant, because (a) differences between the groups only show whether some groups perform better in general, and not the relative difference between zero and positive integer related performance, and (b) the cardinality-principle-knower and subsetknower groups received different tasks (see methods).)This shows that understanding zero is available as soon as one understands positive integers.This is true for both cardinalityprinciple-knowers and subset-knowers.Additionally, subtracting "nothing" was easier than subtracting positive numbers, which might be reasonable, since in a relatively difficult subtraction task it is easier to do nothing than to subtract a positive value.Thus, these results show that handling zero is not harder than handling other numbers for preschoolers in the object version of the tasks.This means that they could understand the value of zero conceptually.
Second, in the purely verbal versions of the tasks, handling the zero value with the natural linguistic version ("nothing") was not more difficult than handling the positive values (right in Figure 3, and right in Table 2).Again, this shows that understanding zero is available as soon as one understands positive integers.Also, subtraction with "nothing" is easier than subtraction with positive numbers.These results repeat the result of the object version.However, handling zero values with the mathematical linguistic version ("zero") revealed difficulties.Importantly, this difficulty was seen mainly in the "zero"-not-knower groups, but not in the "zero"-knower groups.This pattern means that in a trivial way, children who do not know what the word "zero" refers to cannot solve the tasks including that unknown word.This interpretation is confirmed by the fact that these tasks can be solved when the task is also supported by object demonstrations.The results described here are confirmed by the appropriate ANOVAs (see Table 2), except (a) the interaction between groups vs. task in the subtraction task, where although the descriptive statistics show the usual pattern observed in other tasks, the hypothesis test did not reach significance, and (b) in the addition task the "zero" task improvement in the "zero"-knower group did not cause significant interaction.To summarize, the verbal version of the operations tasks also confirms that preschoolers can handle zero as efficiently as positive numbers, with the trivial exception when a child does not know what the word "zero" refers to.Since the same tasks were solved efficiently in the object task, it confirms that the difficulty was linguistic in nature and not conceptual.pairs with positive numbers vs. pairs with zero as a within-subjects factor) × 3 (number knowledge groups as a between-subjects factor); in verbal comparison version: 3 (number types: pairs with positive numbers vs. pairs with "zero" vs. pairs with "nothing" as a withinsubjects factor) × 3 (number knowledge groups as a between-subjects factor); in addition and subtraction tasks: 3 (operand types: positive operands vs. zero vs. nothing as a within-subjects factor) × 3 (number knowledge groups as a between-subjects factor).Post-hoc tests are LSD tests.

Meta-knowledge about zero
Smallest number.Most children thought that the smallest number was "one" (Table 3).
Although some "zero"-knowers proposed the zero as the smallest number, even most of the "zero"-knowers thought that the smallest number was "one."None of the children proposed a negative number as a reply, although in one of our pilot studies it happened once.A chi-square test on the proportion of responses (0; 1; other numbers; nothing; does not know) showed a significant effect, χ 2 (4, N=38)=26.47,p<0.001, reflecting high proportion of "one" responses.
A chi-square test on the responses between the four number-knowledge groups revealed significant difference, χ 2 (12, N=38)=23.5,p=0.024, most probably reflecting the heterogeneous responses in the subset-knower and "zero"-not-knower group, and the relative uniform "one" responses in the other groups.
