The Influence of Social Networks During Study Abroad: Acquiring Non-Standard Varieties

Abstract

Over the past 20 years, researchers have shown increasing interest in social network analysis to understand second language acquisition (SLA), especially in a study abroad (SA) context. To date, few longitudinal studies have examined the joint evolution of the learners’ sociolinguistic competence and socialisation during the SA. By shifting the focus from a global view of the study abroad context to a deep analysis of the composition and structure of each learner’ social networks in the host country, we aim to provide a better understanding of the development of sociolinguistic competence in SLA (Gautier & Chevrot, 2015). We apply the sociological concept of a social network to sociolinguistics. To explore the sociolinguistic competence of 29 learners, we focus on two well-described sociolinguistic variables in French: the optional liaison and the negative ne. We also gathered data on their social networks and provided a deep analysis of each participant’s network. We implemented a quantitative approach to analyse and depict the social networks of the learners. Statistically significant relationships were found between changes in the learners’ personal network and their use of the two sociolinguistic variables. The development of L2-oriented social networks (in terms of size, speaking time, and frequency) over nine months of the SA helps learners to reduce their use of standard variants. Conversely, the development of L1-oriented social networks during the SA is associated with greater use of standard variants.

1. Introduction

Research in second language acquisition (SLA) has been significantly influenced by emergentist theories, including construction grammar, usage-based approaches, connectionism, and dynamic systems theory (Ortega, 2013). These theories have seen strong development in recent years. Mitchell et al. (2013, p. 99) proposed the following definition to qualify emergentist approaches: “From this perspective, the basic idea is that grammatical rules and other aspects of language ‘emerge’ (that is, are constructed and abstracted) from language use and experience, rather than being either innate, or learned as abstract structures.” (Mitchell et al., 2013, p. 99). Ellis et al. (2013) conceived emergentist thinking about language and language acquisition as the conjunction of simple and complex processes. Simplicity relates to constructional grammar, driven by factors such as input frequency or salience, while complexity refers to a dynamic interaction of multiple factors that shape language competence and its conceptualization. The acquisition of first language (L1) and of a second language (L2) is said to be constrained by various psycholinguistic factors, such as the salience or frequency of linguistic forms present in discourse. Numerous studies, for example, have demonstrated the frequency effects at work at several levels in the language acquisition process (Bybee & Hopper, 2001; Ellis, 2002).

Linguistic structures are therefore emergent, governed by certain regular processes, but changeable because they are recreated by individuals in specific conditions of usage (Kemmer & Barlow, 2000; Ellis & Larsen-Freeman, 2006). The language environment is thus the real source from which individuals build their language skills. As a result, the acquisition of a specific expression or word comes from direct contact with its conditions of usage. Exposure to these conditions in the native community will therefore have a positive influence on L2 learners, as it enables them to observe and practice choices and decisions specific to native speakers of the language (Achard, 2008).

Thus, for emergentist theories, the social environment is one of the keys to understanding second language development, but empirical studies on this topic are still rare. One important question raised by such an approach is as follows: How can we measure an individual’s social environment? Social network analysis (SNA) offers a comprehensive view of the social environment and a formal method to determine its influence on language usage. Beckner et al. (2009) recommended this approach to better understand how acquisition processes can be deduced from the structure of networks. SNA aims to highlight the influence of networks on individual behaviour and attitudes. It also offers a methodological approach based on several measures that make it possible to account for connections between individuals in an analytical way.

