Bilingual education: two languages in one classroom? Investigating pupils’ use of Dutch in English Content and Language Integrated Learning (CLIL) classrooms

Bilingual education: two languages in one classroom?

Investigating pupils’ use of Dutch

in English Content and Language Integrated Learning (CLIL)

classrooms

Anne Martens

Thesis

Research Master’s Linguistics

University of Amsterdam

December 16

, 2015

under the supervision of

Dr. Rob Schoonen

and

Prof. Rick de Graaff

third reader

Dr. Ineke Vedder

*

I would like to thank both supervisors for their valuable feedback and support throughout this thesis project. Dr. Rob Schoonen was always willing to discuss the research methodology and progress, and shared his experiences with research in educational settings. The expertise of Prof. Rick de Graaff in the field of bilingual education meant that he had many suggestions in terms of literature and contacts. I am very grateful that Dr. Ineke Vedder was willing to be the third reader of this thesis. I am also indebted to Onno van Wilgenburg (EP-Nuffic), who provided me with a lot of valuable background information on bilingual education in the Netherlands. Last but definitely not least, many thanks to the teachers who made time for interviews and let me attend their classes, and of course to the pupils who participated in the project - their enthusiasm and honest comments on their experiences with bilingual education gave me a lively image of what bilingual education entails.

Abstract

Within the field of second language education research, the role of the first language (L1) in second language (L2) learning has received increasingly more attention. Several researchers (e.g. Cook 2001, García 2009) have challenged the traditional monolingual approach to L2 learning, which stated that the L1 and L2 should be kept separate, and advocate a more hybrid approach. The practice of combining linguistic resources and “engaging in bilingual or multilingual practices” (García 2009:44) is referred to as translanguaging. Recent studies (e.g. Alegría de la Colina and García Mayo 2009, Storch and Aldosari 2010) have directly investigated L1 use among L2 learners, showing that L1 use is higher for learners with a lower L2 proficiency and for metacognitive and vocabulary talk.

The present study investigated how Dutch and English interact in pupils’ classroom interactions in Dutch content and language integrated learning (CLIL) education, with a focus on the languages that are used and switches to the L1 (Dutch). Pupils from year one (about 13 years old) and year three (about 15 years old) of a Dutch bilingual secondary school were audio recorded during English and mathematics classes. For each class, one recording was made during a regular lesson and the other while the pupils executed a collaborative writing task. In addition, pupils participated in an L2 proficiency test, and both pupils and teachers were interviewed about their experiences with L1 use.

Results showed that the main language of communication was English, with first-year pupils using more English than third-years in three of the four settings. This was probably related to motivation: first-years were characterised as being more eager to use the L2. Regular class settings led to a greater use of Dutch than the collaborative writing task settings, possibly due to their more academically demanding concepts and associated more difficult language requirements (Cummins 1980). In terms of language function, L1 use was highest for off-task talk in both years, and vocabulary talk among first-years. Pupils easily switched between the L1 and L2, with switches to Dutch mostly occurring at the beginning of an utterance and coincided with a change in speaker, but not with a change in language function. Switches made at the beginning of an utterance mostly led to a completely Dutch utterances, whereas mid-unit switches were followed more often by switches back to English. Interlocutors mainly replied in English; replies in Dutch were more prevalent when switches were made at the beginning of an utterance.

1 Introduction

Stronger international relations have led to a greater demand for people who are able to speak multiple languages. Over the last decades, this has led to increasingly more interest in language education – from parents and teachers as well as from policy makers. The European Union has made language learning a priority in order to promote mobility and intercultural understanding. The European Commission’s linguistic objective is referred to as ‘MT+2’: ideally, European citizens learn to speak their mother tongue as well as two other languages from a young age (European Commission 2004).

There are several forms in which learning a second language (L21) can take place. For a long time, foreign languages have been part of educational curricula as separate subjects taught in primary and secondary schools. Some schools have gone further and immersed pupils in a foreign language by teaching all of the lessons in a non-native language. Recently, more and more schools in several European countries have adopted an educational model that can be positioned somewhere in the middle, integrating language learning and content learning for a selection of subjects, following an approach called content and language integrated learning (CLIL).

As a country with many international contacts, the Netherlands has valued foreign language skills for a long time. Although it is not an official language, English is omnipresent in everyday Dutch life. Edwards’s (2014) sociolinguistic profile of the Netherlands shows that the use of English is widespread and that English competence is even assumed in the Netherlands, with the language being used in science, business and media – not always accompanied by Dutch translations. Even in the bilingual province of Fryslân, the use of English is more widespread than of the official regional language Frisian (Edelman 2010). Edwards (2014) concludes that, although not officially but definitely functionally, English serves as a second language in the Netherlands.

The importance of foreign languages is reflected in education: all secondary school pupils are taught at least two foreign languages (English and French or German), and English has been an obligatory subject in the last years of Dutch primary school since 1986 (Edelenbos, de Jong & Westhoff 2004; Ytsma 2000). In recent years, a greater emphasis has been placed on productive skills (Edelenbos et al. 2004). This has led to the introduction of CLIL at Dutch secondary schools, combining language acquisition and content learning. Although several studies have investigated CLIL in terms of its L2 outcomes and effects on Dutch and regular subjects, not much attention has been paid to CLIL’s everyday practices. Research on these practices can nevertheless provide valuable information that can be used to further improve bilingual education.

The aim of the current study is to find out more about pupils’ language use in peer interactions in the Dutch bilingual classroom, with a focus on the languages that are used and the functions these languages serve, and on the contexts of pupils’ switches to Dutch. This topic has both theoretical and practical relevance. It is theoretically interesting because of the discussion about the role of the L1 in the L2 classroom. Several researchers (e.g. Cook 2001, García 2009) have challenged the traditional monolingual approach which advocates that use of the L1 should be avoided, and that the two languages should be kept separate. It also meets a more practical need, because professionals in the bilingual education field notice that pupils use Dutch, but the functions the L1 serves are not clear.

1

Before we move on to the actual research project, an overview is provided of relevant literature and research on (language use in) language education, CLIL and multilingual language practices in language learning contexts.

2 Theoretical background

2.1 L2 language education

Instructed second language acquisition is an important social phenomenon that involves an increasing number of people, especially in developed countries (Housen & Pierrard 2005). Whenever the L2 is not a language of the country where it is being used, we talk about a foreign language. Traditionally, foreign or second language education is provided by means of separate language classes, where competence in the target language is the focus of study. Thus the L2 is taught as a subject with an emphasis on explicit language instruction (García 2009). Since acquiring functional knowledge in an L2 requires considerable time investments, policy makers have looked into other ways to teach foreign language skills. An alternative to the rather narrow approach to L2 learning of traditional language education is provided by bilingual education, which combines language and content learning, presenting the L2 in an authentic context that enhances implicit learning (Admiraal, Westhoff & de Bot 2006).

