The Acceptance of Non-Culminating Sentences with an Agent or Cause by Bilingual Children

The Acceptance of Non-Culminating Sentences with an Agent or Cause by Bilingual Children

Elise Vos

e.j.c.vos@student.rug.nl

Student number: S2359839

University: University of Groningen Program: MA European Linguistics First supervisor: Dr Angeliek M. H. van Hout Second supervisor: Dr. Rasmus G. A. Steinkrauss Date of submission: 28

August 2017

Word count: 12.497

Table of Contents

Abstract ... 5

Acknowledgements ... 6

List of Tables ... 7

List of Figures ... 8

1. Introduction ... 9

2. Theoretical Background ... 10

2.1 Telicity and Perfectivity ... 10

2.2 Primary of Aspect Hypothesis ... 11

2.3 Adult L2 ... 12

2.4 Age of Acquisition ... 13

2.5 Manner Bias Hypothesis ... 15

2.6 Locative Verbs ... 16

2.7 Weak Endstate Hypothesis ... 18

2.8 Transparency Principle ... 20

2.9 Acquisition of Aspect ... 21

2.10 Agent Control Hypothesis ... 23

2.11 Agent Control hypothesis in Child Language ... 25

2.12 EFL in Dutch Primary Schools ... 26

2.13 Different Methods to Test Child L2 Proficiency Level ... 28

2.14 GAPS ... 29

2.15 Hypothesis and Research Questions ... 29

3. Method ... 30

3.1 Participants ... 31

3.2 Materials and design ... 31

3.3 Procedure ... 32

3.4 Scoring justifications ... 36

4. Results ... 37

4.1 English at Home ... 40

4.2 Judgement Teachers ... 41

4.3 GAPS ... 42

4.4 GAPS & Judgement Teachers ... 43

4.5 Wrong Answers ... 44

4.6 Special Pattern ... 44

4.7 Particle verbs ... 45

5. Discussion ... 45

5.1 Bilingual and Monolingual children ... 46

5.2 Particle verbs ... 47

5.3 Manner Bias Hypothesis ... 47

5.4 English at home or at school ... 48

5.5 Proficiency level ... 49

5.6 GAPS for proficiency level ... 49

5.7 Limitations ... 50

6. Conclusion ... 50

References ... 52

Appendix A ... 56

Appendix B ... 57

Appendix C ... 63

Abstract

This study uses Strangmann’s (2015) event culmination test, to research the

acceptance of non-culmination of telic-perfective sentences with an agent subject, and with a cause subject in multilingual pre-schoolers. According to the Agent Control Hypothesis, formulated by Demirdache & Martin (2015), non-culmination interpretations are more acceptable when the subject is an agent, but not when the subject is an inanimate cause. The participants were 34 bilingual children from age 4;8-6;10 and 10 adults. Since all children have a different language background and proficiency level, the Grammar and Phonology Screening (GAPS) test, was used to measure individual proficiency in English. Furthermore, the teachers’ intuitive assessment provided an additional measurement. The children all did really well on the culmination test, and made hardly any mistakes. There was a main effect found for situation type, which means that none of the children had no difficulties with differentiating the full and zero results. There was no effect found for age, hence the children performed adult-like. There was neither an effect found for older and younger children, nor for the children who also spoke English at home. There was an effect found for the teachers’

assessment and situation type, which means that the children with a higher proficiency level performed better on the culmination task. This effect was also found for the GAPS, and the GAPS and the teachers’ assessment correlates (0.65). In this study, the Agent Control Hypothesis is not supported for bilingual children.

Keywords: Event culmination, Agent Control Hypothesis, bilingual children, GAPS, telic-

perfective sentences

Acknowledgements

I want to thank the Groningse Schoolvereniging, and especially the international department for letting me test the children. All the teachers were very helpful with providing information about the different language backgrounds of the children. They were all really kind when I came to test the children, and I hope the proficiency test results will be a helpful tool in teaching the children English. Of course, I also want to thank the children. You made me laugh many times with your funny answers, your reactions to the elephant puppet, or to the sticker reward. You even inspired me to start teacher training this September; I enjoyed the testing so much that I now know what I want to do in the future.

I am grateful to my supervisor, Dr Angeliek van Hout. At first, I only had the idea that I wanted to test children, but not what I wanted to test or how. When it was not possible to test the Dutch children, she came up with the idea to test the children of the international department. I also want to thank Iris Strangmann. Even though I never met her, I am very grateful that I could use the test she made for her thesis, and that I could use her thesis as a guideline. I also want to thank Saar Hommes who helped with understanding the test.

Furthermore, I want to thank Mirte Wilhelm for answering all my questions about

SPSS, and especially about ANOVAs. Of course, I also want to thank the control group, and

especially those that provided me with new names of people I could test. Thank you for

participating, even though most of you never met me before. Last of all, I want to thank

Janneke Beenes for answering my endless stream of questions about English grammar and

helping me with the layout.

List of Tables

Table 1. Overview of number of participants for the culmination task and the GAPS, age, age

range and gender for the groups ... 31

Table 2. Number of participant that speak English at home ... 37

Table 3. Distribution of justifications across categories (see scoring justifications for explanations of these categories) ... 38

Table 4. Mean percentage of acceptance for the four conditions for each age group ... 39

Table 5. Distribution of children in different proficiency levels according to teachers’ judgement ... 41

Table 6. GAPS scores. ... 43

Table 7. Wrong answer per condition ... 44

Table 8. Acceptance of non-culmination per verb ... 45

List of Figures

Figure 1 Acceptence of zero result situations with agent (black) and cause (grey) for

Mandarin monomorphemic verbs (Van Hout et al. 2017, p.7). ... 25

Figure 2. Agent in full result situation (blow-out). ... 34

Figure 3. Agent in zero result situation (blow-out). ... 34

Figure 4. Cause in full result situation (blow-out). ... 34

Figure 5. Cause in zero result situation (blow-out). ... 34

Figure 6. Means of yes responses in the four different conditions for the older and the younger children. ... 40

Figure 7. Means of yes responses in the four different conditions for the children that speak English at home or only at school ... 41

Figure 8. Means of yes responses in the four different conditions for five proficiency levels determined by teachers’ judgements. ... 42

Figure 9. Means of yes responses in the four different conditions for the three categories from

the GAPS scores. ... 43

The Acceptance of Non-Culminating Sentences with an Agent or Cause by Bilingual Children 1. Introduction

This study is about event culmination, and the acceptance of non-culminating situations for telic-perfective sentences with an agent subject. For example, when a pirate tried to destroy a plate, but only hit it with a hammer which means there is zero result.

