Exploring overt cross-linguistic influence in lower secondary learners written texts in English and French as a second or third language

ABSTRACT This article explores cross-linguistic lexical influence (CLI), i.e. idiosyncratic borrowings and creations based on other languages, in texts written by pupils learning French and English as foreign languages, and whose language of instruction is German. The study, carried out in German-speaking Switzerland, was based on a plurilingual conception of language learning and drew on research on tertiary languages. Two groups of learners at the end of Grade 7 (age 13/14) were compared before and after a reform in the teaching of English. While both groups had been learning French since Grade 5, one group had been learning English before French since Grade 3 (Group 3/5, n = 220) and the other after French since Grade 7 (Group 5/7, n = 94). The L3 French texts of group 3/5 revealed a higher number of idiosyncratic borrowings and creations based on English L2, while the English L3 texts of group 5/7 borrowed more from L2 French and presented more lexical creations that drew on German. The correlations showed that learners who use lexemes with a detectable cross-linguistic influence in one language also tend to do so in the other. The results are interpreted in relation to the educational context, and to the concept of translanguaging.


Introduction
The demands of globalisation have led many countries to adopt the teaching of English as a foreign language 1 increasingly earlier.The case is no different in quadrilingual Switzerland where, traditionally, the teaching of a second official language has been privileged.Following curriculum reforms, some of the cantons whose official language is German have decided to make the teaching of English compulsory from the third instead of the seventh school year, and French from the fifth school year.Our study analysed texts written in English and French by two groups of learners in the seventh grade, enrolled either before or after the reform.The objective was to understand the extent to which these changes in the learning conditions modified the relationship between the two languages for the learners.The study was based on an integrative conception of the plurilingual repertoire which includes the resources of all the languages acquired by a single individual, which are dynamically connected and available in their entirety (see e.g.Cenoz & Gorter, 2011;Cook, 2016;Coste, Moore et al., 2009;Jessner, 2017;Lüdi & Py, 2009).
This paper focuses on overt cross-linguistic lexical influence, in specific on the insertion of words borrowed without modification from another language and the idiosyncratic use or creation of words bearing traces of the target language and of one or more other languages.The following questions will be explored: First, to what extent do the patterns of cross-linguistic influence in English and French texts (borrowings and lexical creations) differ in the two languages and in the two groups?Second, do learners who show CLI in one foreign language (English or French) also do so in the other?
Numerous studies have analysed cross-linguistic lexical influence (referred to henceforth as CLI), but most have focused on oral, rather than written, production (cf.Jessner et al., 2016) and fewer still have focused on production in more than one language at the same time.The present study investigated these phenomena in texts written in second and tertiary languages.Specifically, it compared two groups depending on when they first began learning the two languages and on the fact that one of the two languages had been learned over a longer period of time.This allowed us to examine not only the impact of this change, but also the intra-individual relationships in CLI use in written production in several languages (Cenoz & Gorter, 2011).We discuss our observations based on the concepts of translanguaging (Canagarajah, 2011;Cenoz & Gorter, 2019;García & Wei, 2014) and pluriliteracy (Moore, 2020) from an educational perspective.

A plurilingual conception of written production in foreign languages
Analysing written production in several languages in the same individual is, in principle, a necessary condition for understanding the specificity of plurilingual functioning and development (Cenoz & Gorter, 2011;De Angelis & Jessner, 2012;Marx, 2020;Rinnert & Kobayashi, 2016).However, the majority of studies have focused on a single language (second, tertiary, or foreign), thus neglecting the other languages in an individual's repertoire.In their ground-breaking study, 'Focus on multilingualism', which analysed texts written in Basque, Spanish, and English by pupils from the Spanish Basque Country, Cenoz and Gorter (2011) observed that these pupils tended to structure texts in a rather similar manner in all three languages.Moreover, in each text, learners borrowed words from the other languages.They interpreted these observations as traces of the permeability of the boundaries between languages, the multidirectional interactions between languages, and the investment of all resources in the complex process of producing written texts in several languages.A growing body of research, which has undertaken similar studies in a bi-or plurilingual context, sheds light on the integratedor even mergednature of plurilingual competence in written production (cf.Reichert &Marx, 2020 andRinnert &Kobayashi, 2016 for recent works).Drawing on complex writing patterns in the tradition of Hayes and Flower (1980), these studies have observed cross-cutting ties between languages in the writing process, i.e. in the fluency of writing, in writing strategies, or in revision processes.They have also revealed the transversality between languages in terms of the written product, in terms of productivity in the different languages and the quality of the text and in the specific structure of the different types of texts; for instance, there is a similar use of connectives.While the implementation of such transversal capabilities results in texts that conform to the standards of each language, the activation of elements attributable to different languages in an individual's repertoire leads to hybrid products.These phenomena have been analysed under the label of CLI in the research on the acquisition of second and tertiary languages, on which our study relied for the analysis of pupils' texts.