The study abroad (SA) context is particularly conducive to the study of the relationship between the language environment and language acquisition. The social and linguistic environment in which learners immerse themselves for a year provides a unique setting to observe their evolving social network. The SA is often regarded as an opportunity for rich socialisation in a foreign language, fostering the development of various linguistic, sociolinguistic, and pragmatic skills. This study will examine the relationship between social networks developed in an SA context and the learners’ language use. To this end, we are particularly interested in the development of sociolinguistic competence. We want to understand how learners acquire the use and social value of the native sociolinguistic variants to which they are exposed in the host community. Language classroom learning essentially advocates standard language use for a great number of sociolinguistic variables (Mougeon et al., 2010; Regan et al., 2009; Dewaele & Regan, 2001; Gautier & Chevrot, 2021). Variationist research in the SA has shown that learners tend to reduce their use of standard variants during a stay abroad (Gautier & Chevrot, 2015, 2021; Thomas, 2004; Sax, 2003). Thus, it is generally accepted that the SA context offers extensive exposure to an L2 and numerous interactions with native speakers, which often leads to linguistic change.

Recent studies have attempted to define the exact composition of the learners’ social environments and to link them to language acquisition. Mitchell (2023) provided a review of various approaches to documenting the SA input and interaction. She shows how studies have evolved from self-report and indirect approaches through questionnaires and language logs to modelling social networks and direct approaches in order to document input and interaction. A growing body of research has demonstrated the utility of SNA in tracking sociolinguistic, phonological, and morphosyntactic acquisition among learners in diverse SA contexts. Kennedy Terry (2017, 2022) has conducted extensive research on the intersection of SLA and sociolinguistics, emphasising the predictive power of social networks during the SA. She has demonstrated that stronger social networks with target-language speakers during the study abroad contribute to the acquisition of sociolinguistic competence. The LANGSNAP project has been instrumental in advancing SNA within the SA research. McManus (2019) investigated the relationship between social networks and language development, demonstrating that stronger host community ties correlate with greater linguistic gains. Earlier studies from LANGSNAP researchers (McManus et al., 2014; Mitchell et al., 2017) have examined how English-speaking sojourners integrate into French-speaking environments, emphasising the role of social relationships in facilitating or hindering language learning. Additionally, Mitchell et al. (2015) explored how different types of placements during the SA impact linguistic progression, revealing that learners embedded in local social structures exhibit more substantial linguistic development. These studies have underscored the success of SNA in analysing linguistic acquisition during the SA. By mapping the learners’ social environments and interactions, researchers have been able to provide a nuanced understanding of the mechanisms that drive language learning beyond formal instruction. The consistent findings affirm SNA’s value as a methodological approach in SLA research, particularly in the SA setting.

In this article, we aim to explore how aspects of social networks during the SA are linked to changes in the use of two well-known sociolinguistic variables in French: the optional liaison and the omission of the negative ne.

2. Materials and Methods

2.1. Participants

The participants were 29 learners of French as a foreign language, spending two semesters in France at the University of Grenoble (Gautier, 2016). They attended the same course at a French-language-learning centre (Centre Universitaire d’Etudes Françaises), where they received fourteen to sixteen hours of French instruction per week, including language and literature/culture components. The participants’ mean age was 21.6 (range: 18–25 years). Half of the students (14) came from the United States and were all native English speakers, while the other half (15) were from China, and all were native Mandarin speakers. At the start of the course, the students’ proficiency in French varied slightly; approximately three quarters of the students were placed at the intermediate level (B1), and the other quarter were placed at the upper intermediate level (B2) according to the Common European Framework of Reference for Languages (Council of Europe, 2001). The French language-learning centre assessed their proficiency at the beginning of each semester. The learners volunteered to participate in the project but were not informed of its specific aim.

2.2. The Optional Liaison

The phenomenon of liaison is common in French speech and occurs when a consonant is produced between two words, if the second word begins with a vowel. Liaison generally forms a syllable with the following consonant. For instance, when the adjective petit (‘small’) is combined with the noun arbre (‘tree’), the sequence can be pronounced /pətitaʁbʁ/. The liaison consonant /t/ appears when the two words are combined. A limited number of consonants are used for liaison: /z/, /n/, /t/, /r/, /p/, and /z/, /n/, and /t/ being the most common (Adda-Decker et al., 1999). Liaison contexts are categorised as either categorical or optional. The optional liaison is a stratified sociolinguistic variable in adult speech. Its realisation varies with speaking style (Ahmad, 1993) and the speakers’ socio-economic background. For example, De Jong (1991) reported that upper-middle-class speakers produce more optional liaisons than lower-working-class speakers.