Demirdache & Martin (2015) formulated the Agent Control Hypothesis, which states that adults are likely to say yes in case of non-culmination when an agent is the subject, but not when the subject is an animate cause. They give examples from Mandarin where non- culmination is allowed. The Agent Control hypothesis was tested for child language by Strangmann (2015) and van Hout et al. (2017). They tested monolingual children that speak English, Dutch, Spanish, Basque, or Mandarin. This study looks at bilingual children, who started learning English when they went to Kindergarten. Since these children just started learning English, it is interesting to see if they do allow the ACH. Some children already speak English at home, and the children’s ages range from 4;8-6;10, which made it necessary to include a proficiency test: the Grammar and Phonology Screening (GAPS) test. This was used to be able to compare the results, since all children had a different language background.

The research question is: Do bilingual children allow non-culmination in telic-perfective sentences?

There has not been much research about bilingual children; most studies focus on adult L2. It is interesting to see if the ACH holds for bilingual children who just started learning their second language. It provides insight about when children acquire telicity and perfectivity, which both are needed to understand the endstate of a verb.

First, a theoretical background is given with relevant literature about telicity, other hypotheses, adult L2, age of acquisition, and second language learning for children.

Furthermore, methods to test children’s language proficiency level will be named. In the

method section, the event culmination task and the GAPS are explained, and in the results section, the results from both tests will be presented. In the discussion, the results will be linked to the literature, and this will be followed by a conclusion. In the first appendix, all the answers that were left out are included, and the second appendix contains the event

culmination test with each three stills from the movies, and the third appendix encompasses the GAPS test.

2. Theoretical Background 2.1 Telicity and Perfectivity

Verbs can have different tenses or have aspect. Saeed (2015, p.114) defines aspect as: ‘a grammatical system relating to time, but the speaker may choose how to describe the internal temporal nature of a situation’. One of the aspect features is telicity. Telicity means that: ‘a verb is telic if its meaning includes a specific moment toward which the event it describes naturally develops’ (Van Hout in press, p.5). Saeed (2015, p.455) defines it as: ‘[a] feature of situation types where the event or process has an inherent end point’. Dowty (1979)

developed a test for telicity with two temporal adverbials: for an hour which indicates an atelic sentence, and in an hour which indicates a telic sentence:

(1) A. Telic: build a bridge in a day /* for a day B. Atelic: watch a bridge * in a day / for a day

If you combine the English perfect or simple past with a telic verb, the sentence will entail an implication of completion:

(2) A. Mary painted my portrait B. Mary has painted my portrait

C. The portrait is finished (Saeed 2015, p.118-119)

If you combine a progressive aspect with an atelic verb, the sentence will not entail an

implication of completion. This is important for this research, since it looks at the acceptance

of non-culmination. The sentences used here are telic-perfective sentences, since this entails completion. Saeed (2015) also points out that the acquisition of tense and aspect is quite hard for foreign language learners, and it is even harder to use it correctly.

2.2 Primary of Aspect Hypothesis

Slabakova (2001) defines aspect as ‘the property which makes it possible for a sentence to denote a terminative (telic) or a durative (atelic) event (Slabakova 2001, p.21) She names in her research about telicity in second language acquisition, the Primary of Aspect hypothesis (POA), formulated by Andersen and Shirai (1996, p.533):

1. “Children first use past marking (e.g., in English) or perfective marking (Chinese, Spanish, etc) on achievement and accomplishment verbs, eventually extending its use to activity and stative verbs.

2. In languages that encode the perfective/imperfective distinction, imperfective past appears later than perfective past, and imperfective past marking begins with stative verbs and activity verbs, then extending to accomplishment and achievement verbs.

3. In languages that have progressive aspect, progressive marking begins with activity verbs, then extends to accomplishment ach achievement verbs

4. Progressive markings are not incorrectly overextended to stative verbs.”

Slabakova looked at multiple studies about telicity in second language acquisition. One of her conclusions about the POA is that

“Learners appear to retain and strengthen the bias of using progressive more with

activity verbs across proficiency levels, while they tend to lose the correlation between

telicity and past tense. In other words, past eventually spreads to all aspectual classes

but progressive does not spread” (Robison 1995; Bardovi-Harlig and Bergström

1996). (Slabakova 2001, p.123)

Another conclusion she draws is: ‘in the early stages of acquisition of tense and aspect morphology, children use more often imperfective marking on activity verbs and past tense marking on accomplishment and achievement verbs, but this is far from exclusive’

(Slabakova 2001, p.142). Saeed (2015, p.119) names activity verbs unbounded processes, and accomplishment and achievement verbs bounded processes and point events, respectively.

The accomplishment and achievement verbs are of interest for this study, and Slabakova (2001) concludes that children use past tense marking, which means that they understand the endstate the verbs entail.

Since this study focusses on bilingual children, and there is not much research available about this subject, the following section will include adult second language acquisition studies.

2.3 Adult L2

Most research about bilinguals is about adults. The acquisition of a language at for example age 20 and age 5 is different. When you are 20, you have already acquired your native language, and therefore you already know more about language, which means that you use different strategies to acquire language. As Bley-Vroman (1989) point out: ‘Foreign language learning differs in degree of success, in the character and uniformity of the resultant systems, in its susceptibility to factors such as motivation, and in the previous state of the organism; the learner already has knowledge of one language and a powerful system of general abstract problem-solving skills’ (Bley-Vroman 1989, p.3-4). He also points out that there is a variation in the degree of success, and there is a change that you will have an accent that can be

perceived as ‘foreign’. According to Bley-Vroman the problem with second language

acquisition is that some people will acquire a native-like level, but that there is a wide range

of variation in the proficiency levels (Bley-Vroman 1989).

Second language acquisition at a later age is no guarantee for success. Cenoz (2003) points out that older learners achieve a higher proficiency faster than the younger learners, but that a lack of input can cause a lower proficiency level than the younger learners, who receive in general more input. The success rate of acquisition is often named together with the age of acquisition, which will be discussed in the next section.

2.4 Age of Acquisition

Johnson & Newport (1989), as well as García Mayo & García Lecumberri (2003) point out that age does matter in second language acquisition. In general, they say ‘the younger the better’. It is also noted that older children acquire a language faster than younger children, but that adults gain a bigger lexicon in a short amount of time (Johnson & Newport 1989). Bley- Vroman (1989) points at the same: ‘Studies which attempt to correlate age of acquisition with degree of ultimate proficiency show that ‘younger is better’ (Bley-Vroman 1989, p.9).

Johnson & Newport substantiate this statement with a study about the acquisition of English as a second language by native Korean and Chinese speakers. The results showed a strong advantage for the earlier arrivals. The results were linear up to puberty. After that, the scores were low with a lot of variance. There was no effect found for the amount of experience with English, American identification, motivation or self-consciousness. They claim that their results support the theory that the critical period for language acquisition also holds for second language acquisition (Johnson & Newport, 1989).

There is a debate going on about the Critical Period theory (CP); the theory that after a certain age, this differs between the age of 4 and puberty, it is very hard to impossible to acquire language, since the ability to acquire language is biologically linked to age.

