Spanish-speaking children do not always overuse estar

Current research has shown that young Spanish-speaking children distinguish between ser and estar but fail to provide clear answers to questions such as which copula is easier/first to acquire. Some studies show that children perform worse with estar while others show that ser is mastered with more difficulty than estar. The current study presents the results of two experiments that tested children’s comprehension of ser and estar. The results indicate that i) children compute the inferences associated with estar when natural contextual information is provided, ii) children’s copular distinction is supported by syntactic factors, iii) children perform poorly when copula choice depends on contextual factors alone, and most importantly, iv) children do not assign estar interpretations to ser compatible environments. It is concluded that Spanish-speaking children do not always overuse estar as opposed to ser.


INTRODUCTION
The Spanish copulas ser and estar have received great attention from theoretical linguists and second language acquisition experts but relatively little is known about how they are acquired by Spanish speakers.With the exception of Silva-Corvalán and Montanari's (2008) study of ser and estar acquisition by an English-Spanish bilingual child, in which the acquisition of estar is reported to be delayed, it is generally accepted that Spanish-speaking children master the basic properties of the copulas and use them in an adult-like fashion.However, studies that test comprehension of the copulas with adjectives have shown that 4-to 6-year-old children (Schmitt, Holtheuer & Miller, 2004;Schmitt & Miller, 2007) have problems with the pragmatic dimension of the copulas.In addition, a recent study of natural production has revealed that 2-to 3-year-old children make some morpho-syntactic errors while using ser and estar (Holtheuer, 2009).Hence, acknowledging that children are highly accurate users of ser and estar should not lead us to think that children's early production and comprehension of the copulas are effortless and adult-like.
Up to date research indicates that young children make a general distinction between ser and estar but fail to provide clear answers to basic questions such as which copula is easier/first to acquire and whether children default to one copula in their early productions.While some studies show that children perform worse with estar than with ser (Sera, 1992;Schmitt et al., 2004) other show that ser is mastered with more difficulty than estar (Sera, Bales & del Castillo, 1997;Schmitt & Miller, 2007).
In order to expand current empirical findings this study presents the results of two experiments, a Picture Selection Task (hence PST) and an Acceptability Task (hence AT) that tested children comprehension of ser and estar sentences in controlled discursive contexts.The objectives of the PST are (i) to examine whether Spanish-speaking children know that estar as opposed to ser is used to describe contextually marked properties and (ii) to investigate whether children learn ser and estar by keeping track of the type of adjective (ser adjectives or estar adjective).The objectives of the AT are (i) to investigate what kinds of information (syntactic, semantic and pragmatic) children use for distinguishing ser and estar and (ii) to examine whether children experience more difficulty with one or the other copula.The results indicate that Spanish-speaking children know the basic properties of the copulas but are still developing their acquisition of the pragmatic properties of ser and estar.Crucially, both experiments indicate that children do not assign estar interpretations to ser compatible environments and therefore do not support the idea that children overuse estar.

Distribution and basic properties of ser and estar
Ser and estar are present in different syntactic constructions (e.g. both ser and estar take adjectival and prepositional phrases as complements, only nominal phrases appear with ser but not estar, and only estar is the auxiliary for the progressive form).However, it is the 'copula + adjective' construction the one that has attracted most interest from researchers because it is in this domain where choosing the appropriate copula becomes a complex task that involves knowledge of syntactic, semantic, pragmatic and discursive factors.Notice that while many adjectives are compatible with both ser and estar, not all adjectives can appear with both copulas (* stands for ungrammatical and? for infelicitous sentences): (1) a. Mariana  Moreover, while some adjectives have different lexical meanings depending on whether they appear with ser or estar (1d), other adjectives keep the same lexical meaning but yield a different interpretation of the predicate (1e).
Traditionally the difference between ser and estar has been casted in terms of permanent or inherent and temporary or non-inherent properties (Bello, 1951;Gili Gaya, 1955;Vañó-Cerdá, 1982), but it is now widely accepted by most linguists that temporal notions alone cannot determine the ser and estar distribution (Schmitt, 1992(Schmitt, , 1996(Schmitt, , 2005;;Maienborn, 2003Maienborn, , 2005;;Arche, 2006).Here, we follow Schmitt's (2005) proposal that views ser and estar as the exponents of the permanent and temporary copula as the result of a pragmatic phenomenon.The two copulas are semantically distinct: while ser is underspecified for aspect, estar is specified for aspectual properties.Since ser is aspectually unspecified it gives rise to an implication that a property holds independent of time while estar is a state that holds at a certain time (time t), and therefore gives rise to an implicature of temporariness.Sera (1992) is the first comprehensive study of the acquisition of the Spanish copulas that shows that children master the distribution of ser and estar with different predicates.Sera (1992) examined a Spontaneous Speech sample of two boys (ages 1;6 and 3;6) and an Elicited Production sample from eleven 3-year-olds, twelve 4-year-olds, eleven 5-yearolds, twelve 9-year-olds and five adults.In her study, children used estar with locative predicates and ser with nominal predicates in an adult manner.However, the children differed from adults in two important ways: first, they overused estar as compared with ser in adjectival predicates and second, they incorrectly used estar in locative predicates with event subjects (estar is used to locate objects and ser to locate events).Given that children did not distinguish between the two copulas based on the semantic properties of the subject (object or event), Sera (1992) suggested that children's ser and estar choice may be determined by syntactic factors.