The term bilingual education is a simple term that refers to a complex phenomenon, encompassing a number of different contexts that have one thing in common: education is provided with multiple languages as languages of instruction (García 2009). The L2 in question can be a minority language, a majority language, a dominant language or a foreign language. To stimulate the use and preservation of minority languages, schools in some areas offer (part of) the curriculum in the minority language (e.g. Basque in Basque Country, Catalan in Catalonia or Frisian in the Dutch province Fryslân). For immigrant children with a different L1, the majority language used for their education is an L2. Bilingual education is also associated with an aspect of cultural awareness, building “tolerance towards other linguistic and cultural groups” (García 2009:6). In the Netherlands, European and international orientation (EIO) can be part of both bilingual and regular education programmes (A. Piketh, personal communication, June 2, 2015).

Within bilingual education there are different strategies that can be distinguished; García (2009) makes a rough division between monoglossic and heteroglossic education types. Monoglossic ideologies assume that the acquisition and development of the L2 are separate from those of the L1, and proficiency according to monolingual norms is the goal in bilingual education. Both languages can be used in education, but the use of the languages involved is strictly separated, for instance by using different teachers or classrooms. Examples of monoglossic education are maintenance programmes, which serve pupils with a cultural minority background in further developing their home language proficiency, and immersion programmes, providing education in a (prestigious) L2 that is not the language of the community. By contrast, heteroglossic ideologies are a response to multilingual environments, claiming that bilingualism involves a dynamic cycle with ever-adjusting language practices and “multiple co-existing norms which characterize bilingual speech” (García 2009:117). Just like monoglossic programmes, heteroglossic education dedicates time to multiple languages; the goal, however, is multilingualism that is not based on monolingual norms but instead “reflects the linguistic fluidity present in the discourse of the

twenty-first century” (García 2009:17). Whereas monoglossic ideologies state that all learners have a similar monolingual starting point, heteroglossic theories consider the learners’ diverse backgrounds and experiences with the languages involved, and accordingly support the interaction of languages (García 2009).

Another distinction that can be made is between immersion and submersion schools, with the former using a language that is not the language of the larger society as the language of instruction, and the latter teaching the majority language to pupils with a different L1 (Admiraal et al 2006). It should be noted that categorising a specific bilingual programme into different types is not always straightforward since the types form a continuum rather than a set of strictly separated categories. A typology of different bilingual education types that can be used to compare and develop bilingual programmes was developed by Piketh (2008; 2006).

2.2 Content and language integrated learning (CLIL)

In several European countries, language and content learning are combined in a heteroglossic bilingual education approach called content and language integrated learning (CLIL)2. CLIL was designed as “a pragmatic European solution to a European need” (Marsh 2003:3), a response to the lack of results obtained from traditional language learning approaches. It is a dual-focused programme, with language-specific goals in content classes, usually applied at mainstream schools. Within CLIL programmes, content is taught by using the target language as the medium of instruction, thus supporting both language and subject learning (Dalton-Puffer 2011). The term CLIL does not refer to a specific programme, but it is an umbrella term for any programme that uses an L2 for the teaching of non-linguistic content (Gablasova 2012). In principle, the L2 that is used can be any language but it is usually a prestigious language, with the lingua franca English playing a dominant role (Dalton-Puffer 2011).

Dale, van der Es and Tanner (2011) give an overview of four theories on language acquisition that form a basis for CLIL. Firstly, the input hypothesis (Krashen 1981) states that it is important that language input is meaningful, relevant and realistic, as well as being abundant and challenging in terms of language level. Ideally, learners are presented with language that is slightly higher than their own language level (also referred to as ‘i + 1’). This may cause a discrepancy between understanding and production, but through language use, learners can move from one level to the other. Secondly, intake theories, which combine input with understanding (Robinson 2003), claim that if learners come across a language frequently and if they have time to process it, they will learn the language. Thirdly, interactionist theories stress the importance of meaningful and modified interaction – adjusted to the proficiency level of the learner in order to make input comprehensible (Lightbown & Spada 1993; Long 1996). Finally, the importance of language production is covered by the output hypothesis (Swain & Lapkin 1995). By producing language, learners notice gaps between what they want to say and what they can say, encouraging them to experiment, make mistakes and be creative.

In addition to these theories related to language learning, cognitive learning theories claim that learning in general is more effective if people spend more time thinking about a task, and when the brain has to work harder to complete it (Dale et al. 2011). Learners are

2

considered to build up knowledge for themselves by comparing what they already know and what is new according to constructivist theories of learning. Extensions of these theories are social constructivist theories, which emphasise the social and dynamic nature of learning and stress the importance of interaction (Williams & Burden 1997).

In terms of language register, Cummins (1980) distinguishes two dimensions: basic interpersonal communication skills (BICS) and cognitive academic language proficiency (CALP). Whereas BICS refer to “cognitively undemanding manifestations of language proficiency in interpersonal situations” (Cummins 1980:112), CALP refers to “proficiency that is related to literacy skills” (ibid.). In CLIL, BICS refer to everyday language that aids understanding in the classroom. More formal terms related to more academically demanding concepts, needed for school, are covered by CALP, which consequently takes longer to acquire (Dale et al. 2011).

CLIL aims at offering learners an environment where they receive plenty of meaningful input and where there is space for interaction with others, thus stimulating both language and content learning. By combining the two, a meaningful context is created, supporting language acquisition. Because CLIL pupils have to work harder to understand content, they may be better at remembering what they have learned – both in terms of language and content (Dale et al. 2011). In terms of language proficiency, the goal of CLIL programmes is to attain functional competence, both in terms of receptive and productive skills (Baetens Beardsmore 2009). This is reflected in the intensity of contact with the L2: contrary to full immersion programmes, only a selection of subject classes is given in the L2 in CLIL education. Moreover, CLIL includes separate L2 lessons that support accuracy development (Baetens Beardsmore 2009) and teachers are often non-native speakers of the target language (García 2009).

2.3 CLIL in the Netherlands

In the Netherlands, CLIL started as a grassroots movement, with parents and teachers taking the initiative for bilingual education, as opposed to a policy imposed on schools. The first secondary school started with a bilingual programme in 1989. A quarter of a century later, more than 120 secondary schools (about 20%) offer CLIL education in the highest track of secondary education (vwo);3 a selection of them also offers CLIL at the lower tracks (havo and vmbo). The L2 used in Dutch bilingual education is usually English, although there are two secondary schools that offer a Dutch-German bilingual programme. Recently a pilot study started, implementing CLIL at primary schools.