Researchers disagree about the age range and the existence of the CP, and if it also holds for second language acquisition. Birdsong & Molis (2001) redid the research from Johnson &

Newport (1989). They used the same test materials and protocol, but only their speakers were

native Spanish. They found that the speakers that started later with learning English, could arrive at a high proficiency level, and some were even nativelike. This rebuts the CP for second language acquisition.

Unsworth (2016) tested children who started between age 1 and 3 with learning their second language, and compared them with children who started between age 4 and 7. She found that there was no significant difference in the acquisition of verb morphology, verb placement, vocabulary and object scrambling. The amount of exposure was the only factor significantly related to children’s scores, so this research proves that exposure is more important than age of acquisition.

Meisel (2009) tested French-German bilingual children. He found that the children started with learning both languages before age 3;6 were classified as 2 L1 speakers (2 first languages), with one exception. The children who started learning French after the age of 3;7 were classified as L2 speakers, with also one exception. This study therefore supports the idea of a critical period.

Another study by Perani et al. (1998) tested a group of bilingual Italian-English speakers, who learned English after the age of 10, and gained a high proficiency level. They compared them with a group of Spanish-Catalan bilinguals, who learned both languages before the age of 4. They looked at which parts of the brain were active during multiple language tasks. They found that: ‘At least for pairs of L1 and L2 languages that are fairly close, attained proficiency is more important than age of acquisition as a determinant of the cortical representation of L2’ (Perani et al. 1998, p.1841). Hence, this study rejects the CP hypothesis.

The children in this study started early with acquiring English as a second language.

Some even started from birth on. Whether or not the CP is true for second language

acquisition, there are multiple studies, like the study from Johnson & Newport (1989) that

prove that an early start is an advantage, but other studies like Unsworth (2016) show that exposure is more important.

This study is not only about bilinguals, but mainly about verb acquisition and knowing that certain verbs entail an endstate. The next studies all tested children’s understanding of the endstate of verbs.

2.5 Manner Bias Hypothesis

Gentner (1978) introduced the Manner Bias hypothesis. Which states that children first learn what an action looks like, and later on what it is supposed to do. Gentner states it as:

‘Children appear to learn the action components of the mixing verbs before they learn the change of state components’ (1978, p.994). She tested this by using the verbs mix and stir in an experiment. Her participants were children between the ages of 5 and 9, and an adult control group. She chose these verbs because:

“the verb mix is strongly functional; mix specifies a change of state (an increase in homogeneity) and is unspecific as to the actions that result in this change. In contrast, stir specifies a certain kind of hand-and-spoon motion (rotary medium slow) and leaves the function unspecified”. (Gentner 1978, p.993)

She found that the youngest children, age 5-7, did not distinguish between mixing a

homogeneous substance or a substance consisting of multiple things. Therefore, she proposed

the Manner Bias hypothesis, since children are driven by what an action looks like (stir),

rather than what an action is supposed to do (mix). This could not be due to the children not

understanding the verb mix, since Gentner’s earlier experiments show the understanding of

the verb at an early age. Instead, the children had the tendency to see it as an action verb. If

the Manner Bias hypothesis is true, children would neglect the endstate often, since they focus

on what an action looks like, and not what it is supposed to do.

2.6 Locative Verbs

Bowerman (1982) noticed her children’s behaviour with locative verbs. They made mistakes between the age of 3 and 7, but before and after that almost no mistakes occurred. The mistakes they made were:

(3)* I filled water into the glass (4)* I poured the glass with water

She found that the children overextended the content-object form to verbs that normally have a container argument as direct object. Taking Bowerman’s observations as starting point, Gropen, Pinker, Hollander & Goldberg (1991) studied the syntax and semantics in the acquisition of locative verbs. They explained that in the case of a content direct object the referent is undergoing a change of location (ex. pour and dump), and in the case of a container direct object, the referent is undergoing a change of state (ex. fill and empty) (Gropen et al.

1991, p.115). They propose object affectedness linking rule, which means that an argument is encoded as the direct object of a verb, if its referent is affected in a specific way in the event described by the verb (Gropen et al. 1991, p.118). They argue that children might think that fill specifies the particular manner in which a substances changes location instead of a particular change of state that a container undergoes (Gropen et al. 1991, p.122).

In their first experiment, 48 children (age 2;6-5;11) divided over three different age

groups and 16 adults participated. They used 24 test items, each consisting of two pairs of

pictures. The first panel showed the manner in which a change of state came about, and the

second panel showed the endstate of a container as a result of the action. The verbs tested

were pour, fill, dump, empty, stuff and splash, of which stuff and splash are not further

discussed in their paper. One example shows a woman pouring water into a cup, which is the

manner, and the second picture shows an empty glass with water spilled next to it, which is an

endstate. Another example shows a woman placing a glass under a dripping tap, which is a

manner, and the second picture showing a filled glass, which is an endstate. Participants had to select which picture pair, pouring-spilled or dripping-filled, matched the particular verb. As expected, the adults made almost no mistakes. The children, on the other hand, had problems with the verb fill. The data suggests that “a child may be more likely to think that the meaning of fill involves pouring, than that it involves something being made full; likewise, that empty means dumping, rather than that it involves something being made empty” (Gropen et al.

1991, p.131). They give two possible explanations for this. The first is Gentner’s (1978) manner bias hypothesis. The second explanation is that the participants had to choose between pouring or filling, but this was not available as one pair in the pictures. Perhaps some children thought that filling must involve an endstate ánd a pouring manner. When they were forced to choose, they chose the endstate over the manner.

They conducted a second experiment with 48 children (age 3;5-8;9) in three different

age groups and 16 adults. This time, participants were shown twelve pairs of pictures. Each

pair had three panels, and four different combinations, namely pouring-spilling, dripping-

filling, pouring-filling, and dripping-spilling. This experiment was mainly used to study

children’s understanding of the verb fill. They had to answer questions about sensitivity

towards manners, sensitivity to endstates, and sensitivity towards manners when contrasted

against endstates. All the picture pairs with three panels, contained the beginning, the middle,

and the endstate of the action. An example for pouring-spilling is the first panel with a woman

pouring water next to a glass, where a puddle of water appears and the glass remains empty,

the second panel shows her still pouring water next to the empty glass, and in the third panel

the woman is gone, leaving behind a large puddle next to the empty glass. An example of

dripping-filling is a first panel where a woman turns on a tap, so that water drips into a half-

filled glass, the second panel shows the woman and the tap still dripping water in the glass,

which is now filled for ¾, and the third panel shows that the woman is gone and that the glass

is full. The results show that all groups had no problems with choosing a full glass over an empty one, and with choosing a full glass over a ¾ full glass, and with choosing a ¾ full glass over an empty one. The performance increased with age, and the difference between the children and adults is significant. The adults treated pour and dump as manner verbs, and not as change of state verbs, and fill and empty as change of state verbs, and not as manner verbs.