Child language acquisition of ser and estar. Previous research
In another study, Sera et al. (1997) tested whether the use of ser and estar affected the ability of Spanish-speaking children (age range 2;11-5;10) to identify real and apparent properties.Even though several children used ser for describing temporary properties, they concluded that Spanish-speaking children correctly matched ser with permanent properties and estar with apparent properties.Unfortunately, in this last study no adults were included and therefore we cannot compare adults' and children's performance.
More recently, Schmitt et al. (2004) examined whether children allow pragmatically appropriate interpretations to ser and estar by using a Picture Matching Task (PMT) and an Acceptability Judgment Task (AJT).There were twenty children (mean age 4;6) in the PMT and sixteen children (mean age 4;4) in the AJT.In the PMT, the child had to choose between two pictures, one depicting a canonical (permanent) property of a character such as a tall giraffe and the other depicting a non-canonical (temporary) property such as a short giraffe standing on top of a table.The participants were asked to point to one of the pictures as an answer to the experimental question (e.g.Which is (ser/estar) tall?).It was found that although children distinguished between the two copulas, unlike adults, they assigned ser interpretations to questions with estar.The authors concluded that children's problems may have been due to their pragmatic inability to rule out estar in typical ser contexts.In their other experiment, the AJT, participants judged whether the ser/estar experimental sentence was acceptable in three different conditions that controlled for the kind of information provided: syntactic, semantic and discursive-pragmatic.The children performed unlike adults in all three conditions but performed worst in the condition that controlled for discursivepragmatic information.Like in the PMT, the children were more willing to accept estar in ser appropriate contexts than vice-versa suggesting that children overuse estar.
In another study, Schmitt and Miller (2007) showed that Chilean and Mexican Spanish speaking children understand some pragmatic properties associated with ser and estar.In one task involving only color adjectives, children (4;5-6;3, mean age: 5;3) approached adult levels in the use of both ser and estar but unlike adults, and similarly to Schmitt et al. (2004), children still allowed estar to be associated with permanent properties.In another task with scalar adjectives (e.g.'tall'), children (4;7-6;0, mean age: 5;5) patterned with adults in the estar but not in the ser condition.
The research findings so far present us with complex and inconsistent data with regard to which copula is overused and/or acquired first.Even though the above studies are not easily comparable (different methods and different tasks), it is obvious that we need to carry out more experiments to understand better how ser and estar are acquired by monolingual speakers.The present study adds to the current body of research by presenting two novel experiments that tackle the comprehension of ser and estar with adjectives in Chilean and Cuban Spanish.

Studies
In order to further explore children's understanding of the copulas we carried out two experiments.The first tested comprehension of estar by using a Picture Selection Task (PST) design and the second experiment tested the use of both ser and estar in specific contexts by using a version of McDaniel and Cairns' (1996) Grammaticality Judgment Task, an Acceptability Judgment Task (AT).Consistent with the Semantic Subset Principle (Crain & Thornton, 1998) it was hypothesized that children may acquire estar before ser.This principle leads children acquiring language to first choose the representation that is true in the smallest set of circumstances.Since estar can be a subset of ser (Luján, 1981), we predicted that if children follow the Semantic Subset Principle, they should acquire estar before ser.In addition, following general experimental findings that children perform significantly worse when the interpretation of linguistic terms requires knowledge of the interaction of syntactic & discourse-related constraints (Avrutin & Wexler, 1992;Noveck, 2001;Papafragou & Musolino, 2003;Schmitt et al., 2004), it was predicted that children would have more problems when the copula choice depends on discourse factors alone.

