The First Language in Science Class: A Quasi-Experimental Study in Late French Immersion
Abstract
This article reports analysis of data collected from a quasi-experimental study in 2 Canadian late French immersion science classes. We examine if, how, and when the first language (L1) is used when students in the first years of their second language learning talk about complex science concepts. We compare differences in groups following a 2-month intervention. Specifically, we study differences in complexity of oral utterances, and differences in use of L1 in oral utterances after participating in an intervention using a literacy-based approach, or the typical, district-prescribed approach. Furthermore, we assess whether increased use of the L1 in complex content statements is positively associated with gains in French and science knowledge. Advanced statistical analyses linked the complexity of student utterances with language use, written skills in French, and performance in science. In this way, this article makes a significant contribution to the existing body of literature on this important topic.
TEACHERS WHO WORK OR HAVE WORKED in French Immersion programs can likely identify with the challenges of teaching complex subjects like science and math when students’ language proficiency is limited, especially during the beginning stages of the programs. These challenges arise as a result of several factors. First, school administrators and parents expect that their children cover the same subject matter as students enrolled in regular English-medium programs. Second, available classroom materials that have been developed for native speakers of French tend to be too advanced for beginning language learners. Moreover, as the following quotation from the Atlantic Provinces clearly indicates, it is expected that teachers and students will only use the target language (TL) in all instances in French Immersion, even at the beginning of the program:
L’apprentissage doit être intensif sans toutefois être une noyade.. Les élèves doivent très tôt pouvoir comprendre le français et l’utiliser pour communiquer. Il est donc essentiel que la seule langue de communication dans la salle de classe soit le français. (Atlantic Provinces Education Foundation, 1997, p. 9)
Learning must be intensive without drowning students. Very early in the learning process, students will be able to understand and use French to communicate. It is therefore essential that French be the only language for classroom communication.
As a result of these challenges, teaching in content-based areas such as science tends to be quite teacher centered. Teachers tend to put the prescribed curriculum documents and textbooks aside and create their own simplified materials, usually in the form of brief, factual notes that students copy from the board and then memorize for a quiz. The focus is principally on the content of the subject. This often results in the fact that students speak and write infrequently in French and when they do, their production is limited to a few words. Moreover, little attention is paid to providing corrective feedback to students about the quality of their linguistic output (e.g., Harley, Allen, Cummins, & Swain, 1990). Researchers such as Lyster (2007) and Swain (1996) suggested that these realities explain, at least in part, why French Immersion students’ skills in French are less developed and less accurate than one would like or expect (Harley, 1998; Lyster, 1987, 2004).
The reality of teaching complex content in French immersion has been the subject of much discussion and debate in professional circles and in the scholarly literature. This topic has surfaced frequently in local and national conferences sponsored by professional organizations like the Canadian Association of Immersion Teachers. Laplante (2000, 2007), Lyster (1987, 2004, 2007), Cormier and Turnbull (2009), Genesee (1991, 2007), Cummins (1998), and Swain (1996) have also made noteworthy contributions in the scholarly literature. Discussions have focused on the need for more integration of language and content in teaching subjects like science, math and social studies.
However, less attention has been paid to first language use as a strategy to facilitate the integration of content and language. In fact, while first language use has long been a topic of much debate and controversy in many second and foreign language contexts beyond French immersion (e.g., Cook, 2001; Levine, 2003; Liebscher & Dailey-O’Cain, 2004; Macaro, 2005, 2009; McMillan & Turnbull, 2009; Turnbull, 2001) the results of this debate have generally been ignored by French immersion professional organizations and policy makers throughout Canada. Nevertheless, some researchers (e.g., Behan, Turnbull, & Spek, 1997; McMillan & Turnbull, 2009; Sanaoui, 2005, 2007; Skerritt, 2003; Swain & Lapkin, 2000; Walsh & Yeoman, 1999) suggested that teacher and learner codeswitching (CS) practice varies significantly in French immersion, especially in teaching and learning content. All scholars call for more research on CS in French immersion.
Therefore, we conducted a quasi-experimental study in late French immersion and focused on two goals. First, we compared two different science teaching approaches: a literacy-based approach (the experimental intervention) and the typical district-prescribed approach (the control group). This part of the study assessed the impact of both approaches on students’ science knowledge and on their written skills. Results that respond to this first goal are briefly summarized here, but are reported in more detail in Cormier and Turnbull (2009). This article reports the findings from our second goal—to examine the role of the first language (L1) and the second language (L2) as students talked about complex science content during the oral interview component of our study. We expected that during the initial oral interview, both groups would rely on their L1 during their responses in a similar way. We wondered if and how groups would respond differently during our final interview, after our 2-month intervention. We expected that as complexity in content increased, use of the L1 as a scaffolding tool to express the ideas would also increase. Finally, our hypothesis was that if the amount of L1 increased as the complexity of the content increased, this trend would be positively correlated with results in French and Science. In other words, we speculated that the increased use of L1 would serve as a scaffold for students to enable them to talk about complex science content and would be positively reflected in students’ achievement in French and in science. Before describing the analysis and results, we begin with a brief synthesis of the literature that has examined the role of the L1 in second and foreign language learning, especially in French immersion.
RELEVANT LITERATURE
Our study draws, in part, on several studies that have used sociocultural theory and a Vygotskyan analysis of verbal interaction to examine learners’ use of the L1 as a cognitive and scaffolding tool to carry out tasks and make sense of content and language in the TL. This body of literature suggests that the cognitive benefits of the first language may be especially relevant for learners with a low level of TL proficiency dealing with challenging tasks and content. This making-sense process most probably begins in the learner's L1, where prior knowledge is encoded and needs to be accessed. Content and language learning happen simultaneously while bridging prior and new knowledge, during learning events that may occur in L1 while bridging towards the L2. Since language is also the tool students use to communicate learning, interactions between L1 and L2 may occur during the making-sense process.