The distinction between categorical and optional liaisons proposed by Durand and Lyche (2008) was applied in this study. The optional liaison contexts reflect the usage of native French speakers, based on a recent analysis of a large corpus. The contexts include the following:

**

After a pre-nominal adjective (e.g., petit ordinateur /pətitɔʁdinatœʁ/ vs. /pətiɔʁdinatœʁ/ ‘small computer’)

**

After a plural noun (e.g., amis heureux /amizøʁø/ vs. /amiøʁø/ ‘happy friends’)

**

After a form of the verb avoir (e.g., avaient essayé /avɛtesɛje/ vs. /avɛesɛje/ ‘had tried’)

**

After a form of the verb être (e.g., c’est à toi /setatwa/ vs. /seatwa/ ‘it’s yours’)

**

After another verb form (e.g., ils fument aussi /ilfymtosi/ vs. /ilfymosi/ ‘they also smoke’)

**

After invariable words, such as prepositions or adverbs (e.g., chez elle /ʃezɛl/ vs. /ʃeɛl/ ‘in her house’)

The realisation of the optional liaison is considered a standard, sometimes prestigious, form, while non-liaison is viewed as non-standard. A study by Adda-Decker et al. (2012) analysed the optional liaison, based on a corpus of recordings of 46 French speakers conversing with friends (Torreira et al., 2010). The speakers of this study (students from the Paris region) rarely used the optional liaison in casual conversations with friends. Based on the same optional liaison contexts as those mentioned above (Durand & Lyche, 2008), the authors note that the rate of the optional liaison does not exceed 11% in the everyday speech of French speakers.

2.3. The Negative ne

Negation in French is expressed using a pre-verbal ne, a verbal form, and a post-verbal particle (such as pas ‘not’ or jamais ‘never’), the most frequent being the item pas (i.e., je ne sais pas; ‘I don’t know’). Although it is required in written speech, in spoken French, the pre-verbal ne is often omitted. Coveney (1996) considered the negative ne to be the most extensively studied and known sociolinguistic variable in French and its omission depends on speaking style and the speakers’ socio-economic background, similar to the realisation of the optional liaison. The research has shown that age also has an influence on the omission rate of ne. For example, Berit Hansen and Malderez (2004) observed the maintenance rate of ne in different age groups. A maintenance rate of 22.3% was observed for people between 51 and 64 years old, while a maintenance rate of 7.5% was observed for people between 24 and 35 years old. Social class has also proved to be a predictive factor in the use of ne. By categorising speakers according to three social classes, Ashby (2001) found that speakers of a low socio-economic background have a slightly lower rate of use than the two other social classes. Similarly, Coveney (1996) observed that lower-class speakers maintain ne at a rate of 8.2%, while upper-class speakers maintain ne at a rate of 16.4%. Despite the differences, the omissions rate reported in these different studies indicate a very low use of the negative ne in the everyday speech of French speakers.

2.4. Acquisition of the Optional Liaison and ne by Second Language Learners in the SA

The SA context has become more widely recognised in the field of SLA, as one which provides good access and exposure to sociolinguistic variables in the target language community. Different studies have investigated the impact of the SA on the acquisition of variation, in the form of both cross-sectional studies (comparing groups at home and groups in the target language community) and longitudinal studies of the evolution of sociolinguistic competence before, during, and after the stay abroad (Lemée, 2002; Sax, 2003; Regan et al., 2009; Dewaele, 2004b). These studies have shown that L2 learners who spend time abroad increase their knowledge of non-standard variants and use them at a higher rate than students who have never spent time abroad. It has also been found that L2 learners underuse non-standard variants even with naturalistic exposure to L2, compared to native speakers.