The children had problems with interpreting the verb fill. Half of the children choose a picture showing a pouring manner without an endstate. For empty they did not find a significant number in the children interpretation of the verb as change of state verb. Therefore, Gropen et al. (1991) conclude that children have problems with differentiating a state change verb and a location verb. If one glass was full after the dripping action, and the other glass was empty after the pouring action, the children showed a preference for the triplet with pouring-spilled, even though the glass was empty. This can be linked to Gentner’s (1978) hypothesis that children have a bias toward the motion component and not to change-of-state component in verbs. Hence, children chose the picture with pour over the picture with drip, since pour includes a manner. Wittek (2002) proposed in her dissertation another hypothesis about the interpretation of endstates.

2.7 Weak Endstate Hypothesis

Wittek (2002) proposed the Transparent Endstate hypothesis. It states that children learning a language that typically expresses caused state changes with complex predicates (English or German), such as cover up, should pay more attention to the endstate when interpreting complex predicates that display the endstate on the linguistic surface, than for verbs that do not (Wittek 2002, p.35). This predicts that the endstate for children is easier to understand for the verb blow out than for the verb destroy. Wittek conducted multiple experiments to test the hypothesis. The first experiment compared verbs with a particle that indicates an endstate (ex.

wachmachen, awake-make) and verbs that do not (ex. wecken, wake). She tested children (age

4;1 – 5;9) and adults. They had to watch movies with two conditions: a change condition and a no-change condition, and in the movie they heard one verb, either the transparent verb with a particle, or the opaque endstate verb. An example of the questions asked is:

(5) Stimmt das? Hat das Mädchen den Mann wachgemacht (or: geweckt)?

Is right that has the girl the man awake-made or: woken

'Is that right? Did the girl make the man awake (or: wake the man)?' (Wittek 2002, p.49)

She found that more errors were made with the transparent endstate verbs, than with the opaque endstate verbs, which does not support the hypothesis. In 25% of the cases children accepted the zero result condition. For the transparent particle verbs, it was 32% acceptance, and for the opaque verbs 17% acceptance. Therefore, she changed the experiment and conducted a second one. This time, she used verb-particle constructions which express state changes, comparing a no-change and a change condition. The children tested were age 4;10- 5;10. Examples of the verbs are wachklingeln, ‘awake-ring’, and auspusten, ‘out-blow’. An example of the questions asked is:

(6) Hat das Mädchen den Mann wachgemacht?

Has the girl the man awake-made 'Did the girl make the man awake? '

She found that the no-change condition was a problem: only 68% of the answers were correct rejections. Therefore, neither this experiment, nor the first supported the hypothesis. In the third and last experiment, she tested children age 4;4-5;10 on verbs in a different syntactic context, such as that the particles appeared sentence finally:

(7) Ich glaube, das Mädchen macht gleich einen Mann wach.

I think the girl makes soon a man awake

'I think the girl is going to make a man awake.' (Wittek 2002, p.69)

She expected the verbs with a particle in sentence-final position to be easier for children than the opaque verbs. But this was not found, since only 63% of the answers in the no-change condition were correct no-answers. 75% of these correct answers included a transparent verb, and 50% of the answers an opaque verb. However, she did not give up the Transparent Endstate hypothesis, since in experiment 3 the children did significantly better on verbs with a transparent particle than on opaque verbs. Instead, she formulated the weak endstate hypothesis: ‘Children interpret change-of-state verbs as if a specific endstate might well come about, but need not’ (Wittek 2002, p.183).

The Transparent Endstate hypothesis can be connected to the Transparency Principle, since both hypotheses argue that something is understood better when its use is overt and transparent. Wittek (2002) used telic sentences for the Weak Endstate hypothesis, just like Gropen et al. (1991) did when testing the Manner Bias hypothesis.

2.8 Transparency Principle

Van Hout (1998) investigated the role of direct objects and particles in learning telicity in Dutch and English. In this study, she tested the Transparency principle, which states that acquiring a certain construction is easier when it is overt and unambiguous (one-to-one), than when it is covert and/or conflated (many-to-one). She tested the acquisition of telicity for Dutch and English. This is of certain interest for this thesis, since the understanding of telicity is necessary for the understanding of the endstate of a verb.

She tested 3-,4- and 5-year-olds and adults. She used four types of sentences:

(8) Intransitive: Did the red/white mouse eat?

(9) Bare transitive: Did the red/white mouse eat cheese?

(10) His transitive: Did the red/white mouse eat his cheese?

(11) Particle verb: Did the red/white mouse eat up his cheese?

(Van Hout 1998, p.6)

The categories contain both atelic (8) and (9) and necessary telic sentences (11). The sentences such as (10) include possessive pronouns, which may also bind the event as a bounded object. The test consisted of stories with flash cards. On the cards were mice that finished eating all the cheese, the telic situation, or mice that ate some of the cheese, but also left something, the atelic situation. The questions asked used the perfective tense form, to see if the participants understood telicity.

Her results show that the adults in Dutch gave a telic reading in 78% of the cases, while the English adults allowed atelic readings in many of the cases, and only 25% restricted to telic. In Dutch, as well as in English, none of the child age groups distinguished the

sentences such as (10) from the atelic sentences such as (8) and (9). The older the children were, the less mistakes they made with the particle verb, which indicates an endstate.

Nevertheless, the three-year-olds choose the atelic situation for the pictures with the endstate in 50% of the times in Dutch, and 34% of the times for English. The 5-year-olds in Dutch only made 10% mistakes, and for English only 9%. In the other categories, all age groups made more mistakes. Therefore, Van Hout (1998) concludes that the data supports the Transparency Principle, since particle verbs are unambiguous and one-to-one notions, and therefore easier to learn than many-to-one notions. The particles are understood as telicity markers, even though the participants were very young. This study connects with the study from van Hout (2005) where she tested the acquisition of aspect in Polish.

2.9 Acquisition of Aspect

Van Hout (2005) tested Polish children on the acquisition of perfective and imperfective

aspect. 56 children participated in the age range 2;0 to 4;11, as well as 20 adults. The

participants were shown pictures with an accompanying story. They had to choose between

two pictures. The picture pairs showed contrasting endstates in three conditions: complete and

ongoing, complete and incomplete, and ongoing and incomplete. Her results show a U-shape

curve; the 2-year-olds and 4-year-olds differentiated perfective and imperfective aspect, while the 3-year-olds did have a preference for the completed situation. Furthermore, looking specifically at children’s interpretation of perfective sentences, the 2-year-olds choose the ongoing situation for perfective aspect in 22% of the time, the 3-year-olds in 29%, and the 4- year-olds in 5%. In condition 2, where they had to choose between complete and incomplete, the 2-year-olds chose an ongoing situation for the perfective aspect in 22% of the times, the 3- year-olds in 16% of the times, and the 4-year-olds in 2% of the times. In condition 3,

contrasting an ongoing and an incomplete situation, 2-year-olds chose the ongoing form for the perfective aspect in 33% of the time, the 3-year-olds in 39% of the time, and the 4-year- olds in 18% of the times. These results show that children make mistakes with interpreting telic-perfective sentences, occasionally choosing a situation without an endstate.