Schools are organised in networks, which are coordinated by the European Platform,4 a Dutch organisation that offers general and financial support, and monitors the quality of CLIL (de Graaff & van Wilgenburg 2015). Besides the orientation on L2 learning, bilingual education supports European and international orientation by incorporating international topics in lessons and encouraging international activities. Although Dutch bilingual education

3 _{Dutch secondary education is divided into three main tracks of education: mavo/vmbo (pre-vocational}

education), havo (higher general education) and vwo (pre-university education). The vwo stream can be further divided into two sub-streams: a main track, atheneum, and a track that adds classical languages to the regular programme, gymnasium.

4 _{Per January 1}st_{, 2015 the Dutch organisations European Platform (concerned with international cooperation in}

primary and secondary education) and Nuffic (concerned with international cooperation in higher education) merged under the name EP-Nuffic. Throughout this text the still more familiar name European Platform is used

is developed collaboratively, guided by the European Platform, schools implement the programme individually. This means that each school has a certain amount of freedom. However, the European Platform has developed standards to guarantee the quality of bilingual schools; schools that meet the criteria are organised in a national network (landelijk

netwerk tto). As for the goals set for bilingual education, pupils are expected to reach B2

(vwo) or B1 level (havo) of the Common European Framework of Reference for Languages (CEFR; Council of Europe 2001) after three years of secondary school, and bilingual education may not negatively influence “students’ proficiency in Dutch” (de Graaff & van Wilgenburg 2015:171) and “national exam results for the subjects taught in English” (ibid.).

Bilingual schools are relatively free to decide how they implement CLIL as long as the standard curriculum is followed and Dutch language proficiency is not affected. The main focus of the content lessons is on content rather than on language, led by the assumption that language use for content learning will enhance implicit learning (Admiraal et al. 2006). Half of the classes may be taught in English, and teachers are expected to have at least a B2 proficiency (according to the CEFR). Pupils who opt for bilingual education are not expected to have a certain proficiency level, but they may be selected on motivation and language aptitude. CLIL education is implemented in the first three years of secondary school; after this period, pupils choose one of four specialisations for further study, making it practically impossible to organise separated classes for regular and bilingual pupils. Apart from the organisational difficulties, the national exams that are taken in the final year of secondary schools are in Dutch, leading to a preference for Dutch subject classes in the final years (Admiraal et al. 2006).

2.4 Previous research on Dutch CLIL education

Research on CLIL is a relatively recent addition to the field of applied linguistics, but collaborations among researchers from different countries have led to interesting insights on CLIL practices and outcomes. Verspoor, de Bot and Xu mention that “a bilingual program is often seen as a threat to the national language and to traditional foreign language teaching” (2015:5). Initially, there were concerns about the development of (academic) L1 proficiency, as well as about whether subject teaching by means of an L2 would lead to less subject competence, since pupils may not fully comprehend what is taught when content is presented in an L2, and teachers may simplify the content material to adapt to their pupils’ language proficiency instead of their content knowledge (Dalton-Puffer 2007). The criteria of the European Platform standard for bilingual education specifically address these issues (de Graaff & van Wilgenburg 2015).

In a longitudinal study, Admiraal et al. (2006) investigated the effects of Dutch CLIL education. The main interest of the project was the pupils’ English language proficiency, but the researchers also looked at Dutch proficiency and results in subjects that were taught in the L2. Five schools that offered Dutch-English bilingual education took part in the study, which were followed for six school years. English language proficiency, receptive vocabulary, reading comprehension and oral proficiency were measured, both in pupils from the bilingual programmes (n=584) and in a control group of pupils who followed a regular programme (n=721). Subject knowledge was measured by using the results in the pupils’ final national exams in geography and history.

It was shown that CLIL pupils outperformed regular pupils on receptive word knowledge at all moments of measurement. However, the growth rates over the years were

similar for both groups, with the highest rate during the first year. This suggests that the CLIL pupils started off with an advantage that they were able to maintain. On the reading comprehension test the bilingual pupils performed better than their regular peers. The oral language proficiency and pronunciation of bilingual pupils, too, were significantly higher than the proficiency and pronunciation of the regular pupils. Scores for the final exams were only available for a small number of pupils (n=57 for English and Dutch; n=29 for geography and history), but analysis based on this data showed that the bilingual pupils obtained better results in the final English exam, while there were no significant differences in the other exams.

In conclusion, the results of the Admiraal et al. (2006) study show no negative effects of CLIL on pupils’ development. The bilingual pupils perform equally well on Dutch and subject knowledge, while their scores for English are higher. It should be noted that the CLIL pupils’ scores on the final primary school test, as well as their initial English proficiency level (as measured by the receptive vocabulary test) were higher than the regular pupils’ scores; possibly a result of the selection process of bilingual schools. A questionnaire administered during the first year also showed that the bilingual and regular pupils differed significantly on home language, language contact and motivation to learn English. It is therefore not possible to make any claims about the additional value of CLIL for subject learning because the test and control group did not perform at an equal level at the beginning of the study.

Recently, Verspoor et al. (2015) looked at Dutch bilingual education in a replication study. Because scholastic aptitude seems to play a role in the decision for bilingual education, the researchers compared the scores of pupils in bilingual education not only to pupils in the regular vwo stream, but also to pupils from the gymnasium stream. These pupils were assumed to be similar in scholastic aptitude (as measured by the final primary school test) and ambition, as well as in linguistic interest, thus serving as controls for higher-aptitude bilingual pupils. Pupils from four different schools and two different school years took part in the study (year 1: n=83 bilinguals, n=84 regulars, n=49 controls; year 3: n=74 bilinguals, n=68 regulars, n=41 controls). English language proficiency was not only measured by means of a receptive vocabulary tests (as was the case in the study by Admiraal et al.), but also by short writing tasks. These tests were conducted three times in each year. A questionnaire was administered at the beginning of the study to assess use of English, out-of-school contact with English, motivation and attitude.