Furthermore, it has been established that the acquisition of categorical liaison is rapid and that advanced learners have a good understanding of this phenomenon (Racine & Detey, 2015). The optional liaison represents greater difficulty because, in addition to the phonological, lexical, and morphosyntactical factors, learners need to grasp the sociolinguistic aspects. Howard (2004) and Thomas (2002, 2004) have observed that the learners’ use of the optional liaison is influenced by two opposing tendencies: maintaining high levels of liaison due to its academic value and decreasing its use to align with native speaker norms. Researchers have been interested in how learners are influenced by the language environment in which they participate. Some studies have compared groups of learners in a language classroom context to groups of learners in an SA context. Thomas (2015) found that the group who had spent ten months in France came closer to nativelike use of the optional liaison. Despite interindividual differences, learners tend to decrease their usage of variable liaison after a stay abroad in France (Howard, 2013; Thomas, 2002).

For the omission of the negative ne (Dewaele, 1992; Regan, 1995; Regan et al., 2009), the rate of ne production depends on the frequency of contact with native French speakers. The rate of learners (11%) is close to that of native speakers for those who are in contact with the latter (Thomas, 2004). However, it can be much higher (up to 81%) for learners with no contact with native French speakers outside the classroom (Thomas, 2004). Regan (1995) showed in a longitudinal study that ne acquisition is influenced by linguistic and extralinguistic factors. She found that, at the end of a year of an SA, learners approach the omission rate of native speakers. This was confirmed in another study (Regan et al., 2009), where the authors noted that the SA learners are more likely to omit the negative ne at a rate similar to that observed in native speakers’ speech. Students who had never been abroad did not adapt their omission rate depending on the situation of communication (Dewaele, 2004a; Regan et al., 2009; Sax, 2003). Prolonged everyday use of French with native speakers seems to foster the development of stylistic variation. However, even if the students decrease their usage of standard variants after the SA, they maintain a higher retention rate than native speakers.

To summarise, studies of L2 acquisition of variable liaison and ne omission are consistent in their findings in two respects. Firstly, the SA context has a positive impact on the L2 learners’ use of sociolinguistic variables, as they evolve from formal to more informal usage. Secondly, even if the learners move towards a less formal usage, they do not generally attain the vernacular norms of native speech.

3. Procedure

Data for this longitudinal study were collected over a period of nine months, with three major data collection points scheduled in September (time 1 [T1]), January (time 2 [T2]), and May (time 3 [T3]). At each data collection point, we scheduled two different meetings with participants, and each observation period followed the same process. Firstly, we conducted an interview to elicit the learners’ sociolinguistic usage. At the end of the interview, they were given a logbook in which to record all their social interactions over the course of a week. Secondly, at the end of that week, we conducted a second interview during which the learners shared information on their personal networks, based on the information recorded in their logbooks throughout the week.

All the interviews were conducted in French on a one-to-one basis, between a participant and one of the researchers involved in the study who is a native French speaker.

3.1. Interviews for Eliciting Sociolinguistic Usage

Sociolinguistic usage was elicited using interviews, which aimed to prompt natural, spontaneous discourse. Informal conversational topics were used, guiding the learners to speak in an unmonitored style, and were chosen to reflect the learners’ interests, such as hobbies and pastimes, holidays, and social life in France. The interviews lasted a mean length of 43 min and were then transcribed into standard orthography using CLAN software (MacWhinney, 2000). Specific codes were included for the realisation or non-realisation of ne and the optional liaison. For the statistical analysis of the usage of the two sociolinguistic variables we used SPSS Statistics version 20.0.1

3.2. Instruments for Social Networks Data Collection

Two types of data were used to assess the participants’ social interactions during the SA period.