Simultaneous activation of multiple languages as a trace of multidirectional cross-linguistic influence in the plurilingual repertoire
The simultaneous activation of several languages in plurilingual individuals concerns experienced subjects as well as beginners.Both orally and in writing, it may reflect the intentional expression, or a more or less conscious compensation strategy in a situation of verbal distress.As Lüdi (1994;2020) argues, although extreme cases seem easy to distinguish at first glance, it is often impossible to identify these different uses with certainty because their status can only be determined by their context and interpretation.In the context of learning, the visible traces of the simultaneous activation of two or more languages at different levels of the language system have been the subject of analyses over several decades (for an overview cf.e.g.Bono, 2008;Jessner et al., 2016;Singleton, 2015).With regard to the cross-linguistic use of the lexcion, focused upon in this study, the same phenomena have been conceptualised and named in different ways.The denomination expresses the point of view adopted; it also diverges, depending particularly on the status granted to the prescriptive norm.This applies to borrowing words without modification ('on my Pult', the German word for desk) as well as to lexical creations bearing the traces of another language ('la wale', in French, created from the English wall, used instead of 'le mur').These phenomena have been conceptualised in different ways: interferences by Weinreich (1953), (negative) transfers leading to errors by Selinker (1972), andmore neutrallyas cross-linguistic influence (Sharwood Smith & Kellerman, 1986) or cross-linguistic interactions (Bono, 2008;Jessner et al., 2016).Even more neutrally, they have been conceptualised as cross-linguistic or transcodic marks or markers (Lüdi, 1987), or as forms of meshing in the more pedagogical and social perspective of the translanguaging concept (Canagarajah, 2011).The latter was created in a context of bilingual education to describe fluid, creative, and individual language practices which, in the context of written production, can be considered as legitimate practices of pluriliteracy, worthy of being educationally reinforced (Canagarajah, 2011;Cenoz & Gorter, 2019;García & Wei, 2014;Moore, 2020).
All these concepts are still common today.We use the non-normative term of cross-linguistic influence 'CLI', which is the term most frequently used in research on the acquisition of second and tertiary languages.It should be noted that many CLI-studies categorise these phenomena as errors, which implies an a priori normative perspective.The concept of cross-linguistic interaction aptly reflects the multidirectionality and dynamics of relationships, but it risks leading to confusion because it may also refer to interactions between communication partners.It has been excluded, as has the concept of transfer, which may imply a unidirectionality of the relationship between languages (from language X to language Y) 2a phenomenon that indeed benefits from being conceived as 'multiple directional relationships between multiple languages' (Cook, 2016, p. 35).In the results section, we will attempt to integrate the various perspectives mentioned above, including that of translanguaging.