Van Hout (2008) did the same study, but now with three languages: Polish, Dutch and Italian. The children had to choose between a complete and an ongoing situation. She used the present perfect for Dutch and Italian and the perfective past for Polish. The Dutch (63%) and Polish (80%) children chose the complete situation, while the Italian children chose the complete situation in 54% of the times. This means that the Italian children have not yet acquired the perfectives of the present perfect. She explains this through the Morphological Salience Hypothesis: the way an aspect is encoded in a language has influence on when it is learned. Since the morphological encoding is not overt in Italian, where aspect encodes the present perfect, it takes more time for the Italian children to learn this.

Van Hout stated that the generalisation was that children acquire event culmination

earlier when the verbal predicate overtly and explicitly encodes a result state (in press, p.3,

1998), which can be seen in van Hout (2008) where the children made less mistakes in the

overt particle verbs. Van Hout (in press) compares studies that looked at culmination by

looking among other things at the interpretation of particle verbs and telic transitive

sentences. She found that there is acceptance of non-culmination, sometimes even 75%. Some acceptance is found for all the studies she looked at, and for different languages and ages. She also finds that less mistakes are made when there is an overt notion used, such as a particle verb or the clitic se. More mistakes are made in the transitive conditions with a

morphologically simple verb (Van Hout, in press).

Van Hout (in press) gives three arguments against the Manner Bias hypothesis. First, the conclusions are based on only two verbs. Second, the Manner Bias predicts that children do not know that verbs include an endstate, but the other studies discussed in van Hout (in press) show that children do know that verbs have an endstate, even though they accept non- culmination to some extent. Third, there is a learnability problem, which means that it is not clear what triggers children to let go of the Manner Bias (van Hout, in press).

Wittek (2002) tested particle verbs and monomorphemic verbs, and her findings also support the Transparency principle. The children performed better on the particle verbs, and Wittek concluded therefore that this goes against the Manner Bias. Instead, she proposes the Weak Endstate hypothesis. She refers to other adult languages that allow non-culmination for monomorphemic change-of-state verbs, which are Mandarin, Japanese, Hindi and Tamil (Wittek 2002, in van Hout, in press). Van Hout (in press) points at the different studies that name agent or cause as the triggers allowance of non-culmination by change-of-state verbs, and points at the Agent Control Hypothesis (Demirdache & Martin, 2015) that looks at this problem.

2.10 Agent Control Hypothesis

The agent-control hypothesis is formulated by Demirdache & Martin (2015). They found that

many languages allow non-culmination readings for perfective accomplishments. This means

that you can say that you destroyed something without it being fully destroyed. But this is

only possible for agents in the subject position, and not for causes. They illustrate this with examples from Mandarin:

(12) Zhangsan guan-le na shan men, dan men hai kai-zhe Zhangsan close-PERF that CLF door, but door still open-DUR

‘Zhangsan closed the door, but the door was still open.’

This can happen with an agent as in (12), but not with a cause as in (13):

(13)# Yi zhen da feng guan-le na shan men, dan men hai kai-zhe One CLF strong wind close-PERF that CL door, but door still open-DUR Intended: ‘The strong wind closed the door, but the door was still open.’

(Liu, in prep.)

They also observed the same phenomenon in English. When there is an agent subject, there is no necessary endstate, but if the subject is a cause, then the endstate is required:

(14) Lipson’s textbook taught me Russian, #but I did not learn anything (15) Ivan taught me Russian, but I did not learn anything.

(Demirdache & Martin 2013, p.32)

Therefore, they formulated the Agent Control hypothesis. The hypothesis that adults in all languages allow perfective accomplishments with an agent in the sentences to have a non- culminating (NC) reading, but that this is not possible with an inanimate cause. They split this hypothesis in two parts: a weak and a strong version. The weak version means that: ‘zero result NC construals only require the predicate external argument to be associated with

‘agenthood’ properties’ (Demirdache & Martin 2013, p.31-32). The strong version states that:

‘zero result and partial result NC construals require the predicate’s external argument to be

associated with ‘agenthood’ properties’ (p.31-32). They found that the weak version holds for

every language family they tested, Romance, Germanic, Salish and Mandarin. This means

that at least zero result non-culmination construals require agent control (Demirdache &

Martin 2015). The strong version is found to be true for Salish languages, whereas French and German only allow the weak version. For Mandarin, only the weak version is found to be true so far (Demirdache & Martin, 2015).

The Agent Control hypothesis is based on adult language, and therefore van Hout et al.

(2017) tested if the ACH also holds for child language. All the previous mentioned studies from Wittek (2002), Gropen et al. (1991) and van Hout (1998) had in common that they all had agent subjects, or animate subjects. None of the subject were a cause. With these agent subjects, the children accepted non-culmination to a certain extend. Therefore, it is interesting to see if the ACH also holds for child language, and if it holds, this might be the explanation why children allow non-culmination in the previous experiments.

2.11 Agent Control hypothesis in Child Language

Van Hout et al. (2017) used the same culmination test as this research, developed by Strangmann (2015). They found that the ACH is not supported for child language. The 5- year-olds who speak Dutch, English, Basque and Spanish all showed adult-like behaviour, which means that the children allow almost no acceptance of non-culmination. Van Hout et al. (2017) also looked at Mandarin, which allows non-culmination in monomorphemic verbs, as can be seen in figure 1.

Figure 1 Acceptence of zero result situations with agent (black) and cause (grey) for Mandarin monomorphemic verbs (Van Hout et al. 2017, p.7).

The monomorphemic verbs in Mandarin show that there is more acceptance of non- culmination with agent subjects than with cause subjects, which supports the ACH. This holds for both the children and the adults, as the black bars are higher for all three age categories.

There was an effect found of subject type; the acceptance of non-culmination was higher for sentences with an agent than for sentences with a cause.

The sentences in Mandarin where the second verb used encoded the result state showed no allowance for non-culmination for the 5-year-olds and the adults, but the 3-year- old allowed there to be some non-culmination.

Since the children tested in this study are all bilingual, literature about language acquisition, and especially second language acquisition will be provided here. Unfortunately, studies about language acquisition with bilingual children are rare. Therefore, the studies about English as Foreign Language in the Netherlands will also be included, to provide information about the success rate of second language acquisition.