Goal of the experiment
The goal of the PST was two-fold.First, we wanted to test whether Spanish speaking children know that estar is connected to the discourse and gives rise to marked interpretations of the subject in adjectival predicative constructions.It is known that estar differs from ser in that the former is always linked to the discourse while the latter is not (Maienborn, 2003(Maienborn, , 2005;;Clements, 2005;Schmitt, 2005;Arche, 2006).If children know the discursive-pragmatic properties of ser and estar they should consistently select the picture that is more appropriate to match a marked interpretation of the predicate.Alternatively, we would expect two outcomes: either children will select any picture at random (therefore, showing no preference for the specific estar representation) or they will choose the picture that is consistent with a ser interpretation.
Our second goal was to find out whether children use typicality cues.Sera (1992) reported that children use certain adjectives with ser and a different set of adjectives with estar and suggested that children might use adjective class (ser adjectives/ estar adjectives) in order to learn the distribution of the copulas.In order to control for typicality effects we used experimental sentences that only contained estar with adjectives that typically appear with ser.The adjectives used in this task were selected based on a norming study in which 19 adult native speakers of Spanish had to fill in the blanks of adjective predicates with either ser or estar (see Appendix A for items used in the norming study).Since the experimental sentences included those adjectives that typically occurred with copula ser in the production of adults, by using estar, we controlled for the potential children's preference for allowing typical ser adjectives to appear only with ser.

Participants
Thirty Cuban-Spanish speaking children with mean age of 60.7 months, an age range of 39-87 months, SD = 11.8 (mean age 5;0, range 3;3-7;3) and eighteen Chilean-Spanish speaking children with a mean age of 58.5 months, an age range of 49-71 months, SD = 7.3 (mean age 4;10, range 4;1-5;11) participated in this task.There were five 3-year-olds, six 4-yearolds, fourteen 5-year-olds, four 6-year-olds and one 7-year-old in the Cuban sample (14 boys and 16 girls).There were twelve 4-year-olds and six 5-year-olds in the Chilean sample (8 boys and 10 girls).The Cuban children attended public kindergartens located in Havana City and came from a mixed socioeconomic background.The Chilean sample consisted of middle class children that attended a private kindergarten center in Punta Arenas.In addition, twenty-five Cuban adults and nine Chilean adults were tested as controls.The difference in the number of subjects between the Cuban and Chilean groups is not a methodological problem since what we are testing is the difference in performance between children and adults that share the same dialect.
The reason for including two dialectal varieties of Spanish is that prior sociolinguistic research has shown that there is variation in the use of the copulas among Spanish dialects (Silva-Corvalán, 1991, 1994;Gutiérrez, 1992;De Jonge, 1993;Fernández, 1999).In general, these studies refer to the overextension of estar in contexts were ser is the canonical copula.
In a diachronic study of the grammaticalization of ser and estar in Romance languages, Batllori and Roca (forthcoming) argue that the progressive reduction of ser and the extension of estar is a strategy used by the child since it leads to structural simplification.Therefore it is important to observe whether children of different dialects of Spanish, that is, different inputs, differ in their representations of both copulas.

Experimental methodology
The experimental design that we used was a PST.Each different trial consisted of two pictures, a description of the pictures (to draw equal attention to both pictures) and a experimental question.The question was always asked with copula estar but the pictures were different in that one of them depicted a specific unmarked characteristic or property consistent with a ser interpretation while the other depicted the same characteristic enhanced by a particular situation, consistent with an estar interpretation as illustrated in (2).
(2) Example of Pictures used in the PST.
A finding that children behave as adults would suggest that children know that estar adds information to a previously known state.In contrast to Schmitt et al.'s (2004) study in which one picture depicted typical properties while the other depicted atypical properties, the current experiment used pictures in which the character retained her normal properties (e.g. a normally tall giraffe on top of a table).That is, we created visual contexts that were consistent with Luján's (1981) account that an estar representation entails a ser representation.On the contrary, the pictures used by Schmitt et al. (2004) used contexts in which the appropriate pragmatic inferences cancelled the entailment relationship (e.g.Picture A depicted a tall giraffe on the floor consistent with both a ser and estar tall interpretation while picture B depicted a dwarf giraffe on top of a table consistent with an estar tall interpretation).That is, the contexts they used gave rise to the implicature that estar + property implies negation of the ser + property.The reader is referred to the extensive literature on scalar implicatures for details (Grice, 1989;Chierchia, 2000;Noveck, 2001;Papafragou & Musolino, 2003;Guasti, Chierchia, Crain, Foppolo, Gualmini & Meroni, 2005).
Another important departure from the study of Schmitt et al. (2004) was that we only tested estar.This was done because we wanted to directly test whether children overuse estar and we also wanted to lower the task demands exerted over the participants.Since in the current task both pictures are compatible with a ser interpretation we simplified the participants' task by using only estar.This way we could be more certain that if differences between adults and children are found they are likely to reflect grammatical knowledge.