3.2.1. Completing a One-Week Logbook

To obtain a picture of each participant’s personal network, first a logbook was used as a name generator, which allows the participants themselves to create a list of contacts with whom they interact on a regular basis. As the sociologists Degenne and Forsé (2004) explained, the choice of a name generator depends on the underlying research question. This study sought to provide a broad view of the different relationships that could be established in the SA context. A name generator was therefore chosen, as it made it possible to capture the learners’ interactions as experienced on a day-to-day basis. A logbook approach is more natural and less intrusive than direct observation (Fu, 2007). Moreover, as Degenne and Forsé (2004) explained, the use of a logbook has a clear advantage over verbal reports, as it minimises both memory bias and any retrospective rationalisation on the part of the participants.

The participants had to fill in the logbook every day for one week. The question of how long learners are asked to keep a logbook is a sensitive issue. Too short a period could lead to the logbook being insufficiently representative, but overly too long a period, the participants could get bored and lose motivation and even give up. In previous surveys using logbooks, the timeframe has ranged from one week to one hundred days (for a review of various logbook studies, see Degenne & Forsé, 2004 or Fu, 2007). For this study, the week-long period was agreed upon for two main reasons. First, the logbook could potentially take between 20 and 40 min to fill out per day (Fu, 2007), which translates into approximately 2.5–5 h per week. This seemed to be a reasonable time commitment to ask of the students. Second, other significant studies have also used week-long logbooks to study personal networks (Héran, 1988; Blanpain & Pan Ké Shon, 1998), and produced interesting results with this timeframe.

Instructions on how to fill in the logbook were given orally, and a written reminder was present on the front page. Learners were asked to keep a record of all face-to-face conversations and phone (or video) conversations that they had during the day, in all languages and with all speakers. The logbook included tables to be completed on each page with the following information: conversation, first name(s) of interlocutor(s), duration, place, and language. Learners were asked to fill in the logbook at regular intervals during the day, rather than all at one time. This instruction was given by pointing out that the ideal way to fill out the logbook was just after an interaction. Learners were asked to exclude from the logbook time spent in class, because the focus of this study was on the interactions that took place outside the classroom in their daily lives. The logbooks were given to the learners at the end of the first meeting in each observation period. Learners were also asked to be as rigorous and accurate as possible when completing the logbook, and an initial reading of the data suggests that they followed this instruction. The number of conversations recorded per day for all learners varied from 0 to 57, with an average of 8. This average number of conversations remained relatively constant for each of the three time points (T1: 8.79, T2: 8.39; T3: 7.60) with a slight decrease at T3. The relative stability of the average number of conversations seems to confirm that the students were conscientious in keeping the logbooks in a regular manner.

Following the logbook-keeping week, interviews about personal networks were conducted using the following methodological approach.

3.2.2. Interviews on Personal Networks and Social Practices

The duration of the interviews depended on the number of people present in each learner’s personal network, but varied from 30 min to 1.5 h. A computer aid was used to guide these interviews, where all the information provided by the learners was recorded.

As an initial step, we listed all the people mentioned in the logbook. Then, a series of questions were asked about each of these people, which can be summarised as follows:

• Personal characteristics: age, sex, language used, and nationality.
• Location of the relationship: in the host country or in the country of origin.
• Type of relationship: family, friend, acquaintance, etc.
• How often they met that person on a six-point scale, ranging from every day to never. We specified that they had to provide a general frequency for the immediate period.
• Type of activities they shared with that person.
• How long they had known each other.
• People they knew in common. We specified that this referred to people who could interact with each other even in the learner’s absence, in order to clarify the question and limit the length of the individual lists.

4. Results

4.1. Sociolinguistic Variables

4.1.1. The Optional Liaison

The

Table 1. Percentages of standard variants produced out of the total of occurrences of the optional liaison according to learner and longitudinal time point.