Bilingual pupils outperformed their regular and control peers from the second moment of testing onwards, with the control pupils obtaining significantly higher scores than the pupils in regular education. However, when initial language proficiency, scholastic aptitude, out-of-school contact with English, and motivation are taken into consideration, the third-year bilinguals did not perform better than their control peers – a result probably due to their already higher proficiency level at the beginning of the year. Although bilingual pupils performed better in an absolute sense, their L2 proficiency did not develop faster than that of pupils in the control group. As for the first-year pupils, they make more progress than their regular and control peers even when the other factors are taken into account. In both school years, L2 proficiency at the beginning of the year had a significant effect on L2 proficiency at the end of the school year. Scholastic aptitude was an important predictor for language learning during the first year, but not during the third year; again possibly as a result of a higher initial language proficiency. Motivation only played a role in final proficiency among third-year pupils, whereas out-of-school contact with English did not contribute significantly to L2 development in either of the groups.

Verspoor et al. (2015) attribute the positive effect of bilingual education to interaction between teachers and pupils, and to the interactions between pupils among each other. At the same time, they admit that it is yet unclear whether a greater amount of input or bilingual teaching methods positively influence L2 development, and they note that the investigated factors are not stable over time. The authors conclude that bilingual education affects L2 development in a positive way, although the role of scholastic aptitude and initial language proficiency should not be ignored.

2.5 Language use in the language learning environment and translanguaging

There are different views regarding the place of the L1 and L2 in bilingual education. A traditional view on bilingual education states that the two languages should be kept separate, and that only the target language should be used in the bilingual classroom (Baker 2011; Creese & Blackledge 2010), resulting in “separate bilingualism” (Creese & Blackledge 2010:105). Upon entering the classroom, pupils cross a so-called ‘magical threshold’ and are supposed to speak English only. Many teachers actively discourage the use of the L1 in the L2 classroom (Dale et al. 2011). By creating a monolingual L2 environment, language learners are provided with plenty of input and are encouraged to use the language they are learning.

However, this view on the ideal language learning environment has been challenged by others who state that insider perspectives on multilingualism should be adopted (Canagarajah 2011). Professionals in the field indeed notice that many pupils continue to speak Dutch to each other during group tasks, while producing English output (bilingual coordinator, personal communication, April 15, 2015). Supporters of multilingual learning environments claim that there is a place for both languages in the classroom, since the use of the L1 can enhance L2 learning by maximising understanding and performance (Baker 2011). Especially when L2 learners share their L1, as is the case in most CLIL settings, L1 use is likely but not excessive (Azakarai & García Mayo 2015; Storch & Aldosari 2010).

The acknowledgement of the role the L1 can play in L2 learning is related to the way bilinguals use the linguistic resources they have; they naturally use both of their languages (Creese & Blackledge 2010). Multilinguals who are proficient in several languages are able to switch between the languages they speak without problems, a phenomenon referred to as

code switching: “the occurrence of elements of language A in stretches of speech of

language B” (Treffers-Daller 2009:59). Although this switching was previously seen as problematic and a sign of weakness, it has been shown that fluent switching serves specific purposes and that it “results from complex bilingual skills” (Milroy & Muysken 1995). By using different codes, multilinguals are able to emphasise the different languages they know to different extents (Canagarajah 2011). Moreover, when people are confronted with cognitively demanding tasks, they often make use of private speech (PS): externalisations of reasoning processes. This speech, which is used spontaneously and often in the L1, is also adopted while working collaboratively, as a means for building knowledge, solving difficulties, guiding behaviour and managing affectivity (Alegría de la Colina & García Mayo 2009).

In research on language learning, the use of the L1 in the process of L2 learning has often been studied as cross-linguistic influence. From this perspective, using the L1 is seen as a source of interference in L2 development (Ringbom 1987). However, more recent studies have suggested that the L1 may actually be a useful tool for L2 learning and that using a shared L1 may support learners to work at a higher level (Alegría de la Colina & García Mayo

2009). These researchers see language as a mediating tool for higher-order mental processing, based on Vygotsky’s sociocultural theory which considers learning and development to originate in social interactions (Antón & Dicamilla 1998; Brooks & Donato 1994). Language is regarded as a social cognitive tool in developing higher mental functions, used for scaffolding: assisting a learner in the learning process (Storch & Aldosari 2010).

Several researchers (e.g. García 2009, Creese & Blackledge 2010, Canagarajah 2011) use the term translanguaging to refer to the bilingual use of languages among speakers of the same bilingual culture. They naturally engage “in bilingual or multilingual discourse practices” (García 2009:44). Canagarajah (2011) also mentions the term code meshing as the use of two languages or modalities that are considered to be part of a single integrated system, unlike code switching, which treats the two languages as separate systems. Translanguaging goes beyond code switching, since the languages involved are not considered to be separate codes with clear-cut boundaries, but they are part of one linguistic repertoire in functional integration (J. Cenoz, personal communication, May 23, 2015). The term was coined by Welsh teacher Cen Williams, describing the practice in bilingual classrooms where two languages are used systematically within the same lesson (Baker 2011). He deliberately used one language for language input and another for language output, switching the languages’ roles across lessons. By activating both languages, learners can maximally benefit from being bilingual.

Translanguaging can be spontaneous, as is often the case when bilinguals who speak the same languages communicate, or intentional, for instance when a language teacher encourages pupils to use all linguistic resources that are available to them. It takes multilingualism as a starting point for language learning, not only as an ultimate goal. By encouraging learners to use multiple languages, content is processed in multiple ways, supporting internalisation (Baker 2011). It is a way to scaffold academic learning, both language- and content-wise, since the use of the L1 can make the target language more salient and comprehensible. Meanwhile, translanguaging is a useful tool in fostering metalinguistic awareness by making connections between the languages involved, and in class management, encouraging class participation (Creese & Blackledge 2010).

Although natural translanguaging is considered to be a useful practice in bilingual education, it is in need of strategic application, since directly translating everything to the L1 would discourage learners from paying attention to explanations in the L2. Canagarajah (2011) considers translanguaging to be a naturally occurring phenomenon that cannot be restrained by monolingual educational policies. He therefore argues for an approach that takes into account discourse and rhetorical strategies when judging whether translanguaging is appropriate and effective. Thus a critical view on assessment and instruction can be developed.

2.6 Previous research on multilingual language use in L2 learning

The use of the L1 in the L2 classroom has been investigated by several researchers, looking both at L1 use by language teachers (e.g. Duff & Polio 1990, Edstrom 2006, Gierlinger 2015) and by L2 learners. In the latter studies, learners from different ages and backgrounds were investigated, supporting the view that use of the L1 can be beneficial for L2 learning. This section addresses a couple of those learner-focused studies, thus giving an overview of research conducted in several L2 contexts.