Procedure and materials
Both pictures were presented to the child at the same time and the experimenter made sure the child looked at both pictures before presenting them with the experimental sentence.The child's task was to look at both pictures and select (by pointing) the one that best answered the experimental question.As the objective of the task was to test the comprehension of estar, all the experimental sentences asked for a specific property with estar.The experimenter took notes of the picture the child pointed to immediately after the child had made the choice.The adults were tested following exactly the same procedure as the one used with children.The experiment was presented to the children individually by one experimenter.The Chilean children were tested in a quiet area of the school that they were attending to in the Punta Arenas region while the Cuban children were tested in a quiet room at their kindergarten in Havana City.The materials were presented in a session of 5 to 10 minutes.
A total of 26 pictures including fillers for Chileans and 24 pictures for Cubans were used in this task.The pictures were presented to children and adults in the same order.Both pictures depicted the same subject but in different situations.Subjects were presented with a brief description of the two pictures.After the story, the child was asked ¿Cuál jirafa está alta?'Which giraffe is (estar) tall'?Given the context, the expected adult response would be to point to the picture in which the giraffe is on the table.The difference between the pictures is that while the giraffe on the left is intrinsically tall, the giraffe on the right is both intrinsically and extrinsically tall.Thus, although it is true that both giraffes ser tall, it is not completely appropriate to say that both giraffes estar tall in both pictures.Hence if children know that estar is marked, they should select the picture in which the giraffe is on the table because this is the picture that adds more information to the already known state of giraffes being tall animals.
and a control item with adjective rubia [blonde].In addition, fillers that asked about the location of an object were included (see Appendix B for all items).

Results
Given that this was a dual choice task, that is, the subject had to select one from two possible pictures, we tested whether the probability of answering correctly (selecting the picture that better matches the experimental sentence) was higher than .5 or at random.The results of the binomial tests show the probabilities of answering correctly significantly above .5 for all the items except for the control item rubia [blonde].Table 1 summarizes the mean The data from the Cuban subjects were entered into a one-way independent Anova with mean percentages correct as the dependent variable and age as the independent variable with three levels (young children aged 3;3 to 4;7, old children aged 5;2 to 7;3, and adults) in order to check whether there were any differences related to age.Table 2 shows the descriptive statistics for the Cuban group.
The assumption of homogeneity of variances was violated (Levene's Test, p= 0.002).Therefore a Welsch F Test (Anova) was performed indicating that the means of the three age groups were statistically non-significant: F (2, 25.9) = 2.87, p = 0.074.
Likewise, the data form the Chilean subjects were entered into a one-way independent Anova in order to compare the means of the three age groups.Table 3 shows the descriptive statistics.
Again, the assumption of homogeneity of variances was violated (Levene's test, p = 0.00).However, the Welsch Anova could not be performed because the variance of the older group of children was equal to zero.Therefore we used the harmonic mean size that reduces the bias that arises when sample sizes are unequal as in this case (Field, 2009).A Tukey HSD test indicated that the means of the three age groups were not statistically different (p= .07).Nevertheless, even though we do not observe significant differences on the means by age group the results indicate that older children perform better than younger children in both dialectal populations suggesting that the younger children are still progressing in their acquisition of estar.

Goal of the experiment
Like in Schmitt et al. (2004), the main goal of the AT was to test whether and how Spanish-speaking comprehension of ser and estar differs in three different conditions: a lexical-semantic condition named Polysemic Free, a syntactic condition named De Complement and a discursive-pragmatic condition named Context Only.This study goes one step forward by adjusting the results for a number of factors such as attention and yes-bias, in a unified regression framework.Hence we were able to obtain estimates of the difficulty associated to processing each type of information: lexical-semantic, syntactic, and discursive-pragmatic.