Studies of lexical cross-linguistic influence in tertiary language acquisition
CLI are tangible indicators of the interaction between the resources of different languages in plurilingual repertoires and of their mutual support, a phenomenon that is much more difficult to demonstrate when the language employed conforms to the rules of a certain variety (Bono, 2008, p. 156).Many studies on tertiary languages have focused on CLI (see, for instance, Jessner et al., 2016). 3These studies have sought to explain, to put it simply, the extent to which the acquisition of L3 + differs from that of L2, and the role played by L2, or other languages previously learned, alongside L1.They have attempted to understand which languages are activated when, by whom, and why, and thus to explain the psycholinguistic, sociolinguistic, and even pedagogical factors or conditions that favour the emergence of CLI.
CLI has been categorised according to different typologies referring to different traditions (see for instance Dewaele, 1998;Ringbom, 2001 andthe discussion in Force-Izzard, 2018).There are two main categories of CLI: on the one hand, the insertion in a language of unaltered forms from other languages (idiosyncratic borrowings, shifts, or switches) and, on the other, the construction of idiosyncratic forms based on one or more other languages, with an adaptation either at the level of the signifier (referred to as lexical inventions, relexifications, or foreignisings) or at the level of the signified (implying, for instance, the extension of the meaning of a word from the target language).CLI has also been analysed according to grammatical categories (roughly speaking, content words and function words have been distinguished).
Most studies have focused on oral production in adult learners (Jessner et al., 2016).For example, the highly influential longitudinal case study by Williams and Hammarberg (1998) analysed, over a two-year period, CLI in the Swedish (tertiary language) production of an English L1 speaker with a very advanced knowledge of German L2, acquired in a school setting, and also her poorer knowledge of French and Italian.According to the authors, her preferred activation of German was influenced by the combination of four factors: recency (whether the language already learned had been used recently or not); proficiency in this language (beginner or advanced level); the degree of typological similarity between L2 and L3 (for instance, German is a Germanic language, as is Swedish); and L2 status (i.e. the preference to rely on a language previously learned as L2).The latter factor was also highlighted by De Angelis and Selinker (2001), who observed in a case study that a French-speaking adult speaking Italian as a tertiary language tended to activate Spanish, learned and perceived as a 'foreign language', to fill lexical gaps, rather than French, which is also close typologically speaking.In 'foreign language mode', concerns about correctness supposedly prevent speakers from using words from their L1, perceived from the outset as erroneous.In their observations, De Angelis and Selinker also confirmed both the importance of psychotypology (Kellerman, 1995), i.e. the subjective perception of the typological proximity between languagesrather than their actual proximityand of individuals' awareness of this perception, which was a determining factor in whether they would capitalise on the correspondences between languages.The perception of a relationship seems to be influenced by metalinguistic consciousness which increases as learners grow older.This was suggested in the study by Cenoz (2001) on the use of CLI in the oral production of stories in L3 English by three groups of Spanish-Basque bilingual learners (2nd, 6th, 9th grade, n = 90), all of whom had been learning for the same length of time.CLI in older pupils, which was just as considerable as in younger pupils, drew more on Spanish than on Basque, the latter being typologically more distant from English than Spanish.The importance of the duration of exposure and the order of learning in L2 was confirmed by Dewaele (1998).He compared lexical inventions in the oral interaction in French of two groups of Dutch-speaking learners (L1) at the advanced level, some of whom had learned French (L2) and English (L3) (n = 32) while the others had French (L3) and English (L2) (n = 7).Lexical inventions were observed in both groups, but those of the L2-French group largely drew on Dutch, while those of the L3-French group, whose L2 was English, primarily relied on English.Dewaele (1998) also observed that lexical inventions, for both groups, decreased in formal situations.Tremblay (2006) emphasised the role of typological proximity, competence, and exposure to L2.She analysed CLI in L3-German oral productions (learners had been learning for 2-3 years) of three groups of English-speaking adult learners (n = 13) with varying levels of proficiency and duration of exposure to L2 French.The results suggest that a higher level of competence must be acquired in L2 to activate it, and that the duration of exposure plays a role.However, in most cases, CLI in all learners was largely based on English, which is typologically closer to German.
Fewer studies have focused on the analysis of these phenomena in the production of written texts.The influence of typological proximity and the dynamics of psychotypology were highlighted by Bono (2008) in her study on lexical inventions in texts written in L3 Spanish (level B2 of the CEFR) by a group of 60 French-speaking university students (L1) whose L2 was English and/or German.Most of the lexical inventions and calques identified in the Spanish texts were literally close to both L1 French and L2 English (such as letra for carta, the corresponding words in French and English being 'lettre' and 'letter'), the formal proximity in L1 and L2 favouring the creation of a lexeme in L3 based on the learning hypothesis L1 = L2 = L3.Other types of inventions have been classified as strategies to overcome difficulties: there are, for instance, the semantic shifts of lexical units that exist in the target language (alta from the German alt for vieja,, preferred to the activation of vieille (old), the word alta existing in Spanish to signify high or large); inventions with no connections to L2 (quoto for I quote from English, instead of the Spanish construction cito) which reflects the L2 effect ( foreign language effect).Lexemes in L1 were also put into quotes, signalling that learners were aware that the choice was unsuited to the unilingual standard (cf.Lüdi (1994) who identified words in L1 written in pencil in copies of L2 exams).Other analyses have confirmed the importance of the activation of L2 in written production in L3.For instance, the study by Pfenninger and Singleton (2016) examined the influence of the age at which individuals began learning English as a foreign language (from Grade 2 or Grade 7 onwards) and its long-term effects on CLI in the oral and written English texts produced by Germanspeaking learners at the end of Grade 7 and Grade 12, who were also learning French (n = 200).In Grade 7, early starters displayed less CLI than did late learners, but while early starters produced more lexical inventions and calques, late learners tended to borrow more.An analysis of borrowed words revealed that only content words were borrowed from French, while in addition to content words, conjunctions, pronouns, auxiliaries, and prepositions were also borrowed from German.In Grade 12, CLI was rare in both groups.According to the authors, the decrease in CLI over time and the predominance of lexical invention in early learners in Grade 7 was a result of learners' more advanced English proficiency.This finding is consistent with the results of the study undertaken by Agustín-Llach (2014).Indeed, Agustín-Llach found that German and Spanish fourth graders (age 9/10) described as low proficient L2 learners of English predominantly turned to borrowing rather than to lexical invention and calques in their written production.
In sum, in oral or written production in language acquisition, the activation of other languages present in an individual's repertoire is influenced by various factors: competence and level of learning (beginner or advanced) in all the languages involved, their typological proximity, psychotypology (the subjective perception of the distance between languages, with a word perceived as similar in language X more likely to be activated in language Y), their status as L2 or as a 'foreign language', the learners' age, their meta-linguistic awareness, the recency and duration of exposure to L2, and the textual or situational typological context.Moreover, borrowings, especially of content words, are more common in the early learning stages, while lexical inventions suppose a certain level of competence.
It is also worth mentioning that all the studies cited above focus only on production in a single target language, L3 (L2 for the research by Agustín-Llach (2014) and Dewaele (1998) who compared French as L2 and L3, respectively), with the notable exception of Cenoz and Gorter's (2011) paper presented earlier, which included qualitative observations on CLI in L1, L2 and L3 within Basque learners of Spanish and English.
Exploring the phenomena of overt cross-linguistic lexical influences found in texts written in French and English as L2 and L3 by students raised in Germanophone Switzerland will permit the further illumination of the interplay between those languages in the plurilingual repertoires of the pupils.

The educational context in Switzerland
In Switzerland, a country in which there are four languages with official status, the use of languages is governed by the principle of territoriality, enshrined in the Federal Constitution.German, the majority language (65%), coexists with three other languages: French (23%), Italian (8%), and Romansh (0.5%).Due to high levels of immigration, the native language of approximately 30% of the children attending compulsory school differs from these official languages (Federal Statistical Office, 2021).Given that Switzerland does not have a Ministry of Education, the making of federalist education policies remains a prerogative of the cantons but The Swiss Conference of Cantonal Ministers of Education (CDIP) is responsible for national coordination.In most cantons, only the regional language is used as the language of instruction.However, in accordance with article 70 of the Federal Constitution (1999), the cantons must promote the acquisition of at least one second national language with a view to supporting cohesion between the different regions.As part of a comprehensive curriculum reform (CDIP, 2004), the cantons agreed to teach two foreign languages from primary school level, predominantly one second national language and English.They were allowed to choose the order of introduction as long as learners attained the same competences by the end of compulsory education.In most cantons of German-speaking Switzerland, including the one in which this study was conducted, English was introduced from the third year of primary school and French from the fifth.In the plurilingual perspective outlined above, these decisions obviously have consequences for both English and French: although neither the age at which pupils begin learning French, nor the teaching hours, have changed, it is no longer the first but only the second foreign language, hence it is a tertiary language.

