L3 English in the German secondary school context: longitudinal development of bilingual heritage speakers’ multilingual repertoire

ABSTRACT

A multilingual experience can be considered a significant asset. However, since the earliest studies in the field, research has reported mixed results regarding potential advantages such as increased cognitive ability and metalinguistic awareness. Moreover, studies investigating the influence of bilingualism/multilingualism on the acquisition of additional languages produced partially conflicting findings. This longitudinal study (three measurement time points within 1.5 years) investigates the impact of social, cognitive, and linguistic variables on English proficiency. We rely on data gathered from 374 unbalanced bilinguals (Russian-/Turkish-German), dominant in German, and 600 monolingual (German) students attending grades seven to ten in Germany. We use multiple linear regression models predicting English proficiency to track longitudinal changes. The results show that German reading proficiency and visual-spatial cognitive ability positively impact L2/L3 English C-test scores. In addition, the impact of heritage language proficiency is not statistically significant and language background did not bring about clear differences between the monolingual and bilingual cohorts per se. We conclude that there are differences between the Turkish-German students and their peers, but not between the monolinguals and Russian-German students. We discuss these findings in light of current teaching approaches prevailing in Germany's foreign language classrooms and perceptions of heritage languages in society.

KEYWORDS:

• Cognitive ability
• English as a foreign language
• heritage bilinguals in Germany
• language acquisition
• language interdependence
• longitudinal development

Introduction

The past years have produced a sizeable envelope of studies dedicated to illuminating the influence of heritage bilingualism on subsequent language learning. It was especially the field of third language acquisition in relation to English that partook in these developments (e.g. Agustín-Llach, 2019; Edele et al., 2018; Rauch et al., 2012; Rothman et al., 2019; Siemund & Lechner, 2015). Migration-induced demographic changes in Western Europe have produced large groups of heritage bilinguals of diverse linguistic provenance for whom the acquisition of English in the respective national education systems has the status of a chronologically third language (Siemund, 2023). With respect to Germany, for example, heritage bilinguals of Turkish, Russian, Polish, Arabic, Farsi, Pashtu, and many other languages represent approximately 50% of the adolescent school population (Gogolin, 2021). These bilinguals speak German as the national and educational language as well as a heritage language as their home language. Often, it is not clear which of these counts as the students’ first language, but they typically develop dominance in German (Hopp, 2019). Research has so far primarily focussed on comparisons of different bilingual groups with monolingual controls and, to a much lesser extent, on comparisons of different bilingual groups with one another (Rothman et al., 2019, 2023). In addition, there are hardly any longitudinal studies. The development of such multilinguals’ repertoires over time can largely be considered unchartered territory. The present contribution aims to fill this research gap by studying the longitudinal development of two bilingual heritage groups and a monolingually raised group in the context of Germany.

Previous studies

The successful acquisition of a heritage language has been a matter of inquiry in a growing body of literature over the last years. Speakers of various heritage languages, including but not limited to Arabic, Turkish, Russian, and German, have been examined in various European countries such as Denmark (e.g. Højen & Bleses, 2023), Finland (e.g. Ansó Ros et al., 2024), the Netherlands (e.g. Stoehr et al., 2018), and Spain (e.g. Vorobyeva et al., 2024). Overall, the majority of these studies have explored the factors affecting heritage language acquisition and the links between heritage and majority language acquisition. Considering that there are still relatively few studies examining how third language (L3) acquisition takes place in heritage speakers, examining heritage bilingualism in the context of L3 could serve as an exciting and important research avenue (Lloyd-Smith, 2023). In line with this view, the focus of attention of previous cross-sectional studies in Germany has been the impact of the educational language German and the respective heritage languages on English proficiency (Edele et al., 2018; Hopp et al., 2019; Klieme et al., 2006; Lorenz et al., 2020, 2021a; Maluch et al., 2015, 2016; Maluch & Kempert, 2019). In Germany, much as in other countries, English is the most important and compulsory foreign language that is introduced earliest. The general question is whether the previous bilingual experience impacts further language acquisition, and if so, in which direction. We here relate to Cummins (1976, 1979) ideas of an ‘underlying proficiency’ and his ‘linguistic interdependence hypothesis’. For example, Hesse et al. (2008) were among the first to point towards positive associations between heritage bilingualism and English acquisition. Maluch et al. (2015) and Maluch et al. (2016) equally report positive effects that, however, are contingent on specific heritage groups and more pronounced in younger age groups. The instructional language German proved the strongest predictor. While heritage language proficiency was only assessed through parental and/or student questionnaires in these studies, Edele et al. (2018) applied direct measures (reading and listening proficiency tests) in the heritage languages. This study found certain group-specific positive effects of bilingualism. Rauch et al. (2010) report no negative effects of Turkish-German heritage bilingualism on the acquisition of English, whereas Rauch et al. (2012) find positive effects amongst fully bi-literate students.