2.12 EFL in Dutch Primary Schools

English as Foreign Language (EFL) in the Netherlands is studied by Kees de Bot, Sharon Unsworth and others. In 2003, Goorhuis-Brouwer and de Bot started testing different cohorts of children. The children were taught 3 hours per week in English. They used the Reynell test as testing material. The children reached a proficiency level of 2;5 years after one year of teaching. Their Dutch proficiency had not been influenced. (Goorhuis-Brouwer & de Bot 2010). The main conclusion is that there is no negative effect of EFL on the Dutch language acquisition, not even for the children who do not have Dutch as L1. The children do acquire some English skills, but the amount of time spent on English does not have negative

consequences for the Dutch language acquisition. (Goorhuis-Brouwer & de Bot 2010, p.299-

300)

To see if an increase in hours has an influence on the English proficiency level, Groningen University and the University of Utrecht ran a project to study this. The project called the FLiPP project; Foreign Languages in Primary Schools. It focussed on two main questions: “How do the English language skills of the children develop over time? What is the relationship between the development of the children’s English language skills and the

development of Dutch as a first and a second language?” (de Bot 2014, p.413).

The method used was the Peabody Picture Vocabulary test (Dunn & Dunn 1981) and the Test of Reception of Grammer (Bishop 2003). To test the proficiency level of Dutch, they used a receptive vocabulary test. The findings for the FLiPP project were that “children in early EFL programmes scored significantly higher than children who were not in such programmes, and when analysed in groups, after controlling for the effect of teachers’

language proficiency, those with more than 60 minutes of weekly classroom exposure scored on average significantly higher than those with 60 minutes or less”. (Unsworth et al. 2014, p.13). “Children who were taught by a native speaker or a teacher with a high level of proficiency scored significantly higher than those who were taught by a teacher with a lower proficiency level” (Unsworth et al. 2014, p.19).

The studies about EFL by Dutch children matter for this thesis, since they show that it does matter how much input the children receive, and if this input is given by a native speaker or the regular teacher. These studies look at Dutch children, but the way they measure the different proficiency levels is useful for this study.

Since the children in this study all have different language background, there was also

a test needed to determine the proficiency level of the children. The test chosen here was the

GAPS test, but there are multiple variants which can be used. An overview of tests used in

previous studies will be given here.

2.13 Different Methods to Test Child L2 Proficiency Level

Umbel et al. (1992) tested the receptive vocabulary of Hispanic children in Miami. They tested both the Spanish and English knowledge. As test material, they used the Peabody Picture Vocabulary Test – Revised (PPVT-R) created by Dunn & Dunn (1981). This test has 175 pages which all contain 4 pictures. The child needs to point to the picture that

corresponds to the word the examiner speaks. Unsworth et al. (2014) also used the PPVT to test vocabulary skills in their research about EFL in the Netherlands. They also used the Test for Reception of Grammar (TROG), to test the children’s receptive grammar skills. The TROG tests the understanding of 20 constructions. The test focusses on the understanding of different sentences, for example; point at the picture were the lady is running, were there is also an option of a lady who stands.

Goorhuis-Brouwer and de Bot (2010) used the Reynell test for measuring language proficiency in their EFL research. This test consists of assignments with toys. The child has to point at objects, point at pictures, manipulate objects, or answer simple ‘yes-no’-questions.

An example includes a situation with a monkey-toy, where the monkey points at pictures and asks the child questions using reflexive and non-reflexive pronouns. The test has standard scores available for every age category, which means that a control group is not necessary anymore. It is available in Dutch for age 1;2 to 6;3 and in English for age 1;9 to 7;3.

Unsworth (2008) used three Dutch tests to determine the language proficiency level of the L2

Dutch. Only one of those test is suitable for children, the other two are for adolescents and

adults. The test for children was the Taaltoets Alle Kinderen, (TAK) or ‘Language Test All

Children’. This test focusses on sounds and pronunciation, passive vocabulary knowledge,

word description, sentence understanding, making a sentence or a word, the understanding of

a text and a story retelling.

Unsworth (2015) tested English children who learned Dutch as a second language.

The first test she used consisted of 24 test items with each two pictures about the same verb, where the children had to complete the sentence using the right verb form, namely the present simple tense. The second task tested the knowledge of L2. The children had to explain to a puppet what you can do with the object pictured. An example is: wat kun je ermee doen?

‘what can you do with it? Aardbeien.. kun je eten. Strawberries… can you eat (lit.)’

(Unsworth 2015, p.618).

2.14 GAPS

The GAPS is a Grammar and Phonology Screening test, which is a quick scan designed for 3;4 up to 6;8 years olds to see if a child might suffer from Specific Language Impairment (SLI). The test was developed by Gardner, Froud, McClelland and van der Lely (2006). The first part of the GAPS consists of 11 test sentences and 3 fillers. The child has to repeat the sentences exactly after the researcher. The sentences contain inflections, morphology and syntax, and test the children’s ability to form words and sentences. The second part consists of 8 complicated non-words, which test the ability to use the sound system and create words. If the child incorrectly repeats a part or a word from the sentences, or a part of the non-words, it is marked as wrong. Gardner et al. (2006) tested 668 children in the UK to make a standard score. This test has not been used before in research to determine children’s proficiency level.

This study uses it, since it is a short test, it takes about 5-10 minutes, which was necessary.

2.15 Hypothesis and Research Questions

This study looks at the acquisition of verbs that entail event culmination. According to the

Agent Control hypothesis (Demirdache & Martin, 2015), adults allow non-culmination when

there is an agent involved, but not when there is in inanimate cause involved. Since the

previous studies from Wittek (2002), Gropen et al. (1991) and van Hout (1998) show

acceptance of non-culmination, it is interesting to see if the ACH also holds for child

language. The test used is the same as van Hout et al. (2017) used to test non-culmination in Dutch, English, Basque, Spanish, and Mandarin, as was developed and used by Strangmann (2015). The study from van Hout et al. (2017) found that the ACH indeed holds for Mandarin.

Both children and adults allowed monomorphemic words with an endstate to not have

culmination as is shown in example (8) and (9). This supports the ACH. The other languages on the other hand did not support the ACH. This study will again test the ACH for children, but with bilingual children who learn and are taught in English in primary school. There is no previous study about the acceptance of non-culmination with telic-perfective sentences with bilingual children.

The research question is: Do bilingual children allow non-culmination in telic- perfective sentences? The subquestions are: Is there a difference in the acceptance of non- culmination by bilingual or monolingual children from the van Hout et al. (2017) study? Is there a difference in the acceptance of non-culmination if the verbs used are particle verbs, which means that they should be more transparent? Is there a difference in the acceptance of non-culmination for the children who only speak English at school and the children who also speak English at home? Does the proficiency level of the children influence the degree of acceptance of non-culmination? Is the GAPS test a sufficient test to determine the proficiency level?

3. Method

This thesis used two different test. The first test is the event culmination task, which

was used and developed by Strangmann (2015). The second test was to determine the

proficiency level of the children, and this test is called Grammar and Phonology Screening

(GAPS), and is developed by Gardner et al. (2006).