At the end of the session (after solving all other tasks) the smallest number task was asked again, repeating basically the very same pattern as at the beginning of the session.This result is in a clear contradiction with the comparison task: in the comparison task children knew that zero was smaller than one, however, they believed that one was the smallest number (even "zero"-knowers believed so).(Although the comparison task analysis presented all the trials with any positive integers collapsed, critically, preschoolers also solved the specific "zero"-1 comparisons correctly: for subset-knower "zero"-not-knowers 80% correct responses in the object version, for cardinality-principle-knowers "zero"-not-knowers 100% in the object version, and for cardinality-principle-knowers "zero"-knowers 83% in the verbal version and 100% in the object version.)This result repeats the pattern that can be seen in the data of Wellman and Miller (1986).One could argue that the children might have misunderstood the task.However, it is hard to imagine that they misunderstood the smallest number task, when cardinality-principle-knower children understand conceptual properties of numbers, such as number words for large numbers with unknown position in the counting list are numbers (Lipton & Spelke, 2006), and they understand how set size change goes parallel with the counting list steps (Sarnecka & Carey, 2008).A possible resolution for this result is that preschoolers believed that zero did not belong to numbers.The next task tested more explicitly whether children regarded zero as a number.
Is it a number?Most groups thought that "two" and "three" were numbers, while "pop" (sound) and "kitten" were not numbers, although the subset-knower and "zero"-not-knower group showed an approximately random performance (Figure 4).This means that except for the latter group, preschoolers understood the task.Critically, while the word "nothing" was evaluated as not a number, the status of the "zero" is more uncertain.This result is in line with the data that even some of the adults do not think that zero is a number (Wheeler & Feghali, 1983).A 4 (number knowledge groups as a betweensubjects factor) × 6 (words to evaluate as a within-subjects factor) ANOVA on the proportion of correct responses (not on the proportion of the "yes" responses as displayed in Figure 4) revealed a main effect of group, F(3,34)=3.008,p=0.044, ηp 2 =0.21, the subset-knower and "zero"-not knower group showing poorer performance than the other three groups, all ps<0.037, and an interaction between the two factors, F(15, 170)=2.12,p=0.011, ηp 2 =0.158.To find the source of the interaction, a similar ANOVA was run, excluding the subset-knower and "zero"not knower group.The 3 (groups) × 6 (words) ANOVA revealed only a main effect of words, F(5,105)=3.34, p=0.008, ηp 2 =0.137.Post-hoc LSD-tests showed that the error rate for "zero" was higher than for "pop", "kitten" and "nothing."Thus, in the three groups evaluating "zero" was ambiguous, while the excluded subset-knower and "zero"-not knower group showed random responses reflecting that they did not understand the task.Figure 4. Proportion of "yes" responses in "is it a number" task at the beginning of the session.Horizontal line at 50% shows the random choice level.Similar to the case of the smallest number task, at the end of the session (after solving all other tasks) "is it a number" task was asked again, and the pattern of the replies did not change.
The results here show that while the positive numbers are considered to be numbers by the preschoolers, they do not think that the word "nothing" is a number, and they are ambiguous about the numerosity of the word "zero."The ambiguity of the word "zero" is partly understandable in the "zero"-not-knower group, since they can only guess.However, the result is more informative in the "zero"-knower group, since they could use the zero value correctly in numerical tasks even when the number was labeled as "zero."How can these differences be interpreted?There could be different considerations that might influence the decision.First, all these words could be and were used by the preschoolers appropriately in numerical tasks (at least in the cardinality-principle-knower and "zero"-knower group), thus, the difference between the number status of the words cannot be caused by the children's numerical knowledge.Second, the word "nothing" can be used more generally in non-numerical cases, and additionally, unlike other number words, "nothing" cannot be used before nouns.These conceptual and linguistic differences might explain, why the word "nothing" is considered to be less numeric than the positive numbers and "zero."Importantly, none of these numerical, conceptual and linguistic considerations could explain why the word "zero" is thought to be less numeric than positive numbers.Thus, we argue that the results reflect that preschoolers do not regard "zero" as a typical number.(Although the word "nothing" is also thought to be less numeric than positive numbers, one cannot tell how strong the mentioned conceptual and linguistic viewpoint influenced this decision, and consequently, how strong the real perceived number status of "nothing" influenced the decision.)This could mean that either (a) at least some of the children do not regard "zero" as a number, or (b) judging the numberness of a value on a continuous scale (e.g., prototypicality of the value as a number) "zero" is not considered as a typical number, or (c) most children are confused about the number status of zero, and they guess in this task.To summarize, this result is consistent with the idea that children do not think that "zero" is a typical number.