In relation to the longitudinal development of heritage bilinguals in Germany, the few existing studies report an increasing uniformity with advancing age respectively years of studying the foreign language English. Siemund and Lechner (2015) demonstrate such increasing uniformity for crosslinguistic influence, Maluch et al. (2016) for general achievement in English, and Lorenz et al. (2021a) for lexical transfer. Cross-linguistic influence including lexical transfer decreases over time in all groups, while general achievement increases. In other words, bilingual and monolingual student groups converge on a common norm, which conforms to the objective of current education policies. Berthele and Udry (2022) is a longitudinal study conducted in the German-speaking part of Switzerland where French is learnt as the first foreign language before English. The study is based on two measurement points. Proficiency in German and French at one point in time was used to predict proficiency in English one year after the first measurement time point. The main predictor variable proved cognitive ability (see also Berthele & Vanhove, 2020, p. 562; Lorenz et al., 2021b; Siemund et al., 2023, 2024). However, Berthele and Udry’s (2022) study did not monitor the students’ multilingual development proper. Hopp et al. (2019) followed 200 monolingual and bilingual heritage students from grades three to four in Germany. The bilingual students outperformed the monolingual students on measures of vocabulary and grammar, but this positive effect diminished with increasing age.

Overall, studies investigating the influence of bilingualism/multilingualism on the acquisition of additional languages produced partially conflicting findings. Cognitive ability and instructional languages have surfaced as strong and significant predictors of English achievement. The influence of the heritage languages appears to diminish over time. Hence, the present study aims to contribute to a growing body of literature on the impact of bilingualism on third language acquisition by employing a longitudinal design and considering a set of personal and contextual factors.

Research questions and hypotheses

The present study addresses the following research questions and, based on these, formulates three hypotheses. The first research question enquires about the personal, linguistic, and contextual variables that explain L2/L3 English proficiency in monolingual (German) and bilingual (Russian-/Turkish-German) students. These variables include cognitive ability, socio-economic status, German proficiency, language background, heritage language proficiency in the bilingual groups, gender, school type, and school year. The studies surveyed in the previous section identify these as (partially) significant factors and, therefore, their impact will be monitored here. Berthele and Udry (2022), for instance, establish cognitive ability as the strongest predictor for language achievement (see also Berthele & Vanhove, 2020). The second research question is interested in how the influence of these personal, linguistic, and contextual variables changes across monolingual and bilingual student groups over time. Regarding our hypotheses, we propose that (H1) cognitive ability remains a strong and significant predictor over time, (H2) the impact of heritage language proficiency decreases and the influence of German proficiency increases over time, and (H3) monolingual and bilingual students converge from measurement time points one to three due to the accumulated time of learning English in the German secondary school context. The second hypothesis, in fact, conflates two hypotheses, but we do not wish to disentangle them, as they belong together. We derive H2 from the observation made in previous studies that the influence of heritage languages on English achievement is typically stronger at a younger age. The influence of German proficiency is supposed to become stronger over time, since it serves as the main instructional language. H3 reflects the aspirations of the school system as they are encapsulated in the relevant curriculum descriptions.

Methodology

The following subsections provide detailed information about the sample, the instruments used, the data collection process, as well as the screening, analysis, and modelling of the data.

The sample

The data used for the current study is drawn from the MEZ project (Mehrsprachigkeitsentwicklung im Zeitverlauf [MEZ], 2014–2019; Gogolin et al., 2017), a large-scale, longitudinal project on multilingual development in the German secondary school context. The project compared monolingual German secondary school students with their bilingual peers (Russian-German, Turkish-German). Data collection occurred between 2016 and 2018 at four measurement time points (MTP) and considered two age cohorts. Cohort 1 started out in school year 7 and Cohort 2 in school year 9 (see Figure 1). The study reported here relies on the data collected at the first three MTPs.

Figure 1. The four MTPs of the MEZ project.

Table 1 presents further information about the data collection periods and the number of students participating at each MTP. Although more students participated during the project (overall sample), only a reduced number (general sample, bilingual sample) could be considered due to participants dropping out of the project or leaving school as well as inconsistent or missing data. The general sample (monolingual and bilingual students) consisted of 600 monolingual German, 160 Russian-German, and 214 Turkish-German students. The bilingual sample comprised 123 Russian-German and 178 Turkish-German students. Some bilinguals had incomplete scores for the heritage language test (reading comprehension test) and had to be excluded from this part of the analysis.

Table 1. The number of students at all four MTPs.