Participants
Twenty-seven monolingual Cuban Spanish-speaking children (mean age 62.2 months, range 39-87 months, SD = 11.8, equivalent to 5;2, range 3;3-7;3) participated in this study.There were four 3-yearolds, five 4-year-olds, thirteen 5-year-olds, four 6-year-olds and one 7-year-old in the sample (13 boys and 14 girls).They had a mixed socioeconomic background with all children attending public kindergartens located in Havana City.In addition, 23 adults were included as controls.

Experimental methodology
The experimental design that we used was an Acceptability Task, a version of the Grammaticality Judgment Task (McDaniel & Cairns, 1996).Normally, children need to judge whether a sentence uttered by an experimenter is well-formed or has a correct interpretation.Every trial consisted of a short story for context-setting purposes and one experimental sentence that matched (or not) the context.

Procedure and materials
The task was presented to the children by one experimenter as a game in which the child had to teach Spanish to a puppet called Pepe.The child's task was to listen to a particular short story read by the experimenter and then judge whether the sentence uttered by Pepe was acceptable by saying sí (yes) or no (no).The child also had to feed the puppet a candy if he gave a right answer or a vitamin if he gave a wrong answer.The adults were given a Paper and Pencil test.The experiment was individually presented to the children at their kindergarten in Havana City and took approximately 15 minutes.Children's acceptability judgments were elicited in three conditions as illustrated in Table 4.
The main goal of Condition 1 was to test whether children know that the use of different copulas correlate with the different meanings of a specific adjective.Therefore the Polysemic Free Condition involved adjectives that have different lexical meanings depending on which copula they are used.
In turn, the main goal of Condition 2 was to test whether children know that estar but not ser takes adjectives with de [of] complements.This condition explores a semantic/syntactic restriction in the use of the copula.When an adjective takes a de complement, the only grammatical option is to use copula estar.Thus, children could rely on syntactic information in order to verify whether the choice of the copula is or not acceptable independent of the story and/or the exact meaning of the adjective.Finally, the main goal of Condition 3 was to test whether children know that estar is used with contingent properties ('now') while ser is used with more permanent or inherent properties.This condition is very similar to Condition 1 (Polysemic Free) except that here the adjectives do not have different lexical meanings.In Condition 3 copula ser is used to express that the property is not bound to the discourse and therefore it qualifies as a permanent property.The copula estar, in contrast, is used to describe a temporary property.In order to show adult-like knowledge in this condition, children would have to rely entirely on contextual information.
The predictions for this task are: a) children will perform better in the experimental sentences with estar, b) children will make more errors in Condition 3 since copula choice is determined by context alone, and c) children will perform better in Condition 2 since copula choice is determined by syntactic factors.