Reliability of the data
Forty preschoolers participated in the study, who were categorized into four groups in the analyses, which resulted in rather small groups.One might question how reliable our results can be.(1) Obviously, significance tests take control of the small sample sizes.(2) Moreover, a similar study was run before the study we reported here.The present study is the improved version of our first measurement.The main differences between our previous and the current measurements are that (a) the order of the tasks was modified, since in the present version all tasks that could potentially train the children were at the end of the session, (b) a few simple control tasks were removed, and (c) there were minor improvements in the stimuli and instructions to have a better control over the stimulus properties and the verbal conditions.Importantly, the pattern of the results is the same in the two studies, thus, the results reported here could be replicated, and the results can be considered to be reliable.(3) Additionally, the Smallest number and the Is it a number tasks were repeated within the session, and as we have mentioned, the results showed the very same pattern, strengthening our claim that our results are reliable.(4) Finally, the small size of the subset-knower, "zero"-knower group (4 children) cannot distort our result, because their data in the operation tasks was not used at all, and almost all of our results rely on the data of the other three groups.

General discussion
Our results shed new light on several important aspects of the understanding of zero by preschoolers.First of all, our data demonstrates that preschoolers understand the handling of empty sets in a numerical context, and they can appropriately apply it in various numerical tasks, such as comparison, addition and subtraction (Aim 1).Importantly, even subset-knowers can process empty sets appropriately: their performance with empty sets was comparable with positive values if the mostly unknown word "zero" or the mathematical linguistic form was not used (Aim 3).This result is in contrast with the result of the Wellman and Miller (1986) study which suggested that handling zero is difficult for preschoolers, and the contradiction partly arises from the fact that in that study children potentially could not understand the word "zero" and the mathematical linguistic form (see also below).Additionally, while Wellman and Miller (1986) propose that children understand zero only when they know that zero is the smallest number, we suggest that handling zero in numerical tasks is a sufficient criterion, while knowing that zero is the smallest number is a knowledge that is irrelevant for evaluating preschoolers' operational capabilities.
Second, while preschoolers can handle empty sets, they have difficulties with the mathematical language (Aim 2).One component of this linguistic difficulty is the knowledge about the meaning of the word "zero."The children who did not know the word "zero", could not solve the tasks in which the number zero was denoted purely by this label, while they could solve the same task with different wording, suggesting that the difference is whether they know that the "zero" label refers to a concept they can use otherwise.Additionally, some children have explicitly noted, that they do not know the meaning of the word "zero."As a second linguistic component, children had difficulty with the "mathematical" formulation of the tasks (e.g., "give zero balls to the bird").Importantly, they could solve the task if (a) zero was denoted nonverbally, and (b) the "natural" linguistic form of the zero-related statements (e.g., "do not give any balls to the bird") was used.This means that again, the difficulties described here are linguistic but not conceptual in nature.It is not surprising that children had problems with the mathematical formulations, since this form is mostly used for mathematical and more formal purposes, and preschoolers probably hear it infrequently.
Third, children had problems with the number status of zero (Aim 4).First, they were unsure whether zero is a number.Second, they showed a contradiction: they thought that 0 is smaller than 1, but they thought that 1 was the smallest number (the same contradiction was apparent in the study of Wellman & Miller, 1986).This contradiction can be thought of as another reflection of not being sure if zero is a number: while 0 is smaller than 1 ("nothing" is less than anything else), 1 is the smallest number, because 0 is not a number.

The dual nature of zero
Why do children not think that zero is a typical number?Several explanations could offer a solution for this problem.