Dalton-Puffer’s (2007) study on CLIL discourse in Austrian bilingual classrooms did not directly investigate discourse in terms of L1 and L2 use, but she noted that pupils in her study preferred to use the L2 (English) for content core questions, while mainly using the L1 (German) for peripheral matters, such as directives to peers (L1 used in up to 80% of the utterances). Moreover, she suggested that age and proficiency may influence the amount of L1 use, with learners with less L2 competence switching to their L1 more often.

Alegría de la Colina and García Mayo (2009) found that the L1 is indeed an important tool for learners with a low L2 proficiency. They investigated undergraduate Spanish learners of English (n=24, mean age 22) who performed three collaborative tasks, looking specifically at the influence of the type of task on the amount of L1 use and the function of the L1. The three tasks – a jigsaw task, dictogloss and a text reconstruction task – all required the participants to complete a text in pairs. The participants were recorded while working on the tasks, in order to analyse their speech. The researchers calculated the quantity of L1 use and segmented the speech into episodes with specific functions, dividing L1 use into three categories: metacognitive talk, metatalk and off-task talk. Metacognitive talk encompassed ‘talk about the task’, such as “planning, organizing and monitoring the activity, setting goals and checking comprehension” (Alegría de la Colina & García Mayo 2009:330). Metatalk is ‘talk about talk’ and “includes segments of the dialogue in which students discuss form” (ibid. 332), such as lexical searches or discussions about grammar. Off-task talk included any talk that was unrelated to the task at hand.

The text reconstruction task and the dictogloss task elicited the highest L1 use (around 75% of the words) and the jigsaw task the lowest amount (55%), although there were considerable differences in the total number of words used across both tasks and pairs. The participants mainly used the L1 for metacognitive talk, particularly for task management, and meaning and content clarification purposes. Of the L1 use, metacognitive talk was used most in the dictogloss task (72% of the episodes) and least in the text reconstruction task (56%). Metatalk was mainly used for talk about grammar, and less for lexical talk (19-33% of the episodes across the three tasks). Overall, metacognitive talk encompassed more episodes, but metatalk covered a greater number of words. Off-task talk was hardly used: less than 1% of all L1 use in all tasks. The authors therefore state that L1 use does not mean that students are not involved in the task. The use of the L1 did support the students in thinking and self-regulating more quickly, as well as in establishing fruitful interaction and transferring skills to the L2, leading to performance at a higher level and enhanced motivation.

In a similar study Storch and Aldosari (2010) looked at L1 use in pair work by Arabic learners of English (n=36). The participants were divided into three groups based on proficiency: pairs with two high proficiency learners, pairs with two low proficiency learners and pairs with a low and a high proficiency learner. These researchers, too, presented their participants with three tasks: a jigsaw, a composition and a text-editing task. The participants’ speech was recorded and transcribed, and subsequently analysed for the quantity and functions of L1 use. Both L2 proficiency and task type were taken as variables in the analyses on L1 use; word counts as well as turn counts were used for calculating the amount of L1 use. Regarding the functions, distinctions were made between task management, discussing and generating ideas, grammar deliberations, vocabulary deliberations and mechanics deliberations. Whenever a turn had multiple functions, the predominant function was counted.

The results showed large differences in L1 use between the pairs within the three proficiency groups; on average, the students produced 7% of the words in their L1. Proficiency did not seem to influence the amount of L1 use in the jigsaw and the composition tasks. A task effect was only found for the editing task, with overall higher percentages of L1 use for all groups, and a more pronounced effect for the low proficiency pairs, which used more L1 (29% of words) than the pairs with higher (9%) or mixed proficiencies (11%). The authors suggested that this could be due to the difficulty and focus of the task, which required the students to understand and improve the written text at hand. The L1 was mainly used for task management (45% of turns), followed by vocabulary deliberations (26%). Since the L1 served important cognitive, social and pedagogical functions, the authors argued that it should be available for L2 learners in class, instead of being prohibited.

Canagarajah (2011) investigated the use of several languages in a written narrative in L2, produced by an Arabic student who also had some knowledge of French. The researcher distinguished four types of strategies. Firstly, recontextualisation strategies were related to shaping ecology in favour of multilingual practices, determining whether mixing languages is possible. Secondly, voice strategies had to do with creating space to experiment with languages, positioning oneself as a multilingual. Thirdly, interactional strategies concerned negotiating meaning with readers, while textualisation strategies were related to considering the text as a multimodal social practice, shaping thoughts, and using composing strategies to develop the text effectively. Canagarajah admits that it is difficult to assess the translanguaging skills of multilinguals, because it is difficult to determine whether deviations from the norm are caused by positive transfer or negative interference (i.e. mistakes). He suggests that the two are separable by looking at intentionality (with codemeshing being an intentional, conscious process, whereas mistakes are unintentional) and the function of deviations: only those that have rhetorical and communicative functionality can be considered instances of codemeshing. Canagarajah claims that it is possible to develop translanguaging skills by raising awareness of the linguistic resources available to multilinguals, and by critically questioning codemeshing choices. Within the classroom, teachers can serve as important models, scaffolding students’ attempts and encouraging them to use their wide linguistic knowledge.

In summary, L2 learners seem to translanguage in the sense that they use their L1 while working on group tasks with peers. Rather than saying that L1 use should be actively encouraged and promoted, researchers seem to agree that use of the L1 should at least not be prohibited in L2 classes, making way for a supporting role for the L1 in L2 learning.

3 Aim of the present study

Previous studies investigating Dutch bilingual education have mainly focused on the learning outcomes of bilingual education, by looking at language proficiency (e.g. Admiraal et al. 2006; Verspoor et al. 2015), as well as motivation (e.g. Mearns 2014). The current study aims to add to our knowledge about bilingual education by investigating the role of the L1 in pupils’ oral communication in the bilingual classroom.

This topic is relevant both from a theoretical and a practical perspective. Theoretically, because it is linked to translanguaging, which advocates that bilingual speakers freely use all of their linguistic resources, switching between languages when they deem it necessary. In how far do pupils in Dutch bilingual classes already translanguage?

Practically, because professionals in the field have noted that pupils do not always speak English, even when they are required to do so (O. van Wilgenburg, personal communication, March 26, 2015). However, it is not clear when pupils switch to Dutch – does language use depend on specific language functions or context, or is use of the L1 necessary to support understanding, as a so-called ‘tool’ in a translanguaging environment?

Since this topic has not been investigated in the Netherlands before, the aim of the present project is to provide an image of pupils’ language use in bilingual classrooms by means of an exploratory research study. Its goal is to investigate to what extent Dutch is actually used and needed, and when pupils switch to Dutch – all in all giving an impression of everyday peer communication practices in Dutch bilingual education. The results may serve as a basis for future research on translanguaging and comparisons of (inter)national CLIL programmes, as well as being a resource to CLIL professionals in the formation of classroom policies regarding language use.