Results
In this task, the subjects were expected to answer either 'yes' or 'no' after each trial according to the context set by the short story.It is a well known fact in the psycholinguistics literature that respondents tend to use 'yes' as a default answer.
That is, when subjects are not sure about what the correct answer is, they tend to respond 'yes' to any question regardless of whether the expected correct response is 'yes' or 'no'.This bias is common in experiments with children (Chien & Wexler, 1990;Schmitt et al., 2004;Waterman, Blades & Spencer, 2004).Therefore, first we provide the means per condition and then we use regression analyses to obtain estimates that are free from the 'yes'-bias phenomenon.Researchers in linguistics normally deal with this problem by eliminating the information for items that have an affirmative answer as the correct response (Schmitt et al., 2004).Although this procedure effectively leaves only observations that are bias-free, this is done at the cost of eliminating substantial information, with the consequent loss in estimating precision.Hence, it is important to note that the results vary depending on the statistics used.Regression analyses are extremely useful for controlling extra variables such as attention, experimental noise and response biases that can affect subjects' performance.Since a major objective of this experiment is to explore whether the experimental variables differentially predict outcomes on ser/estar, the use of regression is ideal since it allows us to rank these factors in order of processing difficulty.The general means of the subjects in the three conditions are presented in Table 5.
A Chi-square test shows that the difference between the means of adults and children is only significant in Condition 3, with χ² (1, N= 38) = 6,01, p= .014.In the other conditions the difference between children and adults was not significant (p>.05).A Chi-square test also shows that the means of condition 1, 2 and 3 for children are not significantly different (p> .05).
Now, when we refer to the regression analysis the results change.Table 6 shows the variables recorded in the database, and Table 7 shows the descriptive statistics.
As said previously respondents tend to use 'yes' as a default answer even when the appropriate answer is 'no'.Therefore, we begin the analysis by testing for a possible 'yes'-bias.Because it is observed in children more than in adults, it leads to higher differences in observed performance between adults and children than those that effectively exist.A probabilistic analysis indicates that the probability of adults for answering yes/no is exactly .5 so no bias is observed in the control group.However, this probability is .23 for children, with the hypothesis of no bias rejected at the 1% confidence level.
Once the bias is established, the question remains of how to obtain estimates that are free from it.The first regression estimates the probability of giving a correct answer controlling for the 'yes' bias.The results indicate that the probability of answering correctly for adults (child=0) when there is no bias (dyes = 0, dyes*child = 0) is .84.For children (child=1), this probability is .84-.41 x 1= .43.The difference in probabilities of answering correctly between adults and children is statistically significant (p =.00).It is important to notice that the probability that adults answer correctly is less than 100%.As we are comparing child knowledge of ser and estar with respect to the adult grammar, we focus on the difference in performance between children and adults.
We will argue that the relevant statistic to focus on is the difference in probabilities between adults and children.We assume that the failure of adults answering correctly is due to the fact that the answers reflect not only understanding of the subject on ser and estar but also the quality of experimental design (See the empirical model in Appendix D).
We proceed to report the results regarding the following questions: i) Do children show different performance between the copulas that would allow us to suggest that estar is acquired before ser?, ii) What kind of information (syntactic, semantic or pragmatic) do children use when distinguishing ser and estar, and iii) Do we observe age effects?
In order to answer the first question, we examined whether children show better performance with estar than with ser.As estar can be a subset of ser, it is reasonable to expect that children acquire estar before ser.To test this hypothesis a variable that captures the difference between children and adults in ser/estar comprehension was created (ser*child).
The result indicated that children have less trouble with estar, but the variable's coefficient was not statistically significant.
In order to answer our second question that asked in which order relative to difficulty our three experimental conditions (syntactic, semantic or pragmatic information) are used by children when employing ser and estar, we used the difference in probability between children and adults (or b1), as the object of interest (Table 8).
A t-test for equality of means shows that coefficient (in condition) 3 is statistically higher than 1 (p=.00), and that coefficient 1 is statistically higher than 2 (p=.00).These results allow us to say that for children it is easier to use syntactic information in order to choose between ser and estar.In other words, children performed better in Condition 2 (de Complement) perhaps because the presence of a de complement helps them to select estar over ser.
The results also indicate that children seem to perform better when they have lexical/semantic cues than when they have only contextual cues.Notice that these results are different from the results we obtained by comparing means.When considering the means children and adults performance was statistically different only in the 'Context Only' condition.Here, we found that children and adults perform differently in all conditions and that the difficulty across conditions is clearly ranked.
In order to answer our third question, we tested whether the comprehension of ser versus estar depends on the age of children, but again the coefficient associated with the interaction ser/ estar x age was found to be small and statistically insignificant.We also found a significant positive correlation between the age of children and their percentage of correct responses in the filler questions (p=.00), showing that children do pay attention as they get older.This should of course be of some concern: it could be that all the differences in correct responses in the main questions are due simply to the fact that adults pay more attention to the questions themselves.To address this issue, we did the previous analysis again controlling for the attention factor summarized in the filler variable.We found that the differences between adults and children become smaller, so the attention effect exists, but is quantitatively very small.With the inclusion of the filler variable, the coefficient of interest (on the variable child) is -.38 instead of -.41, a difference of 3 percentage points.In general, the results strongly confirm that the differences in performance are not driven by differences in attention.Finally, it is worth pointing out that we found significant age effects in the fillers but not in answers to the experimental sentences, again casting doubt on a simple explanation of performance driven by the capacity of concentrating of older versus younger children.