Research questions
In the light of the issues raised in the literature review, the following exploratory research questions were addressed.
(1) To what extent do CLI patterns (borrowings and creations, different source languages) at the lexical level, identifiable in texts written by students of English and French as a foreign language, differ in two groups of learners in grade 7, enrolled in Germanspeaking Switzerland, before and after the curriculum reform described earlier?(2) Do learners who show CLI in one foreign language (English or French) also do so in the other?

Participants
The texts were collected in a canton of Eastern Switzerland from pupils aged 13/14 at the end of the first year of secondary education (grade 7), n = 315.These pupils were learning foreign languages in the diglossic environment of German-speaking Switzerland: they were dialectophones and were enrolled in schools where the language of instruction was standard German.Due to schools' location in the East of Switzerland, there is generally no direct contact with the French language.Two cohorts were compared: pupils who had begun learning French from Grade 5 and English from Grade 7 (these will henceforth be referred to as 'group 5/7') and learners enrolled under the new system, who had begun learning French from Grade 5 and English from Grade 3 (these will henceforth be referred to as 'group 3/5').The two groups had therefore been learning French for three years but the duration of their learning of English differed (one year for group 5/7 vs. five years for group 3/5).Group 5/7 comprised 95 learners randomly selected from 12 classes drawn from eight schools and group 3/5 had 220 learners selected from 38 classes drawn from 19 schools.The study was designed such that both groups were drawn from schools located in the same six municipalities of the canton (out of 77).In both languages and for both groups, the classes were taught in cursory, textbook-based lessons two or three times per week.A questionnaire was used to collect biographical information.A total of 31.9% of the 3/5 group and 26.7% of the 5/7 group also spoke another language at home.Learners who indicated English or French as their family language were excluded ex post from the sample.All the levels of lower secondary education were 4 represented.The pupils were from varied socio-economic backgrounds.

Text collection
The learners were asked to write letters in German, English, and French as part of an exchange programme with partner classes in French-speaking Switzerland and France.They were expected to introduce themselves (name, age, friends, favourite hobbies) and then describe their home (in German their house or apartment, in French their bedroom, in English their living 5 room).The learners wrote their texts on their own, under supervision in their classroom, according to a standardised protocol.They had 20 minutes to write each text.During the writing exercise, the order of languages was maintained: first German, then French, and then English.As the exercise involved a genuine exchange of letters, we can speak of an event of pluriliteracy with an ecological validity (Street, 2003) that went beyond the simple format of a school test.Writing in all three languages was presented as a form of game: an opportunity to learn each other's language, to teach one's language to others, and to communicate in a shared third language, English.When writing in foreign languages, the use of bilingual dictionaries (paper version) was authorised because the main objective was to study the traces of transversality in textual 6 structuring.The learners and their parents were informed on the purpose of the study and their written consent obtained prior to the data collection.