This notion was first introduced by Brooks and Donato (1994) who completed research in Spanish language learning in the United States. They found that learners, especially beginners, often benefit from using the first language when negotiating meaning, allowing learners to initiate and sustain verbal interaction with one another. Drawing on Brooks and Donato (1994), Behan et al. (1997) extended this work into French Immersion. They tape-recorded Grade 7 late French immersion students working in groups to complete a cognitively challenging jigsaw task; they concluded that “L1 use can both support and enhance L2 development, functioning simultaneously as an effective tool for dealing with cognitively demanding content” (p. 41). In 2000, Swain and Lapkin reported that Grade 8 early French immersion students were able to complete a collaborative task more successfully by using some L1 for important cognitive and social functions. Swain and Lapkin concluded that judicious L1 use supports second language learning and production in the second language. They argued that “to insist that no use be made of the L1 in carrying out tasks that are both linguistically and cognitively complex is to deny the use of an important cognitive tool” (Swain & Lapkin, 2000, p. 269).
Although not in French immersion, Anton and DiCamilla (1998), and Dailey-O’Cain and Liebscher (2009) have also drawn on a sociocultural framework to show how second language learners use their L1 to the benefit of their L2 development. Antón and DiCamilla (1998) provided evidence that adult beginner-level learners of Spanish used the first language as a cognitive tool for providing each other with scaffolded help, for maintaining cooperation, and for externalizing internal speech. Dailey-O’Cain and Liebscher (2009) offered detailed discourse analyses of interactions among intermediate and advanced university-aged learners of German and their instructors in a content-based third-year seminar on applied linguistics taught in German. The authors demonstrated that these learners developed naturalistic codeswitching practices even when teachers maximize their TL use. Dailey-O’Cain and Liebscher argued from the perspective of sociocultural theory that codeswitching helps learners to structure discourse in classrooms which promotes many language learners’ goal of aspiring to bilingualism and to interact as fluent bilingual speakers do.
Wannagat (2007) examined the influence of classroom interaction on teaching and learning processes in late immersion history lessons in Germany and Hong Kong. Wannagat conducted a qualitative analysis of student and teacher talk and focused particularly on L1 use. The students’ L1 was used and perceived to be important for “task involvement” (p. 679) and “to tell stories,” that is to focus the attention and to get students emotionally involved (p. 679).
Most of the studies of which we are aware have been qualitative and hence provide rich descriptions of the usefulness and the ways in which L1 use can contribute to learners’ L2 development. However, quantitative studies providing evidence of a facilitative effect of the first language are rare. Macaro (2009) reports two studies (Meng, 2005; Tao Guo, 2007), which attempted to discover whether codeswitching contributes to better learning, either in the short term or the long term. Although the findings of the two studies do not provide conclusive evidence that codeswitching leads to more learning than exclusive TL use does, Macaro argues that banning the L1 from the communicative L2 classroom may be reducing learners’ cognitive and metacognitive opportunities.
OUR STUDY
Participants
The two Grade 7 participating classes were housed at the same medium-sized school located in a middle-class neighbourhood.1 This school was located in a suburb (population 18,000) of a metropolitan centre in eastern Canada (population of 100,000). Students were in the second year of late French immersion and both classes worked with the same team of teachers for all subjects, including science. Students in both classes had generally lived in this same neighbourhood all of their life. The students and their parents, all of European descent, were all first language speakers of English and few students had any experience using French outside school. The experimental group included 25 students—10 girls and 15 boys (mean age of 12. 3 years) whereas the control group included 24 students—11 girls and 14 boys (mean age of 12.4 years).
Teaching Approaches
The teaching approaches in both classes were developed in collaboration with the teacher who accepted to allow us to work in her classes. Drawing on the provincial curriculum document for late French Immersion science, we identified the following outcomes for the two units: (a) students will demonstrate a general understanding of seismic and volcanic activity, including a basic understanding of tectonic plate movement; (b) students will identify seismic and volcanic areas around the world and will explain why these areas are prone to seismic and volcanic activity; (c) students will demonstrate ability to respond to questions, solve problems and make decisions using a variety of techniques; and (d) students will demonstrate an ability to identify questions for further study and practical application (NBME, 2002). Although no language outcomes were proposed in the Ministry document (therefore none were identified or included for the control group), the following language outcomes were covered in the intervention for the experimental class:2 1) students will demonstrate an ability to share and clarify their ideas orally; 2) students will write in French to organize, clarify, and communicate their ideas relating to the science content; 3) students will demonstrate their ability to use reading strategies such as prediction, questioning, clarifying for understanding and summarizing. Moreover, while little explicit focus on form instruction was included in the unit, both approaches provided students with an opportunity to practice and consolidate vocabulary and necessary language structure.
To control for teacher effect, the same research assistant,3 also a certified teacher, implemented the interventions in both classes. Time dedicated to both interventions was practically identical—21 40-minute periods.
Experimental Approach.
Theoretically, the experimental (literacy) approach builds on important work related to content-based teaching in second and foreign language contexts (e.g., Mohan, 1979, 1986; Snow, Met, & Genesee, 1989), especially Lyster's (2007) counterbalanced approach and Laplante's (2000, 2007) language model for teaching science. In addition, our literacy approach draws on literature related to teaching science in first language contexts (e.g., Duit, 1999; Rivard & Cormier, 2008; Ross & Shuell, 1993) and the integration of literacy and science teaching (e.g., Cormier, Pruneau, & Rivard, 2010; Cormier, Pruneau, Rivard, & Blain, 2004; Lee & Fradd, 1996; Lemke, 1990; Norris & Phillips, 2003; Reddy, Jacobs, McCrohon, & Herrenkohl, 1998; Rivard & Straw, 2000; Thier & Daviss, 2002; Wellington & Osborne, 2001). Finally, theoretical and practical literature relating to scaffolding and literacy strategy instruction has also informed the development of our approach (e.g., Giasson, 1990, 2003; Nadon, 2002; Trehearne, 2006). We describe in detail the experimental treatment in practical terms. A more detailed theoretical description of our literacy approach appears in Cormier and Turnbull (2009).