Discussion of the PST and the AT
Our main finding is that children behave adult-like in the PST but differed from adults in the AT.This is not in itself a contradiction since the methodologies used are different.In the PST the child makes a choice between two alternative pictures, while in the AT the child is asked to evaluate the 'Acceptance' of either a correct/felicitous or an incorrect/unfelicitous linguistic string according to a specific context set by the story.The AT is much more demanding than the PST because while in the PST the child selects a preferred option by directly accessing the visual context in the AT the she/he needs to remember the story and also compare the two competing representations (e.g. the correct and the incorrect) in order to evaluate whether the sentence provided matches or not the interpretation of the story.Thus it is expected that children will have more problems when copula knowledge is tested by methods that rely on processing capacities (Grodzinsky & Reinhart, 1993).Our line of reasoning is supported by previous studies (Avrutin & Wexler, 1992) that show that children perform poorly when pragmatics is involved.Similarly in the AT children performed least adult-like when the choice of copula was determined solely by context.Therefore, it cannot be argued that Spanish-speaking children do not distinguish between ser and estar.
Having this in mind we turn to discuss the results of each task separately.The main objective of the PST was to explore whether Cuban and Chilean Spanish speaking children know that estar is the discursively linked copula.It was predicted that if children know the discursive-pragmatic properties of ser and estar they should consistently select the picture that is more appropriate to match an estar interpretation.The first important finding of the PST is that children show an adult-like performance in the understanding of estar.This result suggests that they know that estar relates to a specific situation.Crucially, the different performance between the control (the blonde trial) and the experimental items indicates that subjects experience more difficulty when the context requires to draw the pragmatic inference that estar is used in order to negate ser.That is, even though ser and estar are partially synonyms (Luján, 1981) they can be used to make pragmatically different inferences.For example, by saying Marta está linda hoy we may implicate that Marta is not a pretty person.In this task both Cuban adults and children selected the typically blonde girl consistent with a ser blonde interpretation 48% and 63% of the time respectively (chance behavior) while Chilean children selected the ser picture 89% of the time in clear contrast to Chilean adults who only selected the ser picture 11% of the time.Therefore, processing the inferences associated with estar appears to be more difficult in contexts like the one used in the control item (blonde) than in the contexts used for the experimental items.A possible explanation is that the context in the control item but not in the experimental items gives rise to the pragmatic inference that estar negates ser.This result falls in line with most studies on the acquisition of pragmatic inferences (Noveck, 2001;Papafragou & Musolino, 2003;Chierchia et al., 2004) that processing the appropriate computation of referential elements (i.e.inferences) in specific contexts is costly (Grodzinsky & Reinhart, 1993;Reinhart, 2006).That children could draw the estar inferences in the experimental items but not in the control item (as in Schmitt et al., 2004) is also consistent with Reinhart's (2006) explanation about children's failure to compute pragmatic inferences.Children may be aware of the required computation, but unlike adults, they are unable to execute it because they have limited processing resources.Since the context of the control item was more difficult to process than the context in the experimental items (for both children and adults), children were able to compute the inferences associated with estar in the less demanding contexts (experimental) but failed to do so in the more complex one (control).
Regarding the control item, we also need to explain why Cubans and Chileans vary in their performance.When asked ¿Cuál niña está rubia?, Cuban adults and children selected the picture consistent with a ser interpretation 48% and 63% of the time respectively while Chilean adults and children did so 11% and 89% of the time respectively.That is, Chilean adults are more restrictive than their Cuban counterparts for allowing typically ser properties to be described with estar.This result may be related to the process of estar overextension that has been described in the sociolinguistic literature and could indicate that the encroaching of estar into ser territory is more common in Cuban than in Chilean Spanish.
The second finding of the PST is that children do not seem to use typicality cues of adjectives since they assigned estar interpretations to adjectives that are typically used with ser.Hence, we do not have evidence for Sera's (1992) claim that children acquire adjectives with either a ser or and estar status.
Regarding our second task, the AT, our first finding was that children did not perform adult-like in any condition.However, the results obtained allowed us to rank children's performance in the following manner in the order of most to least adult-like: (1) De-Complement, (2) Polysemic Free, and (3) Context Only.Since the difference in performance between the conditions was found to be statistically significant and because every condition controlled for different properties of ser and estar we can interpret this ranking as showing the kinds of knowledge children exploit in order to use copulas ser and estar.The result that children performed better in condition 2, where they had to rely on the syntactic requirement that estar but not ser can take de complements, confirms Sera's (1992) intuition that Spanish-speaking children may use syntactic cues in order to discriminate between the copulas.The fact that children's performance was better in this condition than on the condition that controlled for lexical meaning of the frame Copula+Adjective, suggests that children rely on syntactic information more than in semantic information.Also, the poor performance of children in the condition that controlled for contextual factors alone, supports our prediction that children have more problems with the pragmatic dimension of copular use.
It was also found that children performed better with estar than with ser.However, the difference between children's performances with ser and estar were not statistically significant and therefore, we could not confirm the hypothesis that estar is acquired earlier than ser.
At last but not least, we found that age did not affect children's performance in the experimental sentences.It is important to note, however, that older subjects performed better than younger subjects in the fillers.We interpret this as showing that the experimental results that we obtained were not highly dependent on attention or other extra-linguistic factors that could interfere with subject's performance.Therefore, we consider the differences in performance to reflect grammatical competence.We also found that attention has a statistically significant role in explaining differences between adults and children.That is, as children pay less attention than adults, part of the difference between these two age groups corresponds purely to differences in attention and not necessarily to different representations of ser and estar.Therefore, an advantage of the current study is that we could isolate the effect of ser and estar comprehension.