3.1 Research questions

The goal of the research project is to gain more insight into the everyday language use of pupils in bilingual education, answering the following research question:

Main RQ How are Dutch and English used among pupils in bilingual classrooms?

In order to answer this main research question, several sub-questions will be investigated. These questions are grouped around two themes: general language use and switches to Dutch. For all questions, comparisons will be drawn between the results from the first-year and third-year pupils; in addition to these group level analyses, the first research question will also be investigated on an individual level in order to look more closely at individual variation in language use.

General language use

RQ1 What proportion of speech is uttered in Dutch and English during peer interactions in the bilingual classroom?

RQ2 Is the proportion of L1 use related to L2 proficiency?

RQ3 To what extent is language use related to class setting, i.e. regular class or collaborative writing task?

RQ4 For which language functions are Dutch and English used?

Switches to Dutch

RQ5 Is the number of switches related to class setting, i.e. regular class or collaborative writing task?

RQ6 When do pupils switch to Dutch, i.e. at what point in an utterance do pupils switch, and are switches related to changes in speaker or language function?

RQ7 Are switches to Dutch salient, i.e. do switchers continue their utterance in Dutch or do they switch back to English?

3.2 Hypotheses

Based on previous research, it is expected that bilingual pupils mainly use their L2 (English) in classroom interactions. L1 use is expected not to be excessive (Azakarai & García Mayo 2015) but related to L2 proficiency, with the first-year pupils using more Dutch than the third-year pupils (Storch & Aldosari 2010). Dutch is expected to be used more in content classes than in language classes, since language is merely a tool in content classes whereas it is the focus of attention in language classes. In a similar fashion, it is expected that the L1 is used more often in a regular class setting than in a collaborative writing task setting. This is related to Cummins’s (1980) theory about BICS and CALP: the regular classes require the pupils to use more academic language, while the collaborative writing tasks give them more freedom to have an informal discussion in coming up with ideas and making decisions about the texts. In correspondence with the studies by Alegría de la Colina and García Mayo (2009), and Storch and Aldosari (2010), pupils are expected to use Dutch mostly in situations where they engage in metacognitive talk, i.e. task management. First-year pupils are predicted to use Dutch more often for metatalk, because of their assumed lower language proficiency and lesser familiarity with specific vocabulary.

Among first-years, switches are expected to occur more often in regular classes than in the collaborative writing task settings – again because of the higher demand for academic language (Cummins 1980). In terms of relation to speaker and function changes, switches are predicted to coincide most often with switches in speaker and function. In accordance with Nortier (1989), it is hypothesised that third-year pupils, who are considered to be more balanced bilinguals due to an expected higher L2 proficiency, show a higher percentage of intrasentential switches (i.e. later in an utterance) and single-word insertions (i.e. switches to Dutch which are less salient). Intersentential switches (i.e. at the beginning of an utterance) are expected to be more salient, leading to more peer replies in Dutch, since they may indicate a change in matrix language.

4 Methodology

4.1 General research set-up

The research questions were investigated by means of a mixed-methods study. The project was primarily qualitative in nature, being based on an interpretive case study. Qualitative research investigates social phenomena in a setting that is as natural as possible, aimed at discovering and understanding the characteristics of a phenomenon, process or the views of the people involved, and revealing “how all the parts work together to form a whole” (Merriam 1998:6). Since the analyses also included frequencies and statistical analyses, the study also had quantitative aspects, leading to a mixed design. The inclusion of participants from different school years makes it a cross-sectional study.

The case under investigation in this project is a Dutch secondary school offering bilingual education. Within this case, a sample of pupils was selected, encompassing one first-year (approximately aged 12-13) and one third-year (14-15 years old) class. A smaller group of pupils was selected and matched in order to make comparisons between the first and third year on an individual level. The pupils were matched in such a way that a maximum variation sample was striven for, in terms of home language (Dutch, English, or another language), language use outside the classroom, as well as experiences with English and English language proficiency (as measured by an L2 proficiency test; see 4.4). The

original selection consisted of sixteen pupils (eight first-years and eight third-years), but two couples had to be removed because pupils were absent during recorded lessons.

For reasons of triangulation, and strengthening internal validity and reliability (van der Donk & van Lanen 2013; Merriam 1998), data related to the research topic was collected in multiple ways. Observations of pupils executing group tasks provided data on communication in actual group work. Interviews with pupils involved in the group tasks revealed pupils’ views on language use in the bilingual educational context. Finally, interviews with teachers provided more information about their experiences and their policies regarding language use in the classroom.

4.2 Study setting and participants

The participants were all pupils from a bilingual secondary school in the Netherlands, which offers both regular and bilingual education at all secondary school levels. The vwo stream of the school is one of 57 ‘senior’ bilingual programmes in the Netherlands, with more than twenty years experience with bilingual education that meets the European Platform’s standards. English is the language of instruction and communication in the subject classes mathematics, geography, biology, history, physical education, drama (year 1 to 3), multicultural studies (year 2), physics (year 2 and 3) and chemistry (year 3). In addition to the bilingual classes, pupils participate in international excursions, exchanges and interactive English-language plays (school guide 2014). Following the European Platform’s bilingual standard, the school has a policy of using the target language as language of communication in CLIL classes; pupils are given a mark for their use of English to stimulate communication in the target language (school’s bilingual coordinator, personal communication, June 2, 2015).

The study was set up in coordination with one of the school’s bilingual coordinators. Since one of the aims of the research project was to look at differences between different years, while avoiding the possible influence of teacher as much as possible, a first- and a third-year class were selected in such a way that they shared their English and subject teachers. While the mathematics teacher was a native speaker of English, the English teacher was a Dutch native with a high proficiency in English, also functioning as language coach for CLIL colleagues. Both teachers had ample experience in bilingual education, having worked at the bilingual department of the school for almost seven (English) and nine (mathematics) years, in addition to multiple years of experience as teachers in regular education in their native countries.

4.3 Questionnaire

The questionnaire consisted of 25 questions about the pupils’ social and linguistic background, as well as their direct and indirect experiences with English (e.g. through holidays and media). The questionnaire was an adapted version of Bonnet’s (2002) pupil questionnaire, which was designed to gather information about thousands of pupils learning English in different European countries, and included questions about the pupils’ specific school curricula. Since the current project investigated pupils from one school only, such information was available from the school; the questions about the school curriculum were therefore deleted. Other questions were slightly adapted to reflect recent changes in (social) media use.