The Literacy approach was implemented in the experimental class in five distinct phases. First, the research assistant assessed the students’ prior knowledge and preconceived notions of earthquakes and volcanoes by using two activities: She asked students to write and/or to draw their understandings of these phenomena (in a personal journal), and then posed a series of true/false questions. Students responded to the questions individually and then compared their answers with their peers in small discussion groups. Reading these questions also introduced students to language structures and vocabulary related to the unit of study. Students negotiated meaning and the answers to the questions in their groups, in which ever language they chose. Students also solicited help from the research assistant to understand the questions. The assistant provided some assistance, especially with isolated terms in L1. However, the assistant intentionally did not provide students with the correct responses to questions, and as a result, created interest and need for further information that would come in the next phase of the intervention.
During Phase two, the assistant led students through a shared writing activity using large flip chart paper. Together, they identified questions that had emerged from Phase one of the unit. Playing the role of secretary and guide, the assistant co-constructed the text with students’ and wrote it on the large paper. Throughout the shared writing process, the assistant drew students’ attention to important, useful and difficult language structures and vocabulary. She also highlighted the cognitive writing process as the shared writing proceeded (Nadon, 2002). This phase also allowed the assistant to identify concepts that had not emerged during the previous phase, but that were included in Ministerial outcomes. The assistant then modeled how to reflect and ask questions throughout the process and provided hints and suggestions to get answers to these questions.
During the third phase of the approach, students were engaged in the following scientific and language-based activities: Students studied examples of volcanic and seismic activity in a variety of pictures and videos and then shared their observations with peers in small groups and wrote about their observations in a personal journal; reading strategy instruction related to establishing a reason for reading and thinking aloud during reading; reading various texts (the textbook, reference books provided by the assistant, Internet) to practice these reading strategies and to locate responses to questions that had emerged in Phase Two of the approach;4 using shared and guided writing strategies (Nadon, 2002), students wrote responses to their questions after reading the texts; students did shared reading (Trehearne, 2006), in pairs, of a text about tectonic plates.
During the fourth phase, students divided into two groups; each group co-wrote an information text that summarized what they had learned during the unit. Together, but with guidance from the research assistant, they decided the plan and flow of the collective text and divided up the writing and illustration tasks. Each group subdivided into six smaller groups and each wrote and illustrated one chapter of the information text. During the writing process, the research assistant provided support with both language and science.
The fifth and final phase of the approach included a publication and public reading of the texts. This shared reading activity was set up in a carrousel format; students circulated around the room and read their texts to one group at a time. In the end, each group read its text six times.
Intervention in the Control Class.
To ensure that the teaching approach implemented in the control class would unfold according to what the Ministry and School District documents recommended, the research assistant planned the teaching unit for the control class in close collaboration with the regular classroom teacher. The unit used with the control class was organized around the following three teaching strategies: language simplification, demonstrations and Question Response Evaluation (QRE)-type discussion (teacher-led question, student response, teacher response and assessment; Lemke, 1990).
The assistant controlled the QRE discussions and then explained the concepts. Students wrote notes from the board at each session; the assistant created simplified notes based on the information contained in the textbook which was considered too difficult for students to read. The textbook was consulted only for viewing pictures of volcanoes, earthquakes and tectonic plates. Once all of the content from the textbook had been imparted to students as notes to copy from the board, students were asked to study for a quiz consisting of true and false questions and a cloze test.
The remainder of the unit engaged students in a research project. Students worked in pairs to create a text, in the format of their choice, that featured a well-known volcano eruption or earthquake disaster. Students were asked to identify and describe the disaster, locate it on a world map and provide five interesting facts about the disaster. Students completed their research using library books and the Internet to locate information. The unit culminated with an oral presentation (mainly, they read their notes) of the projects in front of the class.
Data Collection
Participants in both groups completed a written task (20 minutes) and an individual semi-structured interview (15 minutes) before and after the pedagogical intervention. Orally, they responded to questions, normally asked in French,5 such as: “Pouvez-vous décrire ce qu’est un volcan (et un séisme)? (‘Can you describe a volcano? An earthquake?’) Qu’est-ce qui cause un volcan (et un séisme)? (‘What causes a volcano? An earthquake?’) Où se produisent les volcans (et les séismes) et pourquoi? (‘Where do volcanoes and earthquakes occur and why?’.” The written task consisted of three questions: “Décrivez, dans tes propres mots, un volcan, un séisme et les plaques tectoniques. Tu peux utiliser des dessins aussi, si tu veux. ‘Describe in your own words, and use pictures if you want, a volcano, an earthquake and tectonic plates’.” During the oral interview, the assistant probed students to provide more information when she sensed that the student knew more but had trouble articulating the information. She also informed the students that they could use English when blocked due to limitations in French for oral and written tasks. All students in both classes completed the written task, while a subset of students participated in the oral interview (17 students from the control group, 16 in the experimental class). The researchers chose these participants randomly from the large group of students.6
Data Analysis
Analysis of Oral Interview Data.