Categories of analysis
The categorisation of CLI draws on studies undertaken by Lüdi (1994), Dewaele (1998), Ringbom (2001), Tremblay (2006), Agustín-Llach (2014), and Pfenninger and Singleton (2016).The analysis distinguishes between lexical insertions (words from another language inserted without modification) and lexical creations (words bearing the characteristics of two or more languages).For each form, the language from which an element was borrowed was coded: GE = German, EN = English, FR = French, OTH = other languages.The occurrences whose form could be attributed to a possible influence of two or more languages were coded MULT (multiple).Lexical insertion (INS, idiosyncratic borrowing) refers to the use of lexical units of language X (or, as the case may be, other languages) in a text written in language Y, without adaptation to language Y (examples 1-3).Thus, in example (1), we find the units chair and regal (shelf in German) inserted into a sentence in French.The camera unit (bedroom in Italian) in example ( 2) was coded as OTH; the pink unit (replacing the French word rose, which means pink in German and English) was coded as MULT.
(3) Ma chambre a la couleur blanc et pink.(pink INS_MULT) (student 44231037) My room has the colour white and pink.
Common loan words without formal differences, such as hobby or poster in French, were excluded.However, when the words borrowed were uncommon, such as computer in French, they were included in the coding.Although debatable, this decision takes into account the normative issue of foreign language acquisition in schools, with the word ordinateur (computer in French) being part of the learning glossary commonly used in these classes.By lexical creations (CRE), we refer to the lexical units used in a text written in language Y that bears traits adopted from both language Y and language X (possibly at the same time as other languages), either in terms of meaningas in the examples four to six below or in terms of both the signified and the signifier, as in examples seven and eight.
Thus, in example four, the word wale, created from the English word wall and adapted by the addition of an e and the reduction to a single l, replaces the French word mur; or, in example five, the word stocks, created from the German word Stöcke, replaces the word storeys, the plural s of the English replacing the plural e of the German.In example six, the compound word siège-sac (literally seat-bag, beanbag in English and pouf géant or pouf poire in French), is modelled on the words and composition structure of the German Sitzsack while combining French words.In example seven, hear illustrates the case of a semantic extension or calque from the German word hören, referring to both listen and hear.The absence of capital letters in the spelling of inserted German names, such as regal (example one), was not considered a sufficiently strong indicator to classify a unit in the 'lexical creation' category.Cognates and internationalisms without formal differences between the three languages other than being capitalised in German, such as radio or star (sometimes written in capital letters in the English and French texts), were excluded from the coding.Cognates with formal divergences (such as balcon/balcony/Balkon) were taken into account.Language alternations beyond the lexical level, i.e. the insertion in a text written in language X of sequences of lexical units embedded in the syntactic constructions of language Y, which were very rare, were excluded: (8) ma chambre is very big (student 33071639) my room is very big Lastly, cross-linguistic influence at a purely syntactic level, such as the adoption of the word order common in language X (and not in language Y) in a sequence of language Y, were not examined here; the only exception were compound words (see example six).
Coding and statistical analysis CLI coding was performed following an iterative process, with the support of MAXQDA software, designed for qualitative and mixed methods of discursive analysis (Kuckartz, 2010).An initial version of a coding manual was prepared by the author, based on the studies summarised above and a preliminary coding was applied by one collaborator on approximately 40 texts (for both languages and both groups).Based on this coding, the manual was adapted and all the data was coded in MAXQDA by three collaborators, all fluent in German, French, and English.The codes were filtered by lists and re-analysed to guarantee consistency.Inconsistencies and extreme cases were discussed with specialists in English and French, enabling the coding manual to be further adapted.Subsequently, all the texts were coded a second time by the collaborators, during a workshop, and monitored directly.Any doubts or difficulties which arose over classification were immediately discussed and clarified; the coding manual was adapted accordingly and previously coded texts were revised during the workshop.This adaptive procedure allowed for a thorough, self-consistent data analysis.Once all the texts had been coded, the codes were filtered in MAXQDA.The author then personally monitored them before performing the statistical calculations.
To calculate their length, the texts were segmented into clauses containing a single finite or non-finite central verb and its argumentswith some exceptions, especially for complex constructions with an auxiliary or modal verb or verbal ellipses (Hickmann et al., 1994). 7The Wilcoxon test was used to compare the means within and between groups and Spearman's rank correlation was used to calculate intra-individual links.Due to the small sample sizes (Group 5/7 n = 95; Group 3/5 n = 220), the proportion of students in each class (six to eight students), and the number of classes (50 classes), we ran neither multilevel nor any other analyses targeting class-specific results.