Data gathered via the questionnaire was used to illustrate the background of the pupils from the two classes on a group level, and to match pupils for the analysis of language use on an individual pupil level.

4.4 Proficiency test

It has been shown that proficiency plays an important role in performance in various modalities and in academic success (Gablasova 2012). The participants’ L2 proficiency was therefore assessed using an English grammar test designed for the NELSON project, which investigated reading and writing skills in both Dutch and English classes (Schoonen, van Gelderen, Stoel, Hulstijn & de Glopper 2011). The test assesses pupils’ knowledge of verbal inflection, comparisons, determiners, quantifiers, adjectives, adverbs and sentence structure, and consists of 60 items. Each correct item is awarded with one point, leading to a maximum score of 60. The following examples illustrate items on verb conjugation, comparatives, and adjectives and adverbs.

(1) ____ she ____ (live) with her parents?

(2) This hotel is ____ (comfortable) than the other one.

(comfortabler / comfortablest / more comfortable / most comfortable) (3) She played ____ (bad) and she lost the game.

(bad / badly)

The proficiency results served both to answer RQ2 on the relation between L2 proficiency and L1 use, and to match match pupils for the analysis of language use on an individual pupil level.

4.5 Lesson observations and collaborative writing tasks

The interactions between a subsample of the pupils were audio recorded during English and mathematics classes. Recordings were made using two Roland Edirol R-1 audio recorders and a Marantz Professional PMD620 audio recorder. Both classes participated in two collaborative writing tasks (one during English, one during mathematics) that required the pupils to write a newspaper article about either a poetry festival or a mathematics contest. They were given approximately 25 minutes to write the text in groups of four to six pupils, of which three groups were recorded. The two tasks were conducted on different days.

In addition to the observations in group task setting, the pupils were observed during regular classes; one English and one mathematics class were observed in each year. During the first-year regular English class, pupils worked in pairs on a task that required them to give directions to each other. Their regular mathematics class was focused on working out angles in triangles. The third-years discussed various poems and analysed the lyrics of the song ‘Viva la vida’ during their regular English class; these activities were executed in pairs. In the third-year regular mathematics class, pupils used computers to work on calculations and reports for a project on probabilities.

Additional observations for a general impression of language use were made during physical education classes; for each class one lesson was attended. According to O. van Wilgenburg (personal communication, March 26, 2015), these classes provide a good indication of how comfortable pupils are in using the L2, because they are more occupied

with the activities at hand than with thinking about which language they are supposed to use. Unfortunately recording was not possible during these lessons because of the active nature of physical education classes.

The observations of the English and mathematics classes served as the study’s primary data source, providing data that was used to investigate all research questions. In addition, examples were taken from the observation data to illustrate L1 use.

4.6 Interviews

A subset of the pupils who were audio recorded while performing the collaborative writings tasks was invited for an individual semi-structured interview on their experiences with bilingual education and language use in class. The interviews were conducted during physical education or English classes, or during breaks. The group consisted of the original pupils selected for individual analysis (n=16; see 4.1) and one extra first-year pupil, who was considered to be a suitable substitute match for an absent peer. This led to a total number of seventeen interviews with nine first-years (five girls) and eight third-years (four girls). The two teachers whose classes were involved in the project were interviewed in a semi-structured way about their experiences with bilingual education and their policies on language use in their classes.

Interviews were included in order to investigate whether pupils’ and teachers’ perception of pupils’ L1 use in class corresponded with the actual data from the observations.

4.7 Procedure

All parents and pupils were informed about the research project before its start by the bilingual coordinator, who served as a contact person. They were told that all data would be processed anonymously but nevertheless given the option to object to their child’s participation. The pupils were informed about the study during a mentor class, before they filled out the questionnaire and proficiency test. Both parents and pupils were told that the aim of the study was to investigate bilingual education, without informing them about the exact topic (i.e. language use and switching between L1 and L2 in the bilingual classroom). None of those concerned objected to participation.

The observations and group tasks took place during regular English and mathematics classes, ensuring a natural context. Although all pupils took part in the tasks, execution of the group task was only audio recorded for a small group of pupils. Based on the results from the questionnaire and grammar test, pupils were selected for audio recording and individual interviews. Because a couple of pupils who were pre-selected based on the L2 proficiency test and questionnaire were absent during one or two of the lessons that were observed, the final individual analysis includes less pupils than originally planned. The teachers were interviewed about their experiences with and views on bilingual education on separate occasions.

4.8 Analysis

The audio recordings of the lesson observations were transcribed and analysed using the Analysis of Speech Unit (AS-unit), a unit especially developed for oral data analysis (Foster,

Tonkyn & Wigglesworth 2000). The AS-unit was designed as an alternative means of analysis for spoken (second) language, and is an adaptation of Hunt’s T-unit. The authors provide clear instructions for segmenting data: one unit consists of an independent clause, sub-clausal unit and any associated subordinate clauses. An independent clause is defined as “minimally a clause including a finite verb” (Foster et al. 2000:365); an independent sub-clausal unit is a minor utterance or “one or more phrases which can be elaborated to a full clause by means of recovery of ellipted elements from the context of the discourse or situation” (Foster et al. 2000:366). A subordinate clause minimally has a finite or non-finite verb element plus at least one other clause element. Pauses longer than 0.5 seconds and falling or rising intonation can mark phrase boundaries. Since the observations have a highly interactional nature, a ‘level two’ analysis was applied, excluding one-word minor utterances and verbatim echo responses.

Because the study focuses on communication of pupils among each other, talk between a teacher and pupils was excluded. Utterances that were mumbled were left out as well, just as quotes and names that were uttered separately (e.g. calling another pupil). Expletives and exclamations, such as ‘what the fuck’ and ‘oh my god’, were deleted too. Such expressions were considered to be fairly common in monolingual Dutch adolescent speech, making it difficult to assess whether they are truly English expressions or borrowed chunks that have become part of the Dutch (teenage) lexicon.

All units were coded for speaker, the language used to utter them and the language function of the unit. Based on the studies by Alegría de la Colina and García Mayo (2009) and Storch and Aldosari (2010), units were categorised in six categories: task management, vocabulary talk, grammar talk, mechanics talk, phatics and off-task talk. A seventh category was added for units whose function was unclear. Table 1 specifies the type of units that were part of each of the categories.

Table 1. Overview of the categories used in labelling language function. The studies by Alegría de la Colina and

García Mayo (2009) and Storch and Aldosari (2010) served as a basis for the categories used in this study.