A first group of explanations might suggest that linguistic or cultural factors might cause the difference between zero and positive integers in some tasks.We argue that these explanations cannot account for the present results.(1) The special status of zero might be explained by its rare use in everyday language (Dehaene & Mehler, 1992).However, after understanding the cardinality principle children understand some common properties of numbers, even if that number is beyond the limit of their counting list (Lipton & Spelke, 2006).Critically, the frequencies of those numbers are comparable to the frequency of zero.Consequently, the relatively low frequency of the word "zero" cannot explain why children do not regard zero as a number.(2) Another possible example of a linguistic explanation could suggest that zero is categorized incorrectly, because empty sets are handled linguistically in a different way than positive numbers: mostly the word "nothing" is used instead of "zero", and "natural" sentences are used instead of the "mathematical" versions.However, "zero"-knowers who are familiar with these linguistic forms are as unsure about the number status of zero as "zero"-not-knowers.Additionally, in a similar case of the number one, although children learn the distinction between one and many (i.e., other positive numbers) grammatically which helps them to learn the meaning of "one" (Carey, 2009), they still think that one is a number.Thus, in this case, the linguistic difference does not result in different categorization of different values.(3) A final possible linguistic-cultural explanation could propose that a counting list usually starts with one, which could exclude zero from the set of numbers (Merritt & Brannon, 2013).However, at least in some cases this decision is not based on the counting list.Some adults also think that zero is not a number, and their justifications show that they consider a different information in this categorization: they say that zero is not a number, because "a number is the abstract value of the quantity of a set" or "it has no value" (Wheeler & Feghali, 1983).Additionally, it is unclear why this linguistic phenomenon would have an effect on categorization if the previous potentially stronger linguistic effects do not play a role in the categorization of zero.Importantly, none of these linguistic-cultural explanations can account for the fact that zero is handled differently only in some tasks, while it is handled similarly to the positive integers in some other tasks.To conclude, although we cannot exclude entirely some linguistic or cultural influences, it seems more likely that the cause of the dual nature of zero should be found somewhere else.
A second group of accounts supposes that there are representational causes why zero is not thought to be a typical number.(1) One can imagine that the representation that supports positive values cannot store the value of zero, hence, zero should be stored in a different system.The most frequently referred model suggests that semantic numerical processing is driven by the approximate number system that represents numbers in an imprecise way (Feigenson et al., 2004;Merritt & Brannon, 2013;Moyer & Landauer, 1967).This system might include zero, although this ability can be debated (Dakin et al., 2011;Dehaene et al., 1993;Merritt & Brannon, 2013;Ramirez-Cardenas et al., 2016).It is possible that while the positive numbers and in some cases the zero value could be handled by this system (as in the comparison task), in some other cases the zero value should be handled by another system (as in the number status of zero task).However, it is not clear what this alternative system could be, and why the approximate number system sometimes could not handle zero.Overall, it is hard to give a coherent explanation how the approximate number system could account for the special status of zero.(2) Another possible explanation suggests that understanding numbers relies on the conceptual understanding of the items (e.g., objects) in a set.For example, Carey (2004Carey ( , 2009) ) proposes that children induce the conceptual understanding of how counting can specify the size of a set relying on the set-templates in long term memory which templates are based on the activation of visual indexes.Another model explaining why sign-value notation numbers (e.g., Roman notation) are easier to understand than place-value numbers (e.g, Indo-Arabic notation) proposes that sign-value notations are more similar to a notation in which the powers are denoted by objects or groups (Krajcsi & Szabó, 2012).In the aforementioned models, numbers could be considered as the properties of those items or objects in a set, and since in an empty set there are no items or objects, it would not make sense to talk about the property of the nonexisting items, and accordingly, zero could not be a number.The case of a property for a missing item is similar to the paradox introduced by Lewis Carroll.In his story a smiling cat disappears at some point, while the smile of the cat remains visible.In a similar way, the property (here, the numerosity) of the non-existent items is not meaningful, and the lack of items cannot be described similar to a set of items.This model is also in line with the justifications of some adults about why they do not regard zero as a number, when they mostly state that numbers describe the items in a set (Wheeler & Feghali, 1983).Importantly, this view can explain why zero has a dual nature: while empty sets can be handled in the numerical tasks manipulating the objects or items, the status of zero is special, since it cannot represent the property of the items that are not present, but it represents the lack of those items.Therefore, we conclude that conceptual understanding of numbers as items or objects can explain why zero is handled in a special way.