Exploratory research question 1: CLI in English and French textscomparison between the two groups
Table 1 shows that, overall, CLI is not very frequent: on average, there was approximately one occurrence per text.The standard deviations were high because CLI was distributed unevenly across the texts and, in the texts of group 5/7, varied between zero and seven in French and English; for the 3/5 group, they varied in French between zero and 17 and in English between zero and eight.
A higher average number of CLI was observed in the French texts of group 3/5 and in the English texts of group 5/7.There were also significant differences when CLI was calculated relative to text length.Pupils in group 3/5 wrote longer texts in both languages than did those in group 5/7, and the difference was even more marked in English (see Table 1).It is also worth mentioning that, on average, the English texts of the two groups were longer than the French texts (Z = −10.152,p < .001for group 3/5 and Z = −5.754,p < .001for group 5/7), even though the pupils in group 5/7 had been learning English for only one year (French for three years, like the other group).
A more comprehensive analysis of the insertions and lexical creations reveals the trends below.
As Figure 1 shows, in both groups there were, on average, more insertions of lexical units from other languages in French than in English texts.The difference between French and English was significant in group 3/5 (Z = −5.875,p < .001;Z = −6.735,p = < .001with consideration of text length).In group 5/7 it was significant only when text length was considered (Z = −1.995,p = .046),but not when it was not (Z = −1.749,p = .08).
A comparison of the French texts of the two groups (see Table 1) reveals that insertions were more frequent in group 3/5 than in group 5/7, but although the difference between the total insertions is clearly visible in Figure 1, it was not statistically significant.That said, distinguishing the insertions according to the source languages reveals that group 3/5 used significantly more English words in French texts than did group 5/7.The difference in the average number of German word insertions between the two groups was not significant; insertions of units that could be attributed to a possible influence of two or more languages or insertions of other languages were rare and the differences not statistically significant.Calculations relative to text length tended to reduce the differences but did not lead to substantial changes in the results (see Table 1).
The comparison of English texts shows that the total number of insertions was roughly equivalent in both groups (see Table 1).However, on average, there were significantly more insertions of French words among learners in group 5/7 than in learners in group 3/5.The difference in insertions from German was not statistically significant and there were no units from other languages, or from multiple languages at the same time.
A more qualitative analysis shows that, among the English words inserted into the French texts of group 3/5, there was a greater variety of function words, particularly in terms of prepositions (at, by, near, behind, for, with, on), determiners (the, a), and connectives (then).In the texts of the group 5/7 pupils, we found only determiners (a, one) and a negative particle (no).Regarding content words, we primarily observed cognates in group 5/7 in the three languages (music [musique, Musik], class [classe, Klasse], guitar [guitare, Gitarre]), or between French and English (colour [couleur]), and several verbal form occurrences (mostly is).In the French texts of group 3/5, content words from English also included the same cognates in the three languages.There were also other cognates such as class [classe, Klasse] which appeared frequently, balcony [balcon, Balkon], garden [ jardin, Garten], persons [personnes, Personen], and perfect [parfait, perfekt], and a much wider variety of English-German cognates such as room [Raum, chambre] which appeared six times, and English-French cognates such as mirror [miroir, Spiegel] and many verbs (e.g.look, next to am and has).Words borrowed from German were very similar in the two groups: varied content words (Pult [desk], Zimmer [bedroom]) and, among German function words, connectives (und [and]) and prepositions (an [to], aus [ from]).In the words classified as borrowed simultaneously from English and German, we found only the word computer in both groups.In group 3/5, alongside the word computer, we also found the words pink and pool.
If we focus on the lexical creations based on other languages, Figure 2 reveals that, for both groups and both languages, Germanwhich was the language of instructionhad the greatest impact.
A comparison of the lexical creations in the French texts of the two groups reveals more creations in the 3/5 group than in the 5/7 group; the difference was weaker but remained significant when text length was considered (see Table 1).Regarding the influence of German, the difference was statistically non-significant and was even less important when considering text length; however, the influence of English was statistically significant when both calculation methods were used.The differences regarding the simultaneous influence of multiple languages were not significant, nor were those due to the influence of other languages.
In the English texts, there were significantly more lexical creations among learners in group 5/7 than among those in group 3/5, influenced primarily by German; the influence of German appeared even stronger when considering text length.The other differences (influences of French, of multiple and of other languages, see Table 1) were not significant.As mentioned above, the 'lexical creations' category included lexical inventions (foreignisings), semantic calques, and compositions.The calculation of percentages provides further clarification Table 2.In the French texts of group 3/5, there was a notable proportion of lexical inventions: either from German, for example zimer [Zimmer, chambre], placat [Plakat, affiche], or from Englishwale [wall, mur], vieu [view, vue], chaire [chair, chaise], coleur [colour, couleur]or from both German and English, for example ven [when, wenn, quand], mei (my [mon] with spelling conventions of German).In group 5/7, only one invention borrowed from English (chaire for chaise (chair)) while some inventions drew on German, for example vant [Wand, mur (wall)], cimmer [Zimmer, chamber (room)].However, no inventions borrowed from both German and English.Group 3/5 used more original calques, such as bâton (stick) for étage (floor) [Stock means both stick and floor in German].Lastly, concerning the French texts: both groups used semantic calques, the most common being regarder [used in the sense of see, in German, the verb sehen referring to see and look], as well as compound words, especially in group 5/7 -fleur sticker [flower sticker, fleur adhésive]and in group 3/5, such as siège-sac or sitsac [Sitzsack, beanbag, pouf poire], hautlit [Hochbett, loft bed, lit mezzanine], aimerais-couleur [Lieblingsfarbe, favourite colour, couleur préférée] and several occurrences of en front de [in front of, devant].
In the English texts, both groups used a variety of compound words modelled on German: eattable [Esstisch, dining table, table de salle à manger], sitbag [Sitzsack, beanbag, pouf poire], applbaum [Apfelbaum, apple tree, pommier], highhouse [Hochhaus, skyscraper, gratte-ciel].However, this structure was used more in English than in French for group 5/7.Both groups used a variety of semantic calques, notably when the proximity at the signifier level was clear (for example, the sea [der See instead of the lake, le lac] or to see [sehen, voir] used in the sense of to look).There were also lexical inventions with no linguistic relationship and including the same inventions as those used in French: vant [Wand for wall, mur], cimmer [Zimmer for room, chambre], as well as inventions with a linguistic relationship: jahrs [years, Jahre, années], or those using words existing in the target language: kisses [Kissen, instead of cushions, coussins].The impression that learners in group 5/7 were more creative in English than in French is confirmed by the values in Figure 2.Although the difference was significant (Z = −2.292,p = .022),it was statistically non-significant when text length was considered, (Z = −1.310,p = .190).For group 3/5, the opposite trend was observed (Z = −2.048,p = .041)and the difference was more clearcut when text length was considered (Z = −4.515,p < .001).

Exploratory research question 2: intra-individual correlations
As shown in Table 3, the correlation between the length of texts written in French and English was strong in both groups and higher for group 5/7.This difference was close to the threshold for statistical significance.Taken together, the correlations for CLI were weaker, albeit still significant, and there was no divergence between the two groups.The correlation in the two groups was higher for lexical insertions than for lexical creations: learners who borrowed in one language tended to do so in the other language as well, with the correlation stronger in group 5/7 and the difference between the two groups approaching the threshold for statistical significance; it was also significant when text length was considered (see Table 3).