Category Description

Task management Talk related to the task: clarifying instructions, generating ideas, task division, redirecting attention, tracking progress, solving practical issues, remarks about the task

Vocabulary talk Talk about word (meaning) issues

Grammar talk Talk about grammatical and sentence structure issues Mechanics talk Talk about spelling, punctuation, pronunciation Phatics Establishing contact, expressing social conventions

Off-task talk Talk unrelated to the task at hand: talk about other classes or school obligations, remarks about (peer) behaviour, the research study, free time or jokes

Unclear Language function unclear

Whenever a pupil switched to Dutch, the location of the switch was noted (at the start of the unit or mid-unit), along with whether it occurred together with a switch in speaker and/or language function (both as compared to the previous unit). To answer the question how pupils switch, it was noted whether the speaker switched back to English within the same unit or not. The language of peer response was also coded in order to investigate whether a switch was followed by other pupils.

5 Results

5.1 Participant characteristics

The participants were year and third-year pupils from a bilingual vwo stream. The first-year class solely consisted of pupils from the gymnasium track, while the third-first-years formed a combined class of both atheneum (n=10) and gymnasium (n=17) pupils. All third-year pupils had started bilingual secondary education in the first year. Table 2 provides an overview of the general characteristics of the two classes.

Table 2. General characteristics of the participants in terms of number of pupils, gender, age and home

languages (L1), as well as English language proficiency (maximum NELSON score: 60).

Year Number of pupils (females) Mean age (range) L1 English L1 other than English or Dutch Mean L2 proficiency scores (sd) 1 21 (13) 5 13;2 (12;1 – 13;7) 2 7 51.19 (6.79) 3 27 (15) 15;0 (13;11 – 15;8) 3 3 56.78 (1.99)

The majority of the pupils came from homes with at least one parent who had followed higher education (67% and 89% in year one and year three, respectively). Four pupils in year one and one in year three indicated they did not know their parents’ educational background. All of the pupils with a home language different from Dutch or English had lived in the Netherlands at least since age two. Apart from one first-year pupil who had always lived in the Netherlands, the native English speakers all moved to the Netherlands by age six or seven. Most (n=11) of the first-year non-native English speakers had started to learn English in a non-educational setting (e.g. at home or with family abroad); half of these pupils before the age of eight. Of the remaining eight pupils who had started to learn English at school, two had started before they were eight years old. The majority (n=17) of the third-year pupils had started to learn English at school; all but one after the age of seven. Seven pupils started learning English at home, all of them before they were eight. All pupils acknowledged the importance of being able to speak English, and most of them had a positive attitude towards the language (with the exception of one and three pupils from the first and third year respectively).

Because of the small number of participants, it was not possible to verify whether the NELSON scores were normally distributed. Therefore a Mann-Whitney U test was executed to investigate whether there were any differences in proficiency between pupils from the two streams within the third-year class, or between the first-year and the third-year class. It was shown that the third-year pupils from the gymnasium stream did not differ significantly from the atheneum pupils in terms of proficiency: U = 84, p = .980. The third-years did, however, outperform the first-year pupils: U = 104.5, p < .001. The high mean score of 56.78, where the maximum score was 60, and small standard deviation in among the third-year pupils suggest that they performed at ceiling, possibly concealing the actual variance in language proficiency.

5

5.2 Language use: quantitative data

In total 8,285 AS-units were analysed and categorised in terms of speaker, the language that was used to utter them, and their function. Seven of these units were uttered in languages other than Dutch: German (four times), French (twice) and Latin (once). These cases all concerned off-task utterances, often jokes or literal translations of what a peer had just said in English or Dutch, and were removed from the data set.

The first research question addressed the proportion of speech uttered in Dutch or English. Figure 1 shows the proportions of English, Dutch and mixed English-Dutch units across the eight settings: the collaboration task and regular classes of English and mathematics in the two years. The exact figures are reproduced in Appendix A. Data from the first and third year were compared by means of multiple chi-squared tests in order to see whether the two classes showed a different distribution in the use of English, Dutch and mixed language in any of the four class settings. A Bonferroni correction was applied to adjust the alpha level for multiple comparisons. All statistical results are provided in Table 3.

Figure 1. Language use per year and setting. For each setting, the left column represents data from year 1; the

right column shows the distribution in year 3. The total number of units in each setting is indicated underneath the bar charts.

Table 3. Results of the multiple chi-squared tests that compared the distribution of English, Dutch and mixed

language units across the two years and four class settings; α = .0125.

Setting χ² (df) p Cramer’s V

English class 103.736 (2) < 0.001 .336

English task 2.933 (2) .231 .028

Maths class 174.302 (2) < 0.001 .396 Maths task 52.752 (2) < 0.001 .144

The results showed statistically significant relations between language use and setting in all four settings except for the English collaborative writing task setting. Inspection of the

standardised residuals revealed that the fact that first-year pupils used more English (z = 3.9) and less Dutch (z = -7.2) than statistically expected during the English regular class contributed most to this result, which is a medium-sized effect. Data from the mathematics regular class setting showed that the pupils from the third year used less English (z = -4.7), while also uttering more Dutch (z = 9.6) and mixed units (z = 4.4). Again, this was a medium-sized effect. Finally, the first-years used less Dutch than expected during the mathematics writing task (z = -5.3) while the third-years used more Dutch: z = 4.5. This was a small effect.

In order to look more closely at individual differences in language use and to make a more detailed comparison between the two years in addition to the group level analysis, six first-year pupils were matched with six third-year pupils (see 4.1 for more information on the matching procedure). Figure 2 and Figure 3 show the results for the English and mathematics collaborative writing task settings respectively. Sets 1 and 2 contained girls who either had an L1 different from Dutch or English (set 1) or Dutch as L1 (set 2); all four had started learning English at school at age 9 or later. The L2 proficiency level of the pupils in set 1 was higher than their respective class average, whereas the pupils in set 2 had obtained scores that were lower than average. The boys in set 3 were L1 Dutch speakers who had started learning English at school around age 8; the pupils in set 4 had a similar background but had started learning English at home. The pupils in set 5 were L1 Dutch pupils who had started learning English at home at age 5. Set 6 concerned pupils with English as their L1. In terms of proficiency level, the pupils from sets 3 and 4 had scored lower than average in their corresponding classes, whereas the pupils from sets 5 and 6 had obtained higher scores. 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 1 2 3 4 5 6

Dutch Mixed English

Figure 2. Language use in the English collaborative writing setting for each of the six sets of pupils, showing the

proportion of English, Dutch and mixed English-Dutch units in their speech. For each set of pupils, the left bar represents the first year pupil, and the right bar the third year pupil.