CONCLUSIONS
We conducted two comprehension experiments of ser and estar in Spanish child language.The first experiment was a PST that tested comprehension of the discourse properties of estar and the second experiment was an AT that tested syntactic, semantic and pragmatic properties of ser and estar.The results expanded the findings of Schmitt and Miller (2007) by showing that Spanish-speaking children know that ser and estar have different discourse properties.
Unlike Schmitt et al. (2004) but similarly to Schmitt and Miller (2007), it was found that Spanish speaking children interpret the pragmatics associated to estar just as adults do.Another important finding was related to the question of whether children divide adjectives as having a ser or estar status as motivated by Sera's (1992) claim that Spanish-speaking children draw a distinction between ser adjectives and estar adjectives.Given that in the PST children assigned an estar interpretation to adjectives that typically take ser, it is unlikely that children acquire adjectives by distinguishing those that typically appear with ser from those that typically appear with estar.The results of the AT, in turn, are consistent with Schmitt et al.'s (2004) finding that the pragmatic dimension of copula use is still complex for 3-to 7-year-old children.The fact that children performed much better in the condition in which estar (but not ser) takes de complements than in the conditions that controlled for lexical/semantic factors, suggests that children are more sensitive to distributional cues than to lexical/semantic cues.Moreover, as children performance was least adult-like in the condition in which the only clues for copula usage were given by the context, we argued that children have more problems with discourse factors.Therefore, in the order of more used information to the least used information, we can arrange them as follows: first, syntactic, second lexical/semantic and third contextual information alone.
An important question dealt with the issue of which copula is acquired earlier in Spanish language.
Although we saw a tendency in the right direction, namely that children performed better with estar sentences, the results lacked statistical significance.However, it is still possible that children use the Semantic Subset Principle in early stages of copula acquisition.Further research will need to address if Spanish-speaking children use the Semantic Subset Principle for distinguishing between ser and estar.
Finally, contrary to Schmitt et al. (2004) and Sera (1992) the results do not lend support to the claim that estar is overused.In the PST the children performed adult-like when assigning the pragmatically appropriate interpretation to the experimental questions with estar but were completely different (at chance) in the control item that was similar to the experimental items used in Schmitt et al. (2004).These results suggest that the pictures used in the PMT of Schmitt et al. (2004) (and our control item) were contextually more complex and harder to process.Therefore the estar overuse observed in Schmitt et al. (2004) is more likely to reflect a problem of limited processing resources by children rather than a problem for drawing the right inferences for estar.Similarly, the fact that in the AT children performed better with estar than with ser is clearly incompatible with an overuse of estar by children.The current results highlight the need for further investigating which copula (if any) is overused in the acquisition of Spanish by monolingual children.

Table 1 .
Mean percentages correct per adjective in the Cuban and Chilean groups.percentage of correct responses per adjective in both adult and children in the Chilean and Cuban group respectively.The p-values that are given in parentheses are calculated in order to show if the percentage per adjective is significantly above chance.

Table 2 .
Means by age groups in the PST (Cuban sample).

Table 3 .
Means by age groups in the PST (Chilean sample).

Table 4 .
Experimental conditions and examples of the Acceptability Task.

Table 5 .
Means per condition in adults and children.

Table 6 .
Variable description for the regression analysis.