Summary and discussion of results
The key contribution of this study is its analysis of CLI in texts written in two foreign languages, by the same individuals, in two groups of lower secondary learners, for whom the language learning sequence was changed (L2 vs L3).The start of French teaching being fixed at grade 5, the variation between the two groups was the age at which the teaching of English commenced (grades 3 and 7, respectively).The quantitative and qualitative results indicate patterns of interactions occurring within the plurilingual repertoire, which may possibly be explained by changes in the learning conditions.In response to the first exploratory research question, comparing the two groups revealed that the CLI phenomenon was more often perceptible in the French texts of pupils who had been learning English since the third grade (group 3/5).In the English texts, they were more common among pupils who had been learning this language for only a year (group 5/ 7).The findings concerning English as L2 vs L3 are consistent with the results found by Pfenninger and Singleton (2016).It should be noted that, in general, CLI was relatively rarely used and observed among only a small proportion of the pupils, confirming the findings of other studies (notably Cenoz, 2001;Agustín-Llach, 2014;Bono, 2008).The interpretation of the results must take into account an important feature that was not present in the other studies: i.e. in our study, pupils were allowed to use bilingual dictionaries to overcome lexical difficulties.The differences between the two groups in terms of CLI are thus particularly interesting.
A more in-depth analysis revealed significantly more insertions of English words (including function words) and English-based lexical creations in L3 French texts of group 3/5 than in those of group 5/7.This may be interpreted either in terms of the L2 effect (Dewaele, 1998;Williams & Hammarberg, 1998), the foreign language effect (De Angelis & Selinker, 2001), or prolonged exposure to English and a higher level of language proficiency (Tremblay, 2006).This interpretation is supported by the fact that, on average, the English texts in group 3/5 were significantly longer than those in group 5/7.Moreover, the higher number of insertionsbut not of lexical creationsborrowed from French, identified in the English texts of pupils in group 5/7, corroborates findings from previous studies: indeed, French was the L2 of group 5/7.Rather, more lexical creations resulting from an activation of French would be the sign of a higher level of competence in this language (Pfenninger & Singleton, 2016;Agustín-Llach, 2014).However, in both groups, German was the main source from which the creations were drawn and the French texts of group 3/5 were slightly longer, the exposure to French remaining the same.
By including the calculation of text length in the use of CLI, the differences between the two groups tended to become clearer for the English texts: When divided by the number of clauses, the CLI were on average more frequent in the shorter texts of group 5/7 and less frequent in the longer texts of group 3/5.For the French texts, the calculations relative to text length tended to reduce group differences, but no significant changes were observed.For the within-group comparisons, the differences between English and French were more pronounced in group 3/5, especially for the lexical creations, as they were fewer CLI in the English texts even though these texts were longer.In group 5/7, where the gap between the text lengths was smaller, the differences between the languages were reduced.
Traces of a greater command of English appear in both English and French texts of group 3/5.Indeed, our qualitative analysis revealed a greater diversity of lexical creations in the French textsespecially when we analysed the composition structure used to create compound words in French modelled on compound words in German.The greater experience in English, where the same mechanism is active, may have triggered the use of the composition mechanism in French.This structure, which learners in group 5/7 tended to use only in English for lexical creation, was also activated less in the English texts of group 3/5.Indeed, the latter relied less on the language of instruction because they had a higher level of proficiency in English.
Qualitative analysis also confirmed that both insertions and creations were frequently based on similarities with the signifier, either between two languages or between the three languages (Bono, 2008;Dewaele, 1998;Force-Izzard, 2018).English and German, which are both Germanic languages, share many cognates.However, given its medieval history, a significant proportion of English words has Latin roots, a characteristic shared with French.Lastly, the three languages share many Latin words of a scholarly nature.The relationships between words is more or less easily identifiable depending on the morpho-phonological distance between them (Manno et al., 2020;Pfenninger & Singleton, 2016).
In response to the second exploratory question on intra-individual links in both languages, a calculation of correlations showed that pupils who used lexemes with a detectable cross-linguistic influence in one language tended to do so in the other language as well.This result, consistent with a strong correlation in text length, corroborates the qualitative findings of Cenoz and Gorter (2011) on CLI in multiple target languages and of several studies focussing on textual structure, showing that writing in different languages is based on a shared underlying proficiency (De Angelis & Jessner, 2012;Marx, 2020;Rinnert & Kobayashi, 2016).Among pupils in group 5/7, who were in their first year of learning English, the correlations between CLI in the two foreign languages were stronger than for group 3/5 who had been learning English for five years, the difference being reinforced when CLI was calculated relative to text length.Similar results were reported by Bono (2008, p. 161) who suggested that the distance between languages tends to widen as the learning process progresses.

Limitations
Several limitations must be mentioned and the study must, therefore, be considered as exploratory: The data coding was based on a qualitative, iterative procedure and relatively simple statistical methods were chosen, allowing only indicative trends.Given the relatively small group effects and the large individual differences, it cannot be ruled out that class and teacher-effects also played a role.Future research, must undertake a detailed investigation of teachers' approach to CLI and work with larger sub-samples (approximately 450 students each) from about 40 classes to run multilevel analyses.Indeed, as Maas and Hox (2004) and Akter and Khan (2018) suggest, sampling for multilevel analysis is challenging: only more extensive samples from at least 40 classes with at least ten students in each class can ensure robustness.Although more recent studies (Hox & McNeish, 2020) have used smaller samples in multilevel analysis under certain conditions, the robustness of these studies is not considered to be granted.
Regarding the writing conditions, the use of bilingual dictionaries limits the ability to compare our findings with those of other studies where learners had no external help.This may make it harder to interpret certain results, particularly concerning the differences between the two groups.Future research should assess in greater detail how the use of CLI differs between writers with and without dictionaries (control group) and why learners use CLI to overcome word-finding or -knowledge difficulties, even when they are allowed to use dictionaries.To this end, retrospective interviews may prove beneficial (for a review of recent studies on dictionary use in second language writing, see Lew, 2016).Other decisions relating to our analysis undoubtedly influenced the results, such as the decision to treat German names written in lower case as insertions (borrowings) rather than as creations.Lastly, the possible influence of the plurilingualism of pupils who spoke a different native language at home was not analysed; this will be considered in future studies.