Each oral interview (before and after the intervention) was analysed using the turn as unit of analysis. Each turn was first coded as French only, English only or as a codeswitch (e.g., Il a tremblement de terre sous l’eau et shake et cause a tsumani). Then, percentages of French only, English only and codeswitched turns were calculated in relation to the total number of turns for each student. After this first analysis was completed, a second coding for each turn was completed to assess the level of complexity of the content in each turn. Inspired by Bloom's taxonomy of complexity (e.g., Bloom, 1980), we designed a content complexity scale ranging from 0 to 7 (see Appendix).7 For instance, when a student simply repeated the interviewers’ words, or gave a simple yes or no answer, a score of 0 was assigned. A simple description, containing only one unit of content received a score of 1. To obtain a score of 4, the turn included a more detailed description, and had to include a comparison or an explanation of cause and effect. After coding each speaking turn, the complexity average for each interview was calculated. As a result, for each interview, we identified the percentage of turns coded as English only, French only, and codeswitches. We also calculated an average complexity for each individual interview. Two judges coded each turn independently; interrater agreement level was 92%. The link between language use and lexical complexity was explored through correspondence analyses (Benzécri, 1992) and chi-square analyses. Because of the non-quadricity of the contingency tables, we used the likelihood ratio instead of Pearson's test of independence (Sheskin, 2007). Effect size was estimated using Cramer's V (Sheskin, 2007).
Analysis of Written Tasks.
The students’ written production in French (in the written tasks) was assessed in a variety of ways. Our analyses focused on typically problematic areas for French immersion students, as discussed in the research literature (e.g., grammatical and lexical accuracy, richness of vocabulary, verb choice and conjugation). First we counted the total number of words written in each task.8 Second, we identified the number of words written in French and in English. We were interested not only in the overall length of the students’ texts but also the percentage of the texts they were able to write in French. Third, we did an analysis of lexical, spelling and grammatical errors to assess whether students’ written production became more or less accurate over time. Fourth, we assessed the lexical richness evident in the students’ texts. Finally, we analysed the type and quality of verb conjugations in the students’ texts. Correlation between these four continuous variables was estimated with Spearman's rank-order correlation coefficient (Sheskin, 2007).
Assessment of Science Knowledge.
To assess students’ content knowledge in science, we examined data from students’ drawings, written texts and transcripts from the oral interview. A score was assigned using two rubrics (one for volcanoes and the other for earthquakes) that we created by drawing on the Ministerial curriculum documents and the prescribed textbook.9 These rubrics assessed students’ preconceived notions of volcanoes and earthquakes and allowed us to assess growth in their understanding over the course of the unit. For the purpose of assessing students’ science knowledge, we considered utterances and written responses no matter whether they were completed in French, English or a mixture of both languages.10 Correlation between scores on the science tasks and linguistic variables was estimated with Spearman's rank-order correlation coefficient (Sheshkin, 2007).
Results
A detailed examination of the language and science results is found in Cormier and Turnbull (2009). Overall, the experimental group made more progress than the control group in science and in French. The students in the experimental group wrote longer texts, and more of these texts were written in French with fewer errors. Although both groups made small but insignificant gains in their knowledge of volcanoes, the experimental group made significantly more progress than students from the control group when it came to knowledge of earthquakes.
Building on McMillan and Turnbull (2009), the focus of this paper is to closely examine L1 and L2 interactions during the oral interviews as students talked about complex science content. Since the results in Cormier and Turnbull (2009) suggested that our experimental literacy approach lead to better results in science and in written French, we posed the following hypotheses for this article:
- 1
During the initial oral interviews, that both groups would be similar, and rely mainly on use of English. During these initial interviews, use of French would be associated with lower levels of complexity, and use of English or codeswitches would be associated with higher levels of complexity.
- 2
During final interviews, the experimental group would show greater complexity in their oral responses, but would require more use of English, which would manifest itself through more turns in English and more codeswitches. Therefore:
- a.
Overall, for both groups, the turns coded “English” or “codeswitch” would be coded at a higher level of complexity;
- b.
The experimental group would have more turns that would be coded “English” or “codeswitch”;
- c.
The greater use of “English” turns or “codeswitches” would be positively correlated to increase in complexity, better results in written French and better results in science knowledge.
VERIFYING OUR HYPOTHESES
Our first aim was to verify how both groups compared regarding choice of language use (French, English or CS) during oral interviews and how those choices were related to levels of complexity in the statements, as measured by each turn. A multiple correspondence analysis (Benzécri, 1992; Crucianu, Asselin de Beauville, & Boné, 2004) of group (control and experimental at both time 1 and time 2), language (English, French, and codeswitch), and level of complexity (0 through 6, 6 being the most complex) seems to suggest a shift from exclusive use of English at time 1 toward a more frequent use of codeswitches and French at time 2 for both groups. However, the move away from the exclusive use of English seems to correspond to a loss in complexity (Figure 1). This analysis demonstrates that both groups used less English during the final interviews. However, the French utterances reflected lower complexity levels. The CS turns are associated with a higher level of complexity while English turns show the most complex levels.
Joint Plot of Category Points
This two-dimensional model would account for 88.4% of the total variance (48.8% for dimension 1, 39.6% for dimension 2). Total inertia reaches 2.653 (1.465 for dimension 1, 1.187 for dimension 2). Individual contribution of each variable to both dimensions is presented in Table 1.
| Dimension | Mean | ||
|---|---|---|---|
| 1 | 2 | ||
| Level | .729 | .317 | .523 |
| Language | .726 | .537 | .632 |
| Group | .010 | .333 | .171 |
| Active Total | 1.465 | 1.187 | 1.326 |
| % of Variance | 48.849 | 39.569 | 44.209 |
Measures of association seem to confirm the link between language use and level of complexity. The likelihood ratio chi-square11 is statistically significant (L2[12]= 548.325, p= 0.000, n= 2,114) and the effect size, as estimated with Cramer's V(0.338), suggests a moderate effect according to Cohen (1988). Interpretation of the standardized residuals in Table 2 shows an association of higher levels of complexity with utterances in English; while level of complexity would decrease the more French would be present in textual units of analysis, first with codeswitches, then with exclusive use of L2. Our first hypothesis, therefore, was confirmed. Both groups were quite similar in the beginning and complexity of utterances was associated with English while French attempts were less complex. Measures of association seem to confirm the link between language use and level of complexity. The likelihood ratio chi-square11 is statistically significant (L2[12]= 548.325, p= 0.000, n= 2,114) and the effect size, as estimated with Cramer's V (0.338), suggests a moderate effect according to Cohen (1988). Interpretation of the standardized residuals in Table 2 shows an association of higher levels of complexity with the exclusive use of English; while level of complexity would decrease the more French would be present in textual units of analysis, first with codeswitches, then with exclusive use of L2.
| Level | Total | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 1 | 2 | 3 | 4 | 5 | 6 | ||||
| Language | English | Count | 373 | 215 | 200 | 89 | 59 | 17 | 11 | 964 |
| Std. Residual | −1.1 | .3 | −1.5 | 1.8 | 1.9 | 2.0 | 2.1 | |||
| Codeswitch | Count | 33 | 106 | 207 | 54 | 39 | 6 | 2 | 447 | |
| Std. Residual | −11.1 | .8 | 10.3 | 3.4 | 3.8 | .5 | −.5 | |||
| French | Count | 460 | 142 | 79 | 19 | 3 | 0 | 0 | 703 | |
| Std. Residual | 10.1 | −1.0 | −6.5 | −4.8 | −5.3 | −2.8 | −2.1 | |||
| Total | Count | 866 | 463 | 486 | 162 | 101 | 23 | 13 | 2114 | |
| Level | Total | |||||||||
| 0 | 1 | 2 | 3 | 4 | 5 | 6 | ||||
| Language | English | Count | 373 | 215 | 200 | 89 | 59 | 17 | 11 | 964 |
| Std. Residual | −1.1 | .3 | −1.5 | 1.8 | 1.9 | 2.0 | 2.1 | |||
| Codeswitch | Count | 33 | 106 | 207 | 54 | 39 | 6 | 2 | 447 | |
| Std. Residual | −11.1 | .8 | 10.3 | 3.4 | 3.8 | .5 | −.5 | |||
| French | Count | 460 | 142 | 79 | 19 | 3 | 0 | 0 | 703 | |
| Std. Residual | 10.1 | −1.0 | −6.5 | −4.8 | −5.3 | −2.8 | −2.1 | |||
| Total | Count | 866 | 463 | 486 | 162 | 101 | 23 | 13 | 2114 | |
Changes in language use between time 1 and time 2 would be more pronounced for the experimental group than for the control group. As shown in Figure 2, both groups used less English, more codeswitches and more French at time two. However, this trend is stronger for the experimental group in all cases.
Changes in Language Use × Group
The likelihood ratio chi-square shows a statistically significant (L2[2]= 7.338, p= 0.026) but negligible association (V= 0.083) between language use and time for the control group (n= 1,054). Standardized residuals reveal a slight underrepresentation of exclusive use of L1 at time 2 (Table 3). Therefore, it seems that the control group used slightly less English during the final interviews.
| Time | Total | ||||
|---|---|---|---|---|---|
| Control 1 | Control 2 | ||||
| Language | English | % within group | 45.0% | 36.4% | 42.0% |
| Std. Residual | 1.2 | −1.7 | |||
| Codeswitch | % within group | 22.5% | 25.0% | 23.3% | |
| Std. Residual | −.5 | .7 | |||
| French | % within group | 32.6% | 38.6% | 34.6% | |
| Std. Residual | −.9 | 1.3 | |||
For the experimental group (n= 1,060), the effect is significant (L2[2]= 19.315, p= 0.000) and would be considered small, but not negligible (V = 0.135). Analysis of standardized residuals suggests a move from exclusive use of L1 to codeswitches between time 1 and time 2 (Table 4). The experimental group, therefore, showed an increased use of codeswitches during the final interviews.
| Time | Total | ||||
|---|---|---|---|---|---|
| Exp. 1 | Exp. 2 | ||||
| Language | English | % within group | 53.5% | 39.8% | 49.2% |
| Std. Residual | 1.7 | −2.5 | |||
| Codeswitch | % within group | 16.3% | 24.6% | 19.0% | |
| Std. Residual | −1.6 | 2.4 | |||
| French | % within group | 30.2% | 35.6% | 31.9% | |
| Std. Residual | −.8 | 1.2 | |||
As speculated in our second hypothesis, changes in level of complexity are greater for the experimental group than for the control group during the final interviews. Figure 3 shows, in fact, how complexity of turns assessed at level 4 and 5 increased for the experimental group while there was a decrease in the number of complex turns for the control group. Table 5 (control group) and Table 6 (experimental group) show the strength of the association between initial and final interviews and level of complexity for the utterances. For the control group (n= 1,054), the chi-square doesn't reach significance (L2[6]= 9.348, p= 0.155) and the strength of the association between time and level of complexity would be negligible (V = 0.082). Standardized residuals analysis barely reflects any noticeable discrepancy between observed and expected frequencies (Table 5). For the experimental group (n= 1,060), measures of association describe a small (V= 0.111) but still insignificant relationship (L2[6]= 12.527, p= 0.051). Standardized residuals suggest a modest improvement in level of complexity (Table 6).
Changes in Level × Group
| Time | Total | ||||
|---|---|---|---|---|---|
| Control | Control | ||||
| Level | 0 | % within group | 42.8% | 42.5% | 42.7% |
| Std. Residual | .0 | −.1 | |||
| 1 | % within group | 18.6% | 23.1% | 20.1% | |
| Std. Residual | −.9 | 1.2 | |||
| 2 | % within group | 25.4% | 22.5% | 24.4% | |
| Std. Residual | .5 | −.7 | |||
| 3 | % within group | 8.1% | 8.3% | 8.2% | |
| Std. Residual | −.1 | .1 | |||
| 4 | % within group | 3.3% | 3.1% | 3.2% | |
| Std. Residual | .1 | −.2 | |||
| 5 | % within group | .9% | .6% | .8% | |
| Std. Residual | .3 | −.4 | |||
| 6 | % within group | 1.0% | .0% | .7% | |
| Std. Residual | 1.1 | −1.5 | |||
| Time | Total | ||||
|---|---|---|---|---|---|
| Exp. 1 | Exp. 2 | ||||
| Level | 0 | % within group | 40.7% | 36.2% | 39.2% |
| Std. Residual | .6 | −.9 | |||
| 1 | % within group | 22.7% | 25.8% | 23.7% | |
| Std. Residual | −.6 | .8 | |||
| 2 | % within group | 22.8% | 19.0% | 21.6% | |
| Std. Residual | .7 | −1.0 | |||
| 3 | % within group | 7.2% | 7.1% | 7.2% | |
| Std. Residual | 0 | 0 | |||
| 4 | % within group | 4.7% | 9.8% | 6.3% | |
| Std. Residual | −1.7 | 2.5 | |||
| 5 | % within group | 1.4% | 1.5% | 1.4% | |
| Std. Residual | −.1 | .1 | |||
| 6 | % within group | .6% | .6% | .6% | |
| Std. Residual | 0 | .1 | |||
To further our analysis, and to better comprehend how the changes in levels of complexity occurred, we then examined changes in level of complexity separately for turns coded as English, French, and codeswitches. Figure 4 shows changes in the complexity level for turns in English. Somewhat surprisingly, as also shown in Table 7, the control group (n= 443) seems to show a small (V= 0.151) but significant (L2[6]= 12.817, p= 0.046) decrease in complexity level for English speech units (Table 7).
Changes in Level (English) × Group
| Time | Total | ||||
|---|---|---|---|---|---|
| Control 1 | Control 2 | ||||
| Level | 0 | % within group | 41.7% | 45.0% | 42.7% |
| Std. Residual | −.3 | .4 | |||
| 1 | % within group | 21.2% | 28.2% | 23.3% | |
| Std. Residual | −.8 | 1.2 | |||
| 2 | % within group | 21.8% | 20.6% | 21.4% | |
| Std. Residual | .1 | −.2 | |||
| 3 | % within group | 9.0% | 4.6% | 7.7% | |
| Std. Residual | .8 | −1.3 | |||
| 4 | % within group | 3.8% | .8% | 2.9% | |
| Std. Residual | .9 | −1.5 | |||
| 5 | % within group | .6% | .8% | .7% | |
| Std. Residual | −.1 | .1 | |||
| 6 | % within group | 1.9% | 0% | 1.4% | |
| Std. Residual | .9 | −1.3 | |||
In Figure 4, we notice the increase in complexity, especially at level 4, for the experimental group. Table 8 demonstrates that the experimental group (n= 521), exhibited a small (V= 0.138) but insignificant (L2[6]= 9.537, p= 0.146) increase in level of complexity from time 1 to time 2.
| Time | Total | ||||
|---|---|---|---|---|---|
| Exp. 1 | Exp. 2 | ||||
| Level | 0 | % within group | 36.7% | 31.3% | 35.3% |
| Std. Residual | .5 | −.8 | |||
| 1 | % within group | 19.6% | 26.9% | 21.5% | |
| Std. Residual | −.8 | 1.3 | |||
| 2 | % within group | 21.4% | 16.4% | 20.2% | |
| Std. Residual | .6 | −1.0 | |||
| 3 | % within group | 11.4% | 8.2% | 10.6% | |
| Std. Residual | .5 | −.8 | |||
| 4 | % within group | 7.2% | 13.4% | 8.8% | |
| Std. Residual | −1.1 | 1.8 | |||
| 5 | % within group | 2.6% | 3.0% | 2.7% | |
| Std. Residual | −.1 | .2 | |||
| 6 | % within group | 1.0% | .7% | 1.0% | |
| Std. Residual | .1 | −.3 | |||
Figure 5 shows that changes in level of complexity for codeswitches were greater for the experimental group than for the control group. Table 9 depicts that the control group (n= 246) exhibited a small (V= 0.144) but not statistically significant (L2[6]= 5.665, p= 0.462) association between time and level of complexity for codeswitches.
Changes in Level (Code Switch) × Group
| Time | Total | ||||
|---|---|---|---|---|---|
| Control 1 | Control 2 | ||||
| Level | 0 | % within group | 8.3% | 3.3% | 6.5% |
| Std. Residual | .9 | −1.2 | |||
| 1 | % within group | 17.9% | 17.8% | 17.9% | |
| Std. Residual | 0 | 0 | |||
| 2 | % within group | 50.0% | 48.9% | 49.6% | |
| Std. Residual | .1 | −.1 | |||
| 3 | % within group | 13.5% | 18.9% | 15.4% | |
| Std. Residual | −.6 | .8 | |||
| 4 | % within group | 7.1% | 10.0% | 8.1% | |
| Std. Residual | −.5 | .6 | |||
| 5 | % within group | 2.6% | 1.1% | 2.0% | |
| Std. Residual | .5 | −.6 | |||
| 6 | % within group | .6% | 0 | .4% | |
| Std. Residual | .5 | −.6 | |||
Table 10, however, shows how the experimental group (n= 201), develops a small (V= 0.264) but significant (L2[6]= 14.555, p= 0.024) increase in level of complexity from time 1 to time 2.
| Time | Total | ||||
|---|---|---|---|---|---|
| Exp. 1 | Exp. 2 | ||||
| Level | 0 | % within group | 9.3% | 7.2% | 8.5% |
| Std. Residual | .3 | −.4 | |||
| 1 | % within group | 33.9% | 26.5% | 30.8% | |
| Std. Residual | .6 | −.7 | |||
| 2 | % within group | 46.6% | 36.1% | 42.3% | |
| Std. Residual | .7 | −.9 | |||
| 3 | % within group | 5.1% | 12.0% | 8.0% | |
| Std. Residual | −1.1 | 1.3 | |||
| 4 | % within group | 5.1% | 15.7% | 9.5% | |
| Std. Residual | −1.5 | 1.8 | |||
| 5 | % within group | 0 | 1.2% | .5% | |
| Std. Residual | −.8 | .9 | |||
| 6 | % within group | 0 | 1.2% | .5% | |
| Std. Residual | −.8 | .9 | |||
As expected in our hypothesis, there were virtually no changes in the level of complexity for either group, for turns that were coded “French.” In fact, changes in the level of complexity of French speech units were not significant for either the control group (L2[4]= 7.682, p= 0.104, n= 365) or the experimental group (L2[4]= 6.036, p= 0.196, n= 338).
Use of English, French, and Codeswitches × Results in French and Science
Analyses concerning levels of complexity in turns coded in either English, Codeswitch, and French were performed on the conversational turns as a unit of analysis. For the following analyses on correlations between the use results in French and science as compared to language use as well as levels of complexity as correlated to science results, it is necessary to switch our unit of analysis. Thus, in the analyses that follow, we focus on students as our unit of analysis, which explains the change in sample size. To estimate the correlation between the proportion of turns in English, French, and use of codeswitches, and results in French and science, we used Spearman's rank-order correlation coefficient (Sheshkin, 2007). We present the correlation between language use and results in science and French in Table 11 and the correlations between levels of complexity and science results in Table 12.
| English | French | Codeswitches | ||||||
|---|---|---|---|---|---|---|---|---|
| ρ | p | ρ | p | ρ | p | |||
| Ex | T1 | Words | −0.233 | 0.385 | −0.003 | 0.991 | 0.340 | 0.197 |
| French Words | −0.455 | 0.077 | 0.386 | 0.140 | 0.441 | 0.088 | ||
| Error/Word | −0.053 | 0.845 | 0.049 | 0.858 | 0.068 | 0.802 | ||
| Science | 0.280 | 0.293 | −0.274 | 0.304 | −0.328 | 0.215 | ||
| T2 | Words | −0.242 | 0.426 | 0.374 | 0.208 | 0.298 | 0.324 | |
| French Words | −0.314 | 0.296 | 0.421 | 0.151 | 0.417 | 0.157 | ||
| Error/Word | 0.576 | 0.039 | −0.628 | 0.022 | −0.345 | 0.249 | ||
| Science | 0.239 | 0.391 | −0.053 | 0.851 | −0.474 | 0.074 | ||
| Co | T1 | Words | −0.321 | 0.209 | 0.447 | 0.072 | 0.191 | 0.464 |
| French Words | −0.503 | 0.040 | 0.326 | 0.202 | 0.403 | 0.108 | ||
| Error/Word | 0.305 | 0.233 | −0.140 | 0.593 | −0.233 | 0.368 | ||
| Science | −0.230 | 0.375 | 0.104 | 0.692 | 0.468 | 0.058 | ||
| T2 | Words | −0.278 | 0.297 | 0.186 | 0.490 | 0.287 | 0.282 | |
| French Words | −0.201 | 0.455 | 0.161 | 0.553 | 0.180 | 0.505 | ||
| Error/Word | 0.520 | 0.039 | −0.623 | 0.010 | −0.158 | 0.559 | ||
| Science | −0.307 | 0.230 | 0.177 | 0.498 | 0.606 | 0.010 | ||
| All | T1 | Words | −0.329 | 0.061 | 0.230 | 0.198 | 0.353 | 0.044 |
| French Words | −0.508 | 0.003 | 0.378 | 0.030 | 0.477 | 0.005 | ||
| Error/Word | 0.067 | 0.713 | −0.075 | 0.679 | −0.068 | 0.706 | ||
| Science | 0.045 | 0.806 | −0.082 | 0.650 | 0.062 | 0.733 | ||
| T2 | Words | −0.163 | 0.397 | 0.253 | 0.185 | 0.181 | 0.347 | |
| French Words | −0.279 | 0.143 | 0.337 | 0.073 | 0.245 | 0.200 | ||
| Error/Word | 0.410 | 0.027 | −0.541 | 0.002 | −0.186 | 0.334 | ||
| Science | 0.018 | 0.923 | 0.089 | 0.627 | 0.010 | 0.958 | ||
| n | ρ | p | ||
|---|---|---|---|---|
| Experimental | T1 | 16 | 0.531 | 0.016 |
| T2 | 15 | 0.731 | 0.002 | |
| Control | T1 | 17 | 0.252 | 0.329 |
| T2 | 17 | 0.668 | 0.003 | |
| All | T1 | 33 | 0.352 | 0.044 |
| T2 | 32 | 0.751 | 0.000 |
As we expected, there were no meaningful correlations between language use and results in either French or science at time 1 for either group taken individually. However, using the complete sample, statistically significant correlations were found between the use of codeswitching and the total number of words (ρ= 0.353, p= 0.044) and French words (ρ= 0.477, p= 0.005).
At time 2, the ratio of error/word was positively correlated with the use of English (ρ= 0.410, p= 0.027), and negatively correlated with the use of French (ρ=−0.541, p= 0.002). These correlations held for both the experimental and control groups. In addition, the control group showed a positive correlation between the use of codeswitching and achievement in science (ρ= 0.606, p= 0.010). This correlation goes against the trend observed for the experimental group, for which those correlations are negative, albeit not significantly so.
Level of Complexity × Results in Science
Finally, the level of complexity of propositions (Table 12) was directly and significantly correlated for all groups at all times, with the single exception of the control group at time 1, for which the correlation was also positive, but failed to achieve significance.
DISCUSSION
As stated at the beginning of this article, teaching complex subjects such as science in French immersion, especially when students’ French proficiency is limited, is often challenging for students and teachers. A subject as intricate as science requires higher order cognitive thinking to unpack meaning and make sense of concepts and content. Typically, when reflecting about science, cognitive tools such as diagrams, equations or language can be useful for scaffolding thought processes. Imagine learning the circulation system of the human body without a diagram, or language!
Following Vygotsky's theory of mind (1980), the results from this study show that language acts as an important cognitive tool to help make sense of complex science content. The analyses presented in this article show that during the initial oral interviews, both groups relied mainly on English. During these initial interviews, use of French was linked with lower levels of complexity, and use of English or codeswitches was associated with higher levels of complexity. Indeed, analyses using data drawn from the whole sample showed positive and significant relationships between utterance complexity and use of English or codeswitching. During final interviews, the experimental group's oral output was more complex, but these students still needed English to manage and articulate this greater complexity. During final interviews, when students spoke French only, their utterances were less complex.
Our hypothesis that the greater number of turns coded as “English” or “codeswitches” would be positively correlated to an increase in complexity, better results in written French and better results in science knowledge was generally confirmed. Correlational analyses showed a positive and significant relationship between codeswitching and text length (number of words) and the total number of words written in French. Error rates in written French decreased for both groups as more English or codeswitching were evident in their oral production. While correlations between language use and science results were less clear, there was a positive and significant correlation between utterance complexity and results in science for both groups.
Although it is possible to argue that mixing oranges, apples and bananas, or results from very different types of data (oral and written test scores, oral discourse, qualitative analyses of written language) may be pushing the limits of statistical analyses, our results do provide significant quantitative support for what others have been demonstrating using qualitative data (e.g., Brooks & Donato, 1994; Swain & Lapkin, 2000; Turnbull, 2001).
CONCLUDING REMARKS
This study contributes to the literature grounded in sociocultural theory which investigates the use of L1 as a tool that L2 learners use to construct new meaning and understanding and that allows the learners to improve, not only in their French language skills but also in their thinking and problem-solving abilities in an immersion context. Although additional studies are clearly needed, both quantitative and mixed methods, with larger participant samples and in different second and foreign language contexts, the results from this study should hopefully provoke lively debate in professional and scholarly circles about the use of L1 in teaching content like science in a second language. We believe strongly that it is not responsible or pedagogically sound to avoid discussing the role of L1 in immersion teaching. However, we are not promoting an overuse of the L1 either. In fact, the pedagogical interventions used in this quasi-experimental study did not explicitly include teacher or student use of the L1—it just happened and was necessary for students as they navigated complex tasks and science content presented in French. We believe that the results from our study lend empirical support to Turnbull and Dailey-O’Cain's (2009, p. 183) definition of optimal L1 use in second and foreign language teaching:
Optimal first language use in communicative and immersion second and foreign language classrooms recognizes the benefits of the leaner's first language as a cognitive and meta-cognitive tool, as a strategic organizer, and as a scaffold for language development. In addition, the fi rst language helps learners navigate a bilingual identity and thereby learn to function as a bilingual. Neither the classroom teacher nor the second or foreign language learner becomes so dependent on the first language that neither can function without the first language. Optimal codeswitching practices will ultimately lead to enhanced language learning and the development of bilingual communicative practices. (p. 183)
Open and transparent discussion of codeswitching in immersion settings will help the field and immersion education move forward. Future research that further investigates the relationship between student learning and teacher and student codeswitching, or lack thereof, will as well.
NOTES
Appendix
Complexity Scale (adapted from Bloom, 1980)
Level 0:
The student does not respond to the question because he does not comprehend. The student repeats the interviewer's question, responds with « je ne sais pas (‘I don't know’) », or « je ne comprends pas (‘I don't understand’)», with yes or no, or he verifies whether he has understood.
Level 1: Simple description
The student provides a descriptive response with one detail that was not provided in the interviewer's question.
Level 2: Complex description
The student provides a descriptive response with at least two details not provided in the interviewer's question.
Level 3: Complex description and linkages
The student provides a descriptive response with at least two details not provided in the interviewer's question, and the student provides at least one link between the details.
Level 4: Simple Explanation
The student provides a descriptive response with at least two linked details not provided in the interviewer's question. A cause-and-effect or comparative explanation is provided.
Level 5: Complex Explanation
The student provides a descriptive response with at least two linked details not provided in the interviewer's question. Response goes beyond a simple cause-and-effect or comparative explanation.
Level 6: Complex Explanation and application to another context
The student provides a descriptive response with at least two linked details not provided in the interviewer's question. Response goes beyond a simple cause-and-effect or comparative explanation. Student explains how to apply the explanation to another context or situation.
Level 7: Assessment
The student provides a descriptive response with at least two linked details not provided in the interviewer's question. Response goes beyond a simple cause-and-effect or comparative explanation. Student explains how to apply the explanation to another context or situation.
The student provides an assessment, offers his opinion and/or draws a conclusion based on his analysis of the data.