	MTP1	MTP2	MTP3	MTP4
Overall sample	1818	1568	1698	1224
General sample	974	974	974	–
Bilingual sample	301	301	301	–

All students received education in Germany. This means that for the bilingual participants, the heritage language (Russian/Turkish) is primarily used at home and might be regarded as their weaker language, while German ranks as their majority language (Gogolin, 2021; Lorenz et al., 2020). Furthermore, the additional language English is the second language (L2) for the monolingual German cohort, and it is the third language (L3) for the bilingual Russian-German and Turkish-German cohort, for they had already acquired two languages before learning English. The students are beginner to intermediate English language learners, having typically begun English instruction in primary school (either year one or three). Their English proficiency was not tested prior to the study but only as part of the study (see below).

Instruments

The present study utilised a reading comprehension and fluency test in German and, if applicable, in the heritage languages Russian and Turkish, a C-test in English, and a cognitive ability test. Additionally, information about the socio-economic status, type of school, gender, and age of the students was gathered from parental and student questionnaires.

Reading comprehension and fluency test

Students were given a reading comprehension and fluency test (LGVT – Lesegeschwindigkeits- und Verständnistest, Schneider et al., 2016) in German and the heritage languages Russian/Turkish. The test was adapted for both heritage languages, for it was originally only available in German. Despite not being an exact translation of the German measure, the Russian¹ and Turkish versions employed a similar genre and subject matter. Each test consisted of a text with 47 gaps that had to be filled in using one of the three options within a time limit of six minutes. We followed the testing and scoring procedure recommended by Schneider et al. (2016): a student received two points if they selected the correct response, zero points if they were unable to respond, and a negative point (−1) if they selected the incorrect one. The measurement scale ranged between 47 and +94 points. The test used two components, reading comprehension (fill the gap) and fluency (reading speed based on the total number of words read). We here relied on reading comprehension only, as reading speed is also implicitly included in that measure. We accordingly refer to the test as reading comprehension in what follows. The reliability scores for the German reading comprehension test for each measurement point are as follows; .82 for MTP1, .87 for MTP2, and .90 for MTP3. The reliability scores for the Russian reading comprehension test for each measurement point are as follows; .86 (Cyrillic version) and .15 (the Latin alphabet version) for MTP1, .85 (Cyrillic version) and .15 (the Latin alphabet version) for MTP2, and .84 (Cyrillic version) and .15 (the Latin alphabet version) for MTP3. Finally, for Turkish reading comprehension, the reliability scores are as follows: .64 for MTP1, .72 for MTP2, and .64 for MTP3 (MEZ, 2020a, 2020b, 2020c).

C-tests

English proficiency of the students (lexical and grammatical competence in English) was measured through a set of C-tests, different at each MTP, modelled after those employed in the DESI study (Klieme et al., 2006). The procedure was the same at all MTPs: the four texts included a total of 90 end-clipped words. The first and last sentences were provided in their entirety. The remaining sentences required the students to complete all end-clipped words within 20 minutes. The following example illustrates this task: ‘As they wal___ [-ked] through the zoo th___ [-ey] came to a la___ [-rge] cage with monkeys’. The students were awarded one point if they gave the right response and zero points if they only answered partially, incorrectly, or did not fill in the gap. The internal consistency of the four subtests based on Cronbach's alpha was sufficiently high. The overall Cronbach alpha values across the four subtests in each C-test at three different MTPs are as follows: C-test 1 = .93, C-test 2 = .96, and C-test 3 = .95 (MEZ, 2020a, 2020b, 2020c).

Cognitive ability test

A subset of the KFT 4-12+ R (Heller & Perleth, 2000), a standardised German cognitive ability test that focuses on students’ visual-spatial (non-verbal) skills from grades 4 to 12, was used to assess students’ cognitive ability. It was only assessed during MTP1 and the respective score is used for the two remaining MTPs. The subset contained 25 test items and included sequences of shapes or figures that had to be continued by selecting one out of five options. Accordingly, students earned one point for every correct logical extension. While the assessment scale ranged from 0 to 25, the internal consistency of the measure was sufficiently high, ranging between .85 and .90 (MEZ, 2020a).

Demographic information

Demographic information about participating students was collected by means of parental and student questionnaires to assess and control for a set of extra-linguistic and personal variables, including gender, school year, and type of school (Gymnasium vs other) since these variables might influence the language acquisition process. Gender has been inconsistently shown to impact foreign language acquisition (e.g. Berthele & Udry, 2021; Klieme et al., 2006), which justifies it being added as a control variable in the current study. Moreover, since previous research (Edele et al., 2018; Lorenz et al., 2023) suggested that academic school track may impact students’ academic attainment and foreign language proficiency considerably, school type also featured as a background variable in the current study. Regarding school type, the higher academic secondary track (Gymnasium), preparing students for university admission, and the lower academic secondary school tracks (e.g. Realschule, Hauptschule) were considered.

HISEI-Socio-Economic Index

Socio-economic status has been shown to affect proficiency in a foreign language (e.g. Maluch et al., 2015, 2016), which is why we controlled for it in the following analysis. Socio-economic status was operationalised based on the Highest International Socio-Economic Index (HISEI, Ganzeboom et al., 1992; International Labour Office, 2012) and estimated during MTP1. The corresponding value features in all three MTPs. This index was calculated by using information about the caretakers’professions as gathered through a parental questionnaire. Each student's socio-economic status was represented by the highest value assigned to their family, a lower number denoting a relatively lower socio-economic status. These HISEI values ranged from 16 to 90.

Data screening

The data were initially checked for validity, reliability, and missing cases. The descriptive statistics can be found in Table 2. The instruments yielded values on varying scales, hence, the data were normalised to adjust these to a notionally common scale, i.e. to a value between 0 and 1 for all the participants. The normalisation procedures were carried out using Microsoft Excel. Moreover, a set of ANOVAs and post-hoc Tukey HSD tests indicated that at the starting point, MTP1, there are no statistically significant differences between the three groups concerning English C-test scores. For cognitive ability, the Turkish-German bilingual group is statistically different from the other two groups, whereas the German monolingual group has statistically significantly higher socio-economic status values than the bilingual groups. Finally, the German monolinguals obtain significantly higher scores in the German reading comprehension test.

Table 2. Descriptive statistics (absolute values) at three MTPs (n = 974).

		German monolingual (n = 600)		Russian-German (n = 160)		Turkish-German (n = 214)		ANOVA
		Mean	SD	Mean	SD	Mean	SD	p
MTP1	C-test English	55.38	17.92	53.18	18.52	52.23	17.63	.061
	Socio-economic status	56.55	20.18	43.98	19.10	39.29	18.85	<.001
	Cognitive ability	18.09	5.28	17.61	5.23	15.36	5.91	<.001
	LGVT German	31.85	10.08	29.19	10.07	26.80	11.23	<.001
MTP2	C-test English	51.77	15.48	50.21	16.04	48.01	15.09	.008
	LGVT German	36.14	13.28	32.50	13.52	29.97	13.92	<.001
MTP3	C-test English	53.46	14.87	52.19	15.33	48.51	14.63	<.001
	LGVT German	41.77	14.34	38.58	14.86	36.17	16.78	<.001

Note: Based on one-way ANOVA and post-hoc Tukey HSD results (p < .05). MTP1: C-test English: There are no statistically significant differences among groups. Socio-economic status: German monolingual group is statistically different from the other two groups. Cognitive ability: Turkish–German group is statistically different from the other two groups. LGVT German: German monolingual group is statistically different from the other two groups. MTP2: C-test English: Turkish-German group is statistically different from German monolingual group. LGVT German: German monolingual group is statistically different from the other two groups. MTP3: C-test English: Turkish-German group is statistically different from the other two groups. LGVT German: German monolingual group is statistically different from the other two groups. Statistically significant comparisons are highlighted in bold.

Data analysis

We used multiple linear regressions (stepwise backward model selection)to track the longitudinal changes and examine the strength and contribution of each variable to the model of English proficiency in the general sample (featuring both monolingual and bilingual students) and the bilingual samples (Russian-German/Turkish-German students). The aim was to investigate the relationships between the response variable L2/L3 English proficiency and a set of linguistic and extra-linguistic predictor variables, that is cognitive ability, socio-economic status, German proficiency, heritage language proficiency (if applicable), language group, school type (Gymnasium vs other), school year (7 vs 9; 8 vs 10), and gender. For this, we conducted two separate analyses. The first included all students. In a second step, we considered the bilingual participants separately, as we wanted to add heritage language proficiency as an additional predictor. We ran regressions for each MTP separately allowing two-way interactions between all variables. The final models were checked for linearity, normality, and multicollinearity to ensure that all regression assumptions were met. The analyses were conducted in the R environment (Version, 4.0.2; R Core Team, 2020).

Results

The results section is divided into two subsections. First, the output of the regression analysis is presented for the general sample (all students at three MTPs; n = 974). Second, this is followed by the presentation of the results for the bilingual sample (n = 301) at the three MTPs.

Results general sample

The multiple linear regression analysis returned three models that differ to some extent concerning the variables they include. However, the overall predictive power is similar at the three MTPs, with values slightly lower than .6. This means that all three regression models explain nearly 60% of the variance in L2/L3 English proficiency among the secondary school students (see Table 3). The specific details are untangled below. Note that we had to lower the significance level to adjust for multiple testing. We followed the Bonferroni correction and assumed .05/3 (=.017) as the cut-off point for rejecting the null hypothesis.

Table 3. Multiple linear regression predicting L2/L3 English proficiency at three MTPs in the general sample (n = 974).

	MTP1				MTP2				MTP3
	Estimate	Standard error	t-value	p-value	Estimate	Standard error	t-value	p-value	Estimate	Standard error	t-value	p-value
Intercept	0.354673	0.024551	14.447	<.001	0.43939	0.02509	17.510	<.001	0.35206	0.03171	11.102	<.001
Cognitive ability (KFT)	0.170159	0.020775	8.191	<.001	0.17691	0.02286	7.738	<.001	0.15882	0.02110	7.526	<.001
German proficiency	0.358568	0.048183	7.442	<.001	0.23852	0.04127	5.780	<.001	0.42359	0.05417	7.896	<.001
School type other	−0.29666	0.030697	−9.664	<.001	−0.34324	0.02712	−12.650	<.001	−0.29173	0.03276	−8.906	<.001
School year 9/10	0.140714	0.008636	16.293	<.001	0.12292	0.00934	13.170	<.001	0.18067	0.03120	5.791	<.001
Language group Russian–German					0.01971	0.01740	1.132	.258	0.00792	0.01153	0.687	.492
Language group Turkish–German					−0.03333	0.01534	−2.174	.030	−0.02730	0.01036	−2.568	.010
German proficiency : School type other	0.401116	0.078478	5.111	<.001	0.51789	0.06817	7.597	<.001	0.31293	0.06413	4.879	<.001
German proficiency : School year 10									−0.16170	0.06038	−2.679	.008
School type other : Language group Russian–German					−0.02121	0.02522	−0.841	.401
School type other : Language group Turkish–German					0.05699	0.02319	2.457	.014
R^2/adjusted R^2	.5872/.5851			.5831/.5792			.5689/.5653

Note: Reference level for language group: German; Reference level for school type: Gymnasium; Reference level for school year MTP1: 7 and MTP2/3: 8; Interactions indicated with a colon (:). Statistically significant predictors are highlighted in bold.

Cognitive ability, German proficiency, school type, and school year feature significant predictors in all three models, either as main effects or in interactions. As expected, the influence of cognitive ability (KFT) and proficiency in German is positive, that is, with increasing KFT and German reading comprehension scores, the English C-test score increases. Furthermore, German proficiency is never a main effect, but it interacts with school type and year. Nevertheless, regardless of the interactions, the effect is always positive. This can be seen in the two lower effect plots visualised in Figure 4, visualising all effects at MTP3 (see also Figures 2 and 3 for MTP1 and MTP2).

Figure 2. Effect plots for the general sample at MTP1.

Figure 3. Effect plots for the general sample at MTP2.

Figure 4. Effect plots for the general sample at MTP3.

The effect of school year largely confirms our expectations. There is a positive influence on the English C-test scores with increasing age. At MTP1 and MTP2, this is a main effect (see Figures 2 and 3), and at MTP3, school year interacts with German proficiency (see Figure 4). The regression line in the interaction with German proficiency is steeper for students attending school year 8, and in addition, the regression line starts at a higher C-test score for students attending school year 10.

Moreover, language group belonging does not show a significant influence at MTP1, whereas it appears to have some effect at MTP2 and MTP3. At MTP2, this effect is contingent on school type (MTP3; see Figure 4). It becomes clear that only one of the relationships visualised in the lower right effect plot shows a significant difference. Whereas there is no statistically significant difference across the three language groups among students attending other school types (i.e. lower academic school track), there is a significant difference between the Turkish-German and Russian-German bilinguals as well as the Turkish-German bilinguals and the German monolinguals attending Gymnasium (i.e. the higher academic school track). The difference between the German and Russian-German students does not reach statistical significance. Thus, belonging to the Turkish-German language group has a negative effect. Furthermore, at MTP3, language group turns out to be a main effect. The same relationship as before remains, just without the difference across school types: belonging to the Turkish-German group has a negative effect on the English C-test score when compared to their monolingual and Russian-German peers. Finally, neither socio-economic status nor gender remained in the three regression models after the model selection process, which means that these do not add significantly to explaining the variance in L2/L3 English proficiency once all other variables are controlled for.

Results bilingual sample

This section considers the bilingual students (n = 301). As in the previous section, the three multiple linear regressions predicting L3 English proficiency for the Russian-German and Turkish–German students differ slightly across the three MTPs. The predictive power of the three models is again approximately 60%, with values a little lower than .6 for MTP1 and MTP3 and marginally higher than .6 for MTP2 (see Table 4). As before, we followed the Bonferroni correction and considered .017 as the adjusted significance level.

Table 4. Multiple linear regression predicting L3 English proficiency at three MTPs in the bilingual sample (n = 301).

	MTP1				MTP2				MTP3
	Estimate	Standard error	t-value	p-value	Estimate	Standard error	t-value	p-value	Estimate	Standard error	t-value	p-value
Intercept	0.28481	0.04069	6.999	<.001	0.52668	0.03120	16.879	<.001	0.24499	0.04218	5.808	<.001
Cognitive ability (KFT)	0.14352	0.036	3.987	<.001					0.15539	0.04408	3.526	<.001
German proficiency	0.60906	0.09953	6.120	<.001	0.41298	0.06861	6.019	<.001	0.52916	0.06712	7.884	<.001
School type other	−0.15892	0.01666	−9.537	<.001	−0.35472	0.04416	−8.033	<.001	−0.17999	0.02035	−8.843	<.001
School year 9/10	0.26956	0.04837	5.573	<.001	0.13768	0.01666	8.262	<.001	0.12355	0.01904	6.490	<.001
Language group Turkish-German					−0.07519	0.02199	−3.419	<.001
German proficiency : School type other					0.36740	0.11932	3.079	.002
German proficiency : School year 9	−0.33018	0.12935	−2.553	.011
School type other : Language group Turkish-German					0.09794	0.03280	2.986	.003
R^2/adjusted R^2	.5835/.5764			.6225/.6148			.5483/.5422

Note: Reference level for language group: Russian-German; Reference level for school type: Gymnasium; Reference level for school year MTP1: 7 and MTP2/3: 8; Interactions indicated with a colon (:). Statistically significant predictors are highlighted in bold.

Overall, the regression output demonstrates that cognitive ability, German proficiency, school type, school year, and conditionally also language group at MTP2 significantly impact L3 English proficiency. The corresponding effect plots at the three MTPs can be found in Figures 5–7.

Figure 5. Effect plots for the bilingual sample at MTP1.

Figure 6. Effect plots for the bilingual sample at MTP2.

Figure 7. Effect plots for the bilingual sample at MTP3.

Cognitive ability (KFT) positively influences English proficiency at MTP1 and MTP3, but does not reach statistical significance at MTP2. In addition, the positive effect of German proficiency, as shown in the analyses of the general sample, is also true for the reduced bilingual sample. Whereas at MTP1, the German proficiency score interacts with school year and at MTP2 with school type, it appears as a main effect at MTP3 (see Figures 5–7). The impact of school year and type is also similar to the previously reported results. School year is a main effect at MTP2 and MTP3, meaning that with increasing school year, the English C-test scores increase (see the lower right plot in Figure 7). At MTP1, school year interacts with German proficiency. The regression line of the year nine students starts at a higher point and the increase is overall flatter compared to the year seven students (see Figure 5). Generally, students attending Gymnasium (i.e. the higher academic school track) show an advantage over students attending any of the other school types. At MTP1 and MTP3, this is a main effect, whereas at MTP2 it interacts both with German proficiency as well as language group. As had been demonstrated for the general sample, the significant difference between Russian-German and Turkish-German students attending Gymnasium remains. The Russian-German students enjoy a small, albeit significant advantage. Nevertheless, no such statistical significance can be reported for students attending other school types (see lower right plot in Figure 7). Note, however, that this effect is only significant at MTP2.

Interestingly, the effect of heritage language proficiency did not reach statistical significance in any of the three regression models and was therefore not included. A tentative explanation for this is offered in the discussion section. Moreover, gender and socio-economic status did not feature in the models either due to their lack of explanatory power.

Discussion

The discussion consists of three sections. The first section relates the results to the three research hypotheses. The second section reflects on language interdependence and language dominance from a longitudinal perspective, and the final section seeks explanations for the current findings in the educational setting.

Reconsidering the initial hypotheses

The aim of this study was to answer two research questions, namely (1) which personal, linguistic, and contextual variables explain L2/L3 proficiency, and (2) whether and how these change across monolingual and bilingual English learners over time, and to test whether the three hypotheses can be confirmed with the data at hand. The three hypotheses will be discussed in turn in the following paragraphs.

Our first hypothesis stated that cognitive ability, operationalised with the KFT score assessing visual-spatial (non-verbal) skills, remains a strong and significant predictor over time. The results based on the regression analyses have shown that cognitive ability is indeed a significant factor in all MTPs, with the exception of the bilingual sample at MTP2. Yet, the standardised estimates are lower than for German proficiency and remain below 0.3 in the general sample and even below 0.2 in the bilingual sample (see Tables 3 and 4). This means that the first hypothesis can only be partially confirmed.

The second hypothesis predicted that over time, the impact of heritage language proficiency decreases whereas the impact of German proficiency increases. Such a clear trend could not be observed for German proficiency. Whereas there was an increase from MTP1 (.36) to MTP3 (.42), the estimate was lower at MTP2 (0.24) in the general sample. In the bilingual sample, the standardised estimate was even highest at MTP1 (0.61), decreased at MTP2 (0.41), and increased again at MTP3 (0.53). In addition, no measurable effect of heritage language proficiency could be reported, at neither of the three MTPs of the analyses considering the bilingual learners of English. Therefore, the second hypothesis cannot be confirmed. Nevertheless, the explanatory power of German proficiency remains significant over time and is, even though a systematic pattern is absent, comparably stronger than that of cognitive ability.

The third hypothesis conjectured that the monolingual and bilingual learners of English converge over time owning to the continuing time of studying English in the German secondary school context. At MTP1, language group was not among the statistically significant variables. Moreover, the interaction of language group and school type at MTP2 demonstrated that there was no general difference between the German monolingual learners and their bilingual peers, but that a difference was only identified between the Turkish-German students, on the one hand, and the Russian-German and German students, on the other hand, who attended the higher academic school type Gymnasium. What is more, at MTP3, language group was indeed a main effect. Belonging to the Russian-German or the German group had a positive impact on the English C-test scores when compared to their Turkish-German peers. Thus, language background was shown to be not significant at MTP1, conditionally significant at MTP2, and significant at MTP3, which means that the third hypothesis has to be rejected with the data at hand.

Finally, although nothing specific had been hypothesised about school type, school year, gender, and socio-economic status, these variables had to be controlled for as they are known to affect (foreign) language acquisition (see, for example, Maluch et al., 2015, 2016). Surprisingly, neither socio-economic status nor gender could be shown to contribute significantly to explaining the variance of English proficiency. School type and school year, however, were shown to add to all six regression models. Simply put, attending the Gymnasium and attending a more advanced school year affected the English C-test scores positively.

Interdependence and language dominance over time

The current study can be understood as support for Cummins (1976, 1979) interdependence hypothesis considering the strong positive impact of German proficiency on L3 English proficiency for all students (see also Maluch et al., 2015). However, no measurable-significant interdependence between heritage language proficiency and L3 English was found (but see, for example, Maluch et al., 2016; Maluch & Kempert, 2019). It needs to be stressed again that all participants, including the bilinguals (see Table 2), are dominant in German, which is the language of instruction in school and additionally typologically and structurally similar to English. The corresponding scores in the reading comprehension test in the heritage language (either Russian or Turkish) are comparably low (see Table 5). Note that the scores for German increase over time, whereas no clear pattern can be observed in the heritage language scores. Proficiency in Russian and Turkish might simply be too low to show any meaningful effect on English proficiency, especially as Russian and Turkish are not academically supported in the school context. This point is further discussed in the following section.

Table 5. Descriptive statistics (absolute values) of performance in German and heritage language reading comprehension (LGVT) in the bilingual sample (n = 301).

		Russian–German (n = 123)		Turkish–German (n = 178)
		Mean	SD	Mean	SD
MTP1	LGVT German	29.13	10.15	26.72	11.47
	LGVT heritage language	6.22	9.97	8.72	9.37
MTP2	LGVT German	32.13	13.47	29.81	13.98
	LGVT heritage language	8.92	10.65	15.66	10.84
MTP3	LGVT German	38.54	14.72	35.93	16.51
	LGVT heritage language	7.90	11.41	10.04	11.16

Educational setting and possible explanations

Other than expected, the monolingual and bilingual students did not converge from the first to the third MTP. Whereas there was no group effect at MTP1, a significant difference for the Turkish-German group could be identified at MTP2 and MTP3. The Turkish-German group differed significantly from their bilingual peers (Russian-German students) and their monolingual German peers, whereas the Russian-German students did not differ from the German students. This means that at MTP2 and MTP3, being raised monolingually or bilingually added significantly to explaining the observed differences in English proficiency; however, only concerning the Turkish-German and German students but not the Russian-German students. Belonging to the Turkish-German group showed a negative effect, whereas belonging to the Russian-German group showed no effect at all when compared to the monolingual German students (see also Maluch et al., 2015, 2016). A possible reason for the observed differences between the Turkish-German and Russian-German students could be their variations in cognitive ability scores. Previous research, including our own work, has demonstrated that cognitive ability is a significant predictor of language proficiency (Berthele & Udry, 2022; Lorenz et al., 2023; Siemund et al., 2023, 2024). Therefore, one might argue that the relatively lower cognitive performance of the Turkish-German group contributed to the observed differences. Nonetheless, since we controlled for cognitive ability in the regression models, we argue that the observed group differences need to be taken seriously, particularly as they were absent at MTP1 and only relevant at the remaining two MTPs. We note that there is a growing body of research suggesting that individuals with different heritage language backgrounds perform differently rather than functioning as one homogeneous group (Lorenz et al., 2021a). Furthermore, language interdependence, which is closely linked to cognitive ability, can be notably more pronounced among languages supported by education (Siemund et al., 2024). Overall, the negative effect associated with Turkish-German belonging finds support in numerous studies comparing the general academic success of monolingual and bilingual students in Germany, often featuring students with a Russian or Turkish background as these make up the largest immigrant groups of Germany (e.g. Diehl & Fick, 2016; Nauck & Lotter, 2016; Olczyk et al., 2016). Nauck and Schnoor (2015) discuss the relevance of ‘ethnic residuals' (see also Nauck & Lotter, 2016; Kristen et al., 2016), but it remains to be seen to what extent this concept harbours explanatory power.

Finally, previous research has shown that the current learning context in German secondary schools makes it unlikely for students to access their full linguistic repertoires, particularly their heritage languages: (i) German secondary schools are largely dominated by a monolingual ideology (Gogolin, 2021); (ii) German and other languages are typically avoided in the English language classroom (Elsner, 2018; Fuller, 2020), meaning that (foreign) languages are predominantly taught in isolation (Hopp & Jakisch, 2020; Krumm & Reich, 2016); and (iii) heritage languages are frequently discouraged, not only in school but also in society at large (Fuller, 2020; Wiese et al., 2017). English as a foreign language development thus appears to be comparable across the monolingual and bilingual participants.

Limitations

This study is of course not without limitations. One potential weakness is the methodological necessity of running the regression analyses separately for each MTP. The individual tests (reading comprehension in German, Russian, and Turkish; C-tests in English) differed across the three MTPs which means that they are not directly comparable across the MTPs and only meaningful for each MTP separately.

Moreover, cognitive ability as well as socio-economic status had only been measured or assessed at MTP1. We assumed that these two variables remain stable within the test period of 1.5 years. Moreover, socio-economic status is exclusively based on parental occupation. Additional parameters, such as parental income or cultural capital (number of books per household), might increase the explanatory power of this variable.

In addition, language proficiency in the majority and heritage languages was assessed differently than language proficiency in the foreign language (reading comprehension vs C-tests). Another point worth mentioning is that the Russian–German participants had the option of choosing between a Cyrillic and a transliterated version of the reading comprehension task. This difference was not considered in the current study and might be something to investigate further in the future.

Finally, the study did not employ any measure to assess metalinguistic awareness. Future studies need to include this to explicitly test its assumed mediator effect of bilingual advantages.

Conclusions

The present study set out to determine which personal, linguistic, and contextual variables account for L2/L3 English proficiency and how these variables interact and change in monolingual and bilingual learners over time. Overall, the findings indicate that German proficiency and cognitive ability positively impact L2/L3 English proficiency, while no effect of socio-economic status or gender was observed. Moreover, in the context of the present study, language background (i.e. having a monolingual or bilingual upbringing) did not bring about clear differences between the monolingual and bilingual cohorts per se. There were no significant differences between the Russian-German and German students. However, a negative effect could be measured for the Turkish-German group at MTP2 and MTP3 in the general sample and at MTP2 in the bilingual sample. Thus, whereas English language development was comparable across the monolingual and bilingual Russian-German group, some differences could be observed for the Turkish-German group, possibly associated with their comparably lower cognitive ability scores.

Previous research suggested that bilingual experience can facilitate subsequent language learning experiences, mainly due to heightened metalinguistic awareness (De Angelis, 2007; Jessner, 2008). Our finding that there was no clear ‘bilingual advantage’ for the heritage bilinguals might be linked to the current state of the German secondary school context and bear important implications for the institutional and educational settings. This particular context has previously been characterised as mainly adopting a monolingual ideology (Gogolin, 2021), treating foreign languages in isolation (Hopp & Jakisch, 2020; Krumm & Reich, 2016) or avoiding languages other than English, including heritage languages such as Turkish or Russian, in foreign language classes (Elsner, 2018; Fuller, 2020). Consequently, it might be difficult for students to exploit their bilingual potential or access their full linguistic repertoires. Therefore, changing institutional settings could considerably facilitate language interdependence and lead to heightened metalinguistic awareness. In the multilingual turn in language education (May, 2014), current approaches view multilingual language teaching as an opportunity to fully exploit languages existing in students’ linguistic repertoires and promote multilingual competence (Krumm & Reich, 2016). Similarly, posing a multilingual and plurilingual approach to foreign language classrooms by pointing to cross-linguistic differences and similarities has been shown to promote language development in students (Hopp & Thoma, 2021). We thus propose a more holistic approach to foreign language teaching that recognises and draws on students’ diverse language backgrounds and features pedagogical translanguaging (Cenoz & Gorter, 2021) to metalinguistic awareness in students (Elsner, 2018; Jessner & Allgäuer-Hackl, 2020; May, 2014). Such an approach, in contrast to ‘normative monolingualism’ (Fuller, 2020, p. 171), which seems to be commonly practised in Germany, may benefit both monolingually as well as bilingually raised students. This, however, needs to be backed up with empirical evidence and thus presents an avenue for future research.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

This work was supported by the Federal Ministry of Education and Research (Bundesministerium für Bildung und Forschung) under grant 01JM1406. The financial support is gratefully acknowledged.

Notes

1 The Russian version of the reading comprehension and fluency test was available in Cyrillic and Latin script. Students could choose which one they wanted to take. In the analysis, we did not differentiate between the two versions.