Conclusion and educational perspectives
In some German-speaking Swiss cantons, the introduction of the early teaching of English in primary schools had led to the emergence of a new reality of language learning both for English, which becomes L2 in the school programme, and for French, relegated to L3.Our exploratory analysis of cross-linguistic influence indicates possible changes in the linguistic equilibrium which may go unnoticed if education continues to be conceived from a strictly monolingual perspective.
From a normative point of view, if the traces of the activation of another language are described as errors or negative transfers, CLI of English in French texts can be interpreted as a weakening of French.However, as has been shown above, in the context of research on tertiary languages, these transfers should be considered as traces of the dynamic activation of resources present in the plurilingual repertoire and, at the same time, as a sign of the acquisition of the skills required for language appropriation.As such, from the perspective of pedagogical translanguaging (Cenoz & Gorter, 2019;Leonet et al., 2020), this CLI may be put to use in the teaching of foreign languages, notably L3 French.By drawing learners' attention to their CLI and to their tendency to borrow or create words on the basis of resemblances between the languages present in their repertoire (parallel words), it is possible, particularly at the beginning of learning (Bono, 2008), to develop a strategy that involves using English as a relay or gateway language to French (Jakisch, 2015).In this perspective, CLI would therefore be considered as a resource and as an asset in learners pluriliteracy practices (Moore, 2020), instead of being negatively perceived as signs of incompetence.
However, although it seems necessary to consider these plurilingual writing practices as legitimate, they must not be naively celebrated and taken out of context.Indeed, on the one hand, there is the risk that they will come up against well-established standards of various written modes in a formal situation ('language of distance ' Koch & Oesterreicher, 2008, see discussion in Münch & Noack, 2020).On the other hand, their extensive and careless use can lead to the erosion of a language in a weak position (Cenoz & Gorter, 2019, p. 134).Hence, their inclusion demands systematic pedagogical guidance and requires them to be linked to teaching practices focused on the acquisition of the target language.Cenoz and Gorter (2019) therefore propose the creation of clearly defined breathing spaces in which unilingual norms are consciously eased, but where the metalinguistic and plurilingual consciousness of learners is also systematically reinforced in order to strengthen the catalytic effect of the activation of the language previously acquired during tertiary language acquisition (cf.also Leonet et al., 2020).Encouraging teachers in initial and continuous training to create pluriliteracy events (Street, 2003;Moore, 2020), which simultaneously include English and French, may help them to reconsider French practices exclusively focused on correction and to encourage the casual use of French in a spirit of lingua franca (Seidlhofer, 2017).It may be assumed that, in the Swiss context, the conscious use of borrowings and creations when learning French, drawn from English in particular, will thus eventually be among the plurilingual solutions proposed, which will also include the use of English as a lingua franca, as well as code-switching, translation, and intercomprehension (Backus et al., 2013).Further research involving observation and intervention should investigate the impact of such an approach on learning French as an L3 after English as an L2.Notes 1.While the term foreign language is undoubtedly debatable, it has been traditionally used to designate a language learned by the majority of students in an institutional setting.It is both a subject matter and a working tool for other subjects (cf.Coste, Cavalli et al., 2009).In the concrete case described here, both foreign languages are taught as subjects.2. Several researchers today advocate the use of the concept of transfer in a pedagogical context to name the process of momentary activation, whether promoted or not, of the forms or patterns acquired in another language, transversality thus being the result of a transfer (cf.Marx, 2020).
3. The concept of tertiary language (L3) is not to be understood as the third language learned in a strictly chronological order of acquisition.Rather, it should be perceived as a heuristic concept that makes it possible to grasp the specificities of one or more languages (L3+) learned after one or more L1 and/or one or more L2 in learners' plurilingual repertoires (see, for instance, Hammarberg, 2018, p. 139).This is relevant in the context of this study because many learners bring with them one or more family language(s) in addition to the regional language, characterised by the Swiss German dialect vs. the Standard German diglossia.In this study, the concept is used to distinguish between the first and second foreign languages learned in an institutional setting (L2 vs L3). 4. In the canton analysed, the lower secondary cycle (from the 7th to the 9th grade) distinguishes different types of schools, grouping pupils according to requirement levels: elementary (Realschule), secondary school (Sekundarschule), senior high school (Untergymnasium).5.The letters were sent and the pupils of the partner classes responded.However, the correspondents' letters were not analysed because we had no direct control over the conditions of the written production.The German texts were excluded from this analysis because they did not include CLI. 6.Previous research studies based on these data have focused on transversality in textual structuring (e.g. the use of connectives, see Egli Cuenat, 2022 and Manno et al., 2020).7. The segmentation into clauses (rather than word count) was chosen as a measure of text length because the main objective of the data coding was to analyse the traces of transversality in textual structuring (see note 6).

Figure 1 .
Figure 1.Lexical insertions (idiosyncratic borrowings) in English and French texts: A comparison between the two groups.

Figure 2 .
Figure 2. Lexical creations in English and French texts: a comparison between the two groups.

Table 2 .
Lexical creations: absolute numbers and percentages.