Educational consequences
The results and the interpretation we presented here have some educational consequences.Teaching and understanding the properties of zero is difficult (Frobisher, 1999;Lichtenberg, 1972).Understanding some simple properties, such as the parity of zero or the number status of zero, could be problematic even for mathematics teachers (Hill et al., 2008;Wheeler & Feghali, 1983).Based on our results it is possible to offer some educational recommendations for schools or even for preschools.At the same time, the efficiency of these recommendations might also validate and test our description of zero development.(1) Children could handle empty set, thus, no further effort is needed to introduce this concept to them.(2) However, they are not familiar with the mathematical linguistic formulations of the problem, thus, one might teach these linguistic forms.(a) One linguistic component to teach is that the word "zero" means "nothing" or "doing nothing" in mathematical problem solving.(b) Another linguistic component to teach is the mathematical form of sentences, thus, we can recite both the "natural" and the "mathematical" version of the task, stressing that the two versions refer to the same thing: e.g., "do not give any balls to Suzy Sheep, or you can say, give zero balls to Suzy Sheep."(3) Additionally, children do not know about the number status of zero, they are not sure that zero is a number.According to one possibility, one can explain that zero is also a number, because we can use it in counting and other mathematical tasks.However, this solution raises a problem.If our explanation for the present results is correct, then children are not sure that zero is a number, because they consider that number is a property of items, and zero describes the lack of items, a quantitatively different state.This "no items" state can handle empty sets in numerical tasks correctly, although it is not considered to be a number.Importantly, in the preschooler years (and in most educational systems in the first school years) there is no experience that could demonstrate that the "no items" state is actually a number.For example, most adults know that zero is a number, because it is only a step on the number line among the positive and negative integers, or because in written multi-digit calculations zero is handled in a similar way as the other digits.Still, preschoolers do not have any experience that could reveal the numerical nature of the "no items" state, because for example, they do not know negative integers, or they cannot make multi-digit calculations.Thus, although one can explain, that zero is a number, because we can use it in numerical tasks, still, this information cannot be justified in their experiences.Consequently, it seems more reasonable that it is unnecessary to give any education on this issue before the school years, and it could be better to teach about the number status of zero only when this information can be justified and which can be built up in a reasonable way.To sum up, (1) preschoolers already know how to handle empty sets in basic numerical operations, although (2) mathematical language of referring to empty sets can be taught, while (3) teaching the number status of zero is probably unnecessary at that age.
To summarize, preschoolers can handle zero on the same level as they handle positive integers.Although the linguistic form can cause difficulties for them, this is independent of their conceptual understanding.However, preschoolers are not sure whether zero is a number.This could be caused by the set-based representation of numbers: numbers could be the properties of items in a set, while an empty set does not include any items, consequently, zero cannot be a number in this view.

Figure 2 .
Figure 2. Number of children understanding the mathematical ("giving zero") version of the zero task.

Figure 3 .
Figure 3. Proportion of correct responses in comparison (top), addition (middle) and subtraction (bottom) tasks in the non-verbal (left) and verbal (right) version with positive integers, zero and nothing (columns) within the different number knowledge groups (categories on x axes).Error bars show 95% confidence interval.Horizontal line at 50% in the comparison task shows the random choice level.

Table 2 .
ANOVA results of the operations tasks.All analysis were run on the correct response ratios.The following designs were applied: in object comparison version: 2 (number types: