Comentarios finales

A few experimental studies have examined the interpretation of pronouns and the use of the Φ-features number and gender. Jakubowicz (1984) tested the effects of incongruence in gender and number between the pronoun and the c-commanding antecedent DP on children’s interpretation of pronouns. She used an act-out task in her experiment and found that English three- to five-year-olds performed non-adult-like when there was number incongruence between the main clause subject and the embedded ECM-subject (8).10 According to Jakubowicz, in half of

8 _{In a later study, Baauw & Cuetos (2003) claim that children’s knowledge of features}

of pronouns is intact but that the problems they have interpreting pronouns in ECM sentences are related to the lack of processing resources and children’s inability to use (access) the number feature specification on pronouns.

9 _{I also tested children’s ability to access and use the gender feature in these}

sentences. The person feature was left out because it was difficult to test, especially in the present experimental setting, and it is not directly relevant in the given theoretical framework.

10 _{Lente, van (2002) used a truth-value judgment task to test ECM constructions with}

the instances children failed to take into account the [number] feature of the pronoun and linked it to the closest c-commanding DP. Their performance improved considerably when there was gender incongruence, such as in (9).

(8) Cathy said that Sue and Mary wanted her to kick the balls. (± 50% adult like performance)

(9) John said that Cathy wanted him to pet the cat. (± 80% adult like performance)

There is an alternative explanation to the one offered by Jakubowicz for children’s chance-level performance in sentences such as (8). It is not necessarily the inability to use the [number] feature that causes children to allow the pronoun her to refer to the closest c-commanding DP; in the case of (8) the closest c-commanding DP is Mary. The problem may lie in children’s interpretation of plural DPs, such as [Sue and Mary], which can be interpreted as a collection of individuals (collective interpretation), meaning that both Sue and Mary wanted Cathy to kick the balls at the same time. The plural DP can also quantify members of a collection, in which case the plural DP is accompanied by a distributive operator (for details, see Roberts, 1987 and Heim, Lasnik & May, 1991). The members of a group [Sue and Mary] are then treated separately (distributive interpretation), meaning that Sue wanted Cathy to kick the balls now and Mary wanted the same thing yesterday. Avrutin & Thornton (1994) found that children have problems with the collective interpretation of plural DPs. The distributive interpretation seems to be less problematic for them. The errors children make in assigning reference to her in (8) could then be related to their preference for a distributive interpretation and not their inability to access or use the [number] feature. They could be interpreting the intended collective DP [Sue and Mary] as distributive DP [Sue] and DP [Mary], quantifying each member of the collection.

In the distributive scenario in (8), each of the DPs would be represented as a separate individual file card in discourse. Creating one event card would violate the number of arguments (individual file cards) the verb want selects. In that scenario the verb would have three antecedent in French. The children she tested performed at chance level in all ECM conditions, with congruence and incongruence in number. She interpreted these results as providing evidence for the number underspecification hypothesis. The main problem with this experimental study is the fact that van Lente used only one item for each experimental condition. Needless to say, it could well be the case that a particular item could have affected children’s responses.

arguments, two subject DPs Sue and Mary, which correspond to two individual file cards and one complement event card. The event of

wanting must therefore be split. An event card would be created for each

of these individual cards (DPs), resulting in two separate events of

wanting. In Jakubowicz’s experiment testing sentences such as (8),

children are possibly allowing the pronoun to corefer with DP [Mary] as being a part of one event of wanting and with DP [Sue] as belonging to another event of wanting.11 In that case they would be allowing a

discourse dependency between the pronoun and each of the individual file cards. Therefore, the cause of children’s mistakes in Jakubowicz’s study with sentences such as (8) does not necessarily need to be related to their inability to interpret the number feature as such. It could be the case that they are unable to interpret plural DPs collectively.

In another study, Scholes (1981) examined whether children are sensitive to gender or number information on the pronoun using a picture selection task. He tested sentences such as (10) and (11) among others.

(10) Someone is touching him. (11) Someone is touching them.

He found that children younger than four allow him in (10) to refer to a female individual or to several male individuals, and them in (11) to refer to several individuals or to just one. He concluded that this age group does not have the knowledge of gender or number. He further indicated that five-year-old children can use gender information, and at the age of six they become sensitive to number distinctions. The problem with this study lies in the materials used to test the relevant conditions. For example, the sentence in (10) was paired with two pictures from which children had to choose the one that corresponded to the target sentence best. The first picture represented a male individual touching with each of his hands two other male individuals. The second picture depicted

11 _{Independent evidence from previous studies indicates that children tend to}

quantify over events rather than individuals (for data on children, see Philip, 1995, for data on Broca’s aphasics, see Avrutin & Philip, 1998). Adult speakers also quantify over events. Take for example:

(i) 10 cars passed by a tollbooth.

Most of the adult speakers would agree that it was 10 cars that passed by a tollbooth because the event of passing by a tollbooth is what counts and not the number of individual cars. After all it could be the case that only one car passed by the same tollbooth 10 times.

two different male individuals each one touching the same third male individual. Both of these pictures are correct responses to the target sentence since in each picture someone is touching him (another male person). The only difference between the two pictures is that the action of touching is distributed over two individuals in one picture and in the other it is collective in the sense that one and the same individual is being touched by two other men.

The two studies discussed in this section fail to provide conclusive evidence that children have problems using the information carried by [number] and [gender] features of third person pronouns. In order to test children’s knowledge of these two features and their use of this knowledge, I tested their interpretation of pronouns in simple transitive and ECM sentences, where I manipulated the number and gender of the pronoun and of the antecedent DP. If children possess the knowledge of these features and can use it, then the incongruence between the pronoun and the local DP should help them correctly reject a dependency between the two elements. If this is not the case then their performance should not be different in sentences where there is such incongruence versus sentences where the pronoun and the local DP have the same number and gender.

5.3.2 Experiment 6

5.3.2.1 Subjects

In the experiment testing incongruence in number and gender in simple transitive sentences 33 Dutch children (mean age = 5.2) were tested. Their performance was compared to the performance of a control group of 11 Dutch adults (mean age = 31).

In the experiment testing the knowledge and use of the number feature in ECM sentences I tested 19 children (mean age = 5.3) and 11 controls.12 In these sentences only the incongruence in number between

the pronoun and the local antecedent DP was tested. It was impossible to include these sentences in the experiment testing the incongruence of features in simple transitive sentences because of the truth-value judgment task used to test the latter. In order to test the effect of number

12 _{Like in the experiment testing number incongruence in ECM sentences in}

agrammatism discussed in Chapter 2, ECM sentences testing the same incongruence in children were a part of another experiment. These sentences served as fillers in the experiment testing children’s’ interpretation of pronouns in VP ellipsis, which will be discussed in section 5.5.

incongruence in ECM constructions the same subjects had to be tested with the same method as in Experiment 1, i.e. a picture selection task.

5.3.2.2 Materials and procedure

The method used to test pronouns in transitive sentences is the same as the method used in Experiment 2. Two versions of a test (version A and version B) were designed. Each version contained 5 items for each condition in YES and NO form (see Appendix F for all test sentences). The total number of experimental items per version was 40. There were 15 filler sentences per version. The verb that was used in the YES version of a condition in version A was used in the NO version of the same condition of version B in order to control for the meaning of each verb (for examples of the experimental pictures, see Chapter 2, section 2.2.4.3).

(12) Number/gender congruence (baseline): SINGULAR-SINGULAR Target sentence:

De jongen spuit hem nat.

The boy squirts him (wet).

(13) Number incongruence: SINGULAR-PLURAL Target sentence:

De man spuit hen/hun nat. The man squirts them (wet).

(14) Number incongruence: PLURAL-SINGULAR Target sentence:

De boeven wijzen hem aan. The thieves point at him.

(15) Gender incongruence: MASC.-FEM./FEM.-MASC. Target sentence:

De prinses wijst hem aan. The princess points at him.

The ECM sentences were tested with a picture selection task, as described in section 2.2.4.3, in Chapter 2 (see also Figure 2.1 and Figure 2.10 for examples of pictures that accompanied the Number congruence and the Number incongruence conditions respectively). There were 15 items per condition, exemplified in (16) and (17).

(16) Number congruence ECM:

Eerst hebben de man en de jongen gegeten en daarna [zag de man

hem dansen].

First the man and the boy ate and then the man saw him dance. (17) Number incongruence ECM:

Eerst hebben de man en de jongens gegeten en daarna [zag de man

hun dansen]

First the man and the boys ate and then the man saw them dance.

5.3.2.3 Results

The results testing number and gender in transitive sentences will be discussed first. I examined the results of all 33 children that were tested in this experiment. As we can see in the first row of Table 5.2, the performance of the whole group of 33 children on the baseline condition with a pronoun in a transitive sentence congruent in number and gender with the c-commanding antecedent is significantly above chance. Examination of the individual data revealed, however, that there were children who performed at chance level and children that were above chance.

The assumption that children are supposed to have problems with morphosyntactic features had originally been postulated to account for the errors in the cases where there is congruence between the local antecedent and the pronoun in these features. In order to examine whether incongruence between these two elements had an effect on children’s interpretation of pronouns, I had to look at children who had problems with pronoun congruence (baseline) condition. Therefore, I decided to separate children who made no errors in the baseline condition (n=21) from those who made at least one error on this condition (n=12). In Table 5.2, the results of the 12 children who made errors with pronouns in the baseline condition are given in the second row (for individual data, see Appendix F, Table F.2).

Table 5.2

Percentages correct responses on the NO version of the four experimental conditions for the whole group of tested children (n=33) and the group where each child had made at least one error in the baseline condition (n=12).

Singular-Singular congruence NO Singular-Plural incongruence NO Plural-Singular incongruence NO Gender incongruence NO Children (n=33) 85.5 91.5 78.2 90.9 Children (n=12) 60.0 88.3 68.3 86.7

Table 5.3 shows the percentages of correct responses of children (n=12), controls and agrammatic patients (tested in the experiment discussed in Chapter 2, section 2.2.4.) in the conditions with NO as the expected answer. First, I will analyse the children’s results on the conditions with NO as the correct response. In order to test whether these children performed differently on the singular-singular congruence condition (baseline) versus the three incongruence conditions, a Wilcoxon Signed Rank test was used. Children scored significantly better on two incongruence conditions in comparison with the singular-singular congruence condition (singular-singular vs. singular-plural - Z= -2.555, p<. 011; singular-singular vs. gender incongruence – Z= -2.873, p<. 004). There was no significant difference in their performance on the singular- singular vs. plural-singular incongruence conditions (Z= -.829, p<. 407). As was already discussed in sections 2.2.4.4. and 2.2.4.5. in Chapter 2 (see also Table 2.2, Chapter 2), agrammatic aphasic patients exhibit exactly the same pattern.

Table 5.3

Percentages correct responses on the NO version of the four experimental conditions for the group where each child made at least one error in the baseline condition (n=12), controls and the agrammatic patients discussed Chapter 2, section 2.2.4.4.

Singular-Singular congruence NO Singular-Plural incongruence NO Plural-Singular incongruence NO Gender incongruence NO Children (n=12) 60.0 88.3 68.3 86.7 Controls 98.9 98.9 100 100 Broca’s 85.4 100 93.8 96.9

Table 5.4 presents the percentages of correct responses of children, controls and agrammatic patients (tested in the experiment discussed in Chapter 2, section 2.2.4.) in the YES conditions. There was no difference between the singular-singular congruence condition and the three incongruence conditions (Wilcoxon Signed Rank test - singular-singular vs. singular-plural - Z= -1.000, p<. 317; singular-singular vs. plural- singular – Z= -1.000, p<1.000; singular-singular vs. gender incongruence – Z= -1.000, p<1.000). Like for the NO conditions, the same pattern was found for agrammatic aphasic patients (see Chapter 2, section, 2.2.4.4, Table 2.3).

Table 5.4

Percentages correct responses on the YES version of the four experimental conditions for the group where each child made at least one error in the baseline condition (n=12), controls and the agrammatic patients discussed Chapter 2, section 2.2.4.4..

Singular-Singular congruence YES Singular-Plural incongruence YES Plural-Singular incongruence YES Gender incongruence YES Children (n=12) 100 98.3 100 100 Controls 100 100 100 100 Broca’s 98.9 95.8 96.9 97.9

The results of the experiment testing number in ECM sentences are given in Table 5.5. Children chose the correct picture in the Number congruence condition 46.7 % (chance level) significantly less often than in the Number incongruence condition 85% (chi-square = 33.654, p<. 001). The same pattern was found in the agrammatic patients (for details, see Chapter 2, section 2.2.4.4, Table 2.4).

Table 5.5

Percentages correct picture chosen on the two experimental conditions testing number in ECM sentences in children, controls and agrammatic patients discussed in Chapter 2, section 2.2.4.4.

ECM constructions

Number congruence Number incongruence

Children 46.7 85.0

Controls 97.8 100

Broca’s 56.7 85.6

It should be noted that the two conditions were tested with two different groups of children but using the same methodology. The congruence condition was tested in Experiment 5, discussed in section 5.2.2. The incongruence condition was tested as part of the experiment that will be discussed in section 5.5, again using the same methodology. It was impossible to test the congruence condition with this group of children because that would have increased the total number of items, which would have been impossible to test in one session. I assume that this should not be a problem since the chance level performance on ECM sentences is a robust effect found in children in many different languages (see Chapter 2, footnote 4), so there is no reason to assume that this particular group would not have problems with these constructions. 5.3.3 Discussion

The results obtained in this experiment coincide with the results obtained from the agrammatic patients and do not support the

underspecification account. Incongruence in number between the pronoun and the local antecedent improve children’s performance in both simple transitive and ECM sentences (Experiment 5 and Experiment 6). Children are sensitive to the information about number and gender in both structures and use this information to correctly reject a dependency between the pronoun and the local antecedent DP.

In Chapter 2 I argued against the claim that the source of errors with pronouns in simple transitive and ECM constructions in agrammatic patients lies in their failure to implement Rule-I and the A-chain condition. I proposed that syntactic operations are not available on time in agrammatic patients because of their limited processing resources. This makes other, extra-syntactic ways of establishing referential dependencies, such as coreference interpretations, possible in environments where they are normally ruled out. I would like to propose here that the data on children’s interpretation of pronouns are also better captured within the same economy-based approach as has already been claimed by Avrutin (2004a).

The Primitives of Binding model distinguishes between ECM constructions, where the pronoun and the local antecedent are not coarguments (arguments of the same predicate), and transitive constructions, where these two elements are coarguments (for details see section 2.3.3, Chapter 2). The economy hierarchy plays a role in both structures, and in transitives an additional constraint must also be considered. According to the economy hierarchy, syntactic operations are cheaper for the language processor than extra-syntactic operations (e.g. bound variable or coreference) and are applied first. Only if syntactic operations are not available can other operations be applied. The economy hierarchy is essential in accounting for both aphasic and child data with pronouns. As argued in the beginning of this chapter, syntactic operations are not the cheapest route in child’s language system. Like in Broca’s aphasia, I assume that children’s syntactic processing is delayed. When a child encounters a pronoun in an ECM sentence, syntactic operations will not be available on time. There will thus be no way for the system to check whether a syntactic dependency would in principle be possible and which would thus block a coreferential dependency. Therefore, allowing a coreferential dependency between a pronoun and the local antecedent would in that case not be a violation of the economy hierarchy.

As is the case for agrammatic patients, pronouns in transitive sentences cause fewer problems for children because of lexical restrictions on the verb, which prohibit arity reduction. These restrictions

do not apply in ECM constructions because the pronoun and the matrix subject DP are not coarguments in these structures. Because syntax is not ready on time, an extra-syntactic referential dependency through discourse becomes possible for children as well. All featural specifications and knowledge about when syntactic dependencies can and cannot be established are part of children’s syntactic knowledge, which is in principle available. However, a reduced capacity to use