3.1 Participants

In total, a group of 44 people participated in this study. 34 children between age 4;8 and 6;10, who all attend the Groningse Schoolvereniging in Groningen and are taught in English at the international department. At home they speak another language with their parents, and sometimes also English at home. All children have a different language background: there were children who speak Dutch and German, but also Romance languages as French, Italian, Spanish, and Portugese, Slavic languages as Polish, Czech, Bulgarian and Russian, but also Greek, Kazakh, Arabic, Chinese, Japanese, Malaysian, Lithuanian, Hindi, and Urdu. All children have had at least one year of English instruction.

Besides the children, 10 adults with English as their mother tongue participated as a control group.

Non-culmination task

GAPS Age range Gender

Children N = 34 (5;8) N = 34 (5;8) 4;8 – 6;10 23 female Adults N = 10 (25;2) Not applicable 19;5-56,2 2 female

Table 1. Overview of number of participants for the culmination task and the GAPS, age, age range and gender for the groups

3.2 Materials and design

For this thesis, two tests were conducted. The first one was the same as Strangmann (2015) developed for her thesis. The test consisted of 28 short movies and 6 test movies. The movies had no spoken language. The test took about 15 to 20 minutes. All the tests were audio recorded, except for one child where something went wrong with the recording. It was decided to include the child and use only the notes the researcher made.

Seven verbs were used in the test sentences: blow out, break, cover up, destroy, open,

shut and take off. The design of the movies has two types of subjects; namely agent and

cause, and two types of endstates, namely full result and zero result. This results in four different conditions:

Condition 1. Agent - full result Condition 2. Agent - zero-result Condition 3. Cause – full result Condition 4. Cause – zero result

The agent was portrayed by a pirate or a clown. The cause was portrayed by an explosion or wind, which was made visible through a fire or waving flags in the background and sounds. In the full result state, agent or cause in the movie finished the task. In the zero result state, the agent or cause in the movie tried to finish the task, but failed, which led to zero result.

The verbs chosen are particle verbs and some simple verbs. This was done to see if the sort verb influenced the interpretation of the sentences. All sentences are telic-perfective, which entails culmination.

The children were introduced to a hand-puppet called Mister Elephant, who had big ears, but small eyes. Therefore, he could not see the movies well, and the children had to help him understand what happened in the movies. The puppet was used to make sure that the children would dare to say no to the researcher.

3.3 Procedure

The children were tested between April 20 and May 11, 2017. Because of the lack of a

separate room, the children were tested in their school’s corridor or kitchen. The corridor was

sometimes noisy, but the only place we had. This resulted in bad audio recordings, which is

the reason that some of the answers were coded as ‘inaudible’. The adults were tested

between May 10 and June 7 2017. They were tested at various locations in Groningen.

The culmination task included 28 items. All items had one yes/no filler question, and the actual question concerning the verb. There were question object fillers and question object appearance fillers. The question object concerned the name of the object, and the question object appearances the colours of the objects shown. These questions were easy and used to see if the child payed attention to the movie.

Six practice items were shown first. These items included one movie with each agent, and two movies with each cause to make the children familiar with the explosion and the wind. During these test items, the children were corrected if they gave the wrong answer. The questions about the verbs included a different verb, to see if the children were familiar with the verbs. The filler questions were about the objects shown in the movies.

After the filler question, the question concerning the verb was asked. This was done in a transitive construction in simple past tense: ‘Did the agent/cause verb the object?’ In figure 2 and 3, the question was: ‘Did the clown blow out the candle?’, where figure 2 has a full results and figure 3 a zero result. The question in figure 4 and 5 was: ‘Did the wind blow out the candle?’, where figure 4 has a full result, and figure 5 a zero result. If the child answered no, the question ‘What happened’ was asked to elicit an explanation. Figures 2 to 5 show examples of the movies. As is shown in the pictures, the movie started with a wide shot of the situation, then a close-up to make the intention clearer (which is not the case in figure 2), and then another wide shot with the resulting state. For eleven movies, there is no close-up

included, since this did not add to the clarity of the movies. These movies included the movies

with the explosion, where no close-up was needed. All the other movies are included in

appendix B.

Figure 2. Agent in full result situation (blow-out).

Figure 3. Agent in zero result situation (blow-out).

Figure 4. Cause in full result situation (blow-out).

Figure 5. Cause in zero result situation (blow-out).

Since all children had different language backgrounds and different levels of

proficiency, there needed to be a test to determine their proficiency level, in order to compare them with each other. Children first did the event culmination test and then the GAPS. The older children did it all in one session, except for one child. The younger children did it in two sessions, and there was one child who did the GAPS in two times and had three sessions in total.

The GAPS consists of two parts. The first part is reading a story where the child has to

repeat 2 practice sentences, 11 test sentences and 3 fillers for an alien doll named Bik, who

can only hear children’s voices. The sentences include all kinds of inflections and syntactic

elements which the child also needs to repeat, otherwise the sentence is marked as wrong. The

second part consists of the repetition of non-words, with first 2 practice words, and 8 more

complicated non-words. The prosodic and metrical complexity in increased, for example drepa is relatively easy, but padrep is hard (the bold parts are stressed) (Van der Lely et al, 2011, p.2). Examples of the non-words are klest, padrepper bademper and difrimple. Gardner et al. (2006) tested 668 children from across the UK to make a standard score. All scores are percentile scores and tells you how many children are likely to score higher or lower. If the child gets a low score, depending on their age, this might be an indicator for SLI. For this thesis, it was used as a quick measurement for the English proficiency level of the children.

Unfortunately, there are no standard scores available for bilingual children.

Why did this study choose the GAPS, instead of one of the other methods? The first idea was to use the Reynell test, but this test takes quite long. If the children had to do this test and the culmination task, it would have been more than one hour per child, which is too much for the children, and also for the teachers whose lessons are being interrupted. The studies mentioned above focus mainly on proficiency level in different circumstances, while this study focusses on event culmination, but needs a proficiency level test. The GAPS takes around 5 to 10 minutes per child, so this is a quick screen. The length was the main reason for choosing the GAPS. It was not used before on bilingual children, but it works for them as well. Only the percentile scores based on the British children are not available, but the raw scores give a good indication about the language proficiency.

Furthermore, the teachers were asked for a proficiency judgement as a means to group the children based on their proficiency level. They divided the children into five categories:

low, low/middle/, middle, middle/high, and high. These intuitive judgements and the GAPS

are used in the results section to provide measurements of the English proficiency level of the

children. Their proficiency level from their native language(s) was not taken into account.

3.4 Scoring justifications

Before the results were calculated, the answers from the event culmination test were coded.

The answers from category 4 and 5 were left out. This was done because item trouble and different description show that the participant did not understand the meaning of the movie, and therefore gave a different answer than expected. The given answers were coded in six categories:

1. Reference to the agent’s intentions or efforts: ex. ‘The clown did not destroy the plate because it was too hard’.

2. Reference to the cause: ex. ‘It was trying to open’.

3. Reference to the result state: ex. ‘It moved’, or ‘it could not because there was something in the front’.

4. Item trouble. This category was if there was a misunderstanding about the intention of the video. Ex. ‘there was an invisible string tight to it, they just waved the flag and then the other person with the invisible string shut the door’. Here, the child saw the invisible string for the wind items and therefore she did not believe that the wind did the actions. The answers coded with item trouble are excluded from further research.

5. Different description. The participant accepted the event that took place, but gave a different explanation about the event. Ex. ‘water can only take the candle off’, where the child did not accept the wind blowing to be a possibility.

6. Other. This category is made for the items that do not fall under the other categories. It was decided to include these items in the research. Ex. ‘the hammer didn't break the plate because it is not stable’, ‘because there was a hard and there was a cold’, ‘it made noise’.

A list of all the answers from item trouble, different description and other is included in

appendix A.

4. Results

Before the results were calculated, two children were excluded from further research.

There were filler questions in the beginning to see if the child paid attention or not. As in Strangmans (2015) thesis was decided, the children should not make more than three mistakes. This was the case for none of the children. One child was excluded from further research because she answered in 27 out of 28 times ‘yes’, which makes her an outlier. She answered the practice items correctly, but afterwards she only answered ‘yes’. She gave explanations for her answers, and these will be discussed in section outlier. Another child was excluded from further research, since this child has SLI and is under treatment.

One child has autism, but his language skills are excellent, and he is therefore included in the research. Another child has problems with her sight, but we solved this by putting the laptop with the movies right in front of her, so she could see what was happening. In total, 32 children are included in this research.

Bilingual from birth Early child L2

Participants (N=32) 8 24

Table 2. Number of participant that speak English at home

In total, 8 children speak English at home, and for 24 children English is an early child L2. The children who speak English at home also speak another language at home. For

example, they speak English with their father and Dutch with their mother. The other 24 children started around the age of 4, when they first attended Kindergarten.

There was one item from the filler questions that was also left out from further

analysis. The question, ‘is the candle purple’, got so many different answers that was decided

to count everything as correct. Even the adults gave multiple answers for this colour. There

was a filler question about an object, where the children made many mistakes. Almost half of

the children answered ‘yes’ to the question ‘is this a glass’, whereas the movie showed a

plate. None of the children made more than three mistakes, including this question, so based on the filler questions no children were left out.

When the children answered ‘no’ there was always a follow-up question asked, namely ‘What happened?’. Unfortunately, this was not asked when the answer was ‘yes’ in a zero result situation. Luckily, some children explained their lines of reasoning anyway. These answers will be discussed in the section 4.6 special pattern.

The last thing that should be mentioned here is that we decided to include a child that is 2 months too old for the GAPS test. The GAPS is designed for children up to 6;8 years old, and this child in 6;10. It was decided to include her, since she only started learning English in the autumn, and therefore her proficiency level was similar as that of the other children that just started.

As can be seen in table 3, most answers referred to the result state. There are only 8 answers that are excluded from further research. The no response or inaudible answers were included in the research, but they were not coded. The items where a child did not say yes or no, even not after multiple encouragements, were excluded from further research. This includes 4 answers.

Item category Number of answers

Reference to agent’s intentions or efforts 124

Reference to the cause 76

Reference to the result state 329

Item trouble 5

Different description 3

Other 5

Table 3. Distribution of justifications across categories (see scoring justifications for explanations of these categories)

Multiple ANOVAs were conducted to get to the results. The first was a two-way ANOVA between subject type (agent or cause), situation type (full or zero), and acceptance.

A two-way ANOVA showed that on average, the children were more likely to say yes when the situation type was full (M= 0,97; SD= 0,175), than when the situation type was zero (M=

0,02;SD=0,141).

There was a main effect of situation type, F (1,884) = 199,578, p < 0.001., but no main effect of subject type, F (1,884) = 0,010, p = .530, and no interaction. There also was no effect of age. This means that both adults and children reject non-culmination situations.

Age group Condition Mean

proportion of yes responses

Number of yes responses (total)

Std. Deviation

Children Agent_full 0.97 216 (223) 0.175

Children Agent_zero 0.01 3 (223) 0.115

Children Cause_full 0.97 214 (221) 0.176

Children Cause_zero 0.03 6 (221) 0.163

Adults Agent_full 1.00 69 (69) 0.00

Adults Agent_zero 0.00 0 (70) 0.00

Adults Cause_full 1.00 70 (70) 0.00

Adults Cause_zero 0.00 0 (70) 0.00

Table 4. Mean percentage of acceptance for the four conditions for each age group

The adults all gave the same answers, so SPSS did not have enough variation to

provide F or p values. Therefore, they are not included in further analyses. Since there was an

age range from 4;8 years old up to 6;10 years old, another two-way ANOVA conducted to see

if there was an effect of age. The children were divided into a group with the 16 youngest

children, named Young, and a group of 16 older children, named Old. Figure 6 shows the distribution of yes answers, divided by the younger and older children.

Figure 6. Means of yes responses in the four different conditions for the older and the younger children.

A two-way ANOVA showed that there was a main effect for situation type F (3,880) = 2635,805, p < 0.001, but no main effect for the interaction of situation type and age F (3,880)

= 1,708, p = 0.164. The older children did not perform significantly better than the younger children.

4.1 English at Home

Eight children use English also at home, with one or both parents. Figure 7 shows that the children who speak at home score slightly higher than the children who only speak English at home, but this effect was not found to be significant. A two-way ANOVA showed only a main effect for situation type F (3,880) = 2001,243, p < 0.001.

0 20 40 60 80 100 120

Agent_Full Agent_Zero Cause_Full Cause_Zero

Mean 'Yes'-respons

Four Conditions

Old Young

Figure 7. Means of yes responses in the four different conditions for the children that speak English at home or only at school

4.2 Judgement Teachers

The teachers were asked for their intuitive judgement about the language proficiency of the children. They divided the children into five categories: high, middle/high, middle,

low/middle, and low.

Judgement teachers about the proficiency level

Number of children

High 15

Middle/high 3

Middle 7

Low/middle 4

Low 4

Table 5. Distribution of children in different proficiency levels according to teachers’ judgement

A two-way ANOVA was conducted with the proficiency levels, the acceptance and situation and subject type. In figure 8, it is clear that the children with a low judgement said more often ‘yes’ in the zero condition, and less often ‘yes’ in the full condition. There was a main effect found for situation type F (1,868) = 126,583, p < 0.001. There was also a main

0 20 40 60 80 100 120

Agent_Full Agent_Zero Cause_Full Cause_Zero

Mean 'yes'-reponse

Four conditions

English at home English at school