		T1		T2		T3
Learners	Nationality	Percentage of Optional Liaison Realised (Number/Total Number of Occurrences)		Percentage of Optional Liaison Realised (Number/Total Number of Occurrences)		Percentage of optional Liaison Realised (Number/Total Number of Occurrences)
AND	American	80.90%	(72/89)	63.93%	(39/61)	42.37%	(25/59)
BEV		50.00%	(1/2)	83.33%	(5/6)	44.44%	(4/9)
EMI		28.57%	(2/7)	33.33%	(13/39)	31.25%	(5/16)
HEA		81.48%	(22/27)	75.00%	(21/28)	50.00%	(13/26)
JAC		26.87%	(36/134)	31.82%	(28/88)	23.94%	(45/188)
JEF		60.71%	(34/56)	61.18%	(52/85)	47.50%	(38/80)
KAT		26.15%	(17/65)	36.84%	(49/133)	14.91%	(17/114)
KRI		63.77%	(44/69)	50.60%	(42/83)	47.22%	(34/72)
MAT		59.15%	(42/71)	63.74%	(58/91)	44.44%	(28/63)
MEL		68.42%	(13/19)	51.22%	(21/41)	31.67%	(19/60)
MIC		20.00%	(2/10)	60.00%	(9/15)	11.76%	(2/17)
ROB		83.72%	(36/43)	38.10%	(16/42)	22.73%	(5/22)
SAM		20.37%	(11/54)	36.17%	(34/94)	26.67%	(20/75)
SEA		32.14%	(9/28)	33.33%	(26/78)	30.88%	(21/68)
CHE	Chinese	66.67%	(14/21)	51.61%	(16/31)	62.50%	(5/8)
HAI		30.30%	(10/33)	26.67%	(8/30)	28.57%	(6/21)
HUA		57.50%	(23/40)	25.81%	(24/93)	28.89%	(13/45)
HUI		11.67%	(7/60)	23.36%	(25/107)	29.03%	(18/62)
LEI		28.57%	(8/28)	43.75%	(21/48)	27.42%	(17/62)
LUO		10.81%	(4/37)	21.62%	(8/37)	12.12%	(4/33)
MEN		25.00%	(7/28)	66.67%	(10/15)	25.00%	(8/32)
RON		67.31%	(35/52)	75.86%	(44/58)	81.03%	(47/58)
SHU		36.00%	(18/50)	22.62%	(19/84)	43.75%	(35/80)
WEN		48.21%	(27/56)	46.15%	(18/39)	67.86%	(19/28)
WUY		25.61%	(21/82)	35.77%	(44/123)	25.61%	(21/82)
XIE		75.00%	(36/48)	81.52%	(75/92)	69.05%	(58/84)
YAP		57.50%	(23/40)	57.78%	(26/45)	42.86%	(18/42)
YAX		24.29%	(17/70)	27.12%	(32/118)	10.09%	(11/109)
YIN		16.67%	(7/42)	2.86%	(1/35)	18.00%	(9/50)
Mean (Standard deviation)		44.25% (23.04)		45.79% (20.37)		35.92% (18.04)

provides our results for the optional liaison, presented as percentages of standard variants (realised liaison) produced by each learner (n = 29) at the three time points. The table shows the results sorted by nationality, then alphabetically. The realisation percentages were calculated by dividing the number of occurrences of the optional liaison realised by the number of occurrences of possible liaison for each subject.

The mean rate of realisation of the optional liaison decreased markedly between T1 and T3, and this difference was significant (Wilcoxon: z = −2.258, p = 0.002, with a medium effect size, r = 0.42). The difference was mostly noticeable between T2 and T3 (Wilcoxon: z = −2,714, p = 0.007, with a large effect size, r = 0.5), the difference was not significant between T1 and T2.

4.1.2. The Negative ne

The

Table 2. Percentages of standard variants produced out of the total of occurrences of negation according to learner and longitudinal time point.

		T1		T2		T3
Learners	Nationality	Percentage of Negative ne Realised (Number/Total Number of Occurrences)		Percentage of Negative ne Realised (Number/total Number of Occurrences)		Percentage of Negative ne Realised (Number/Total Number of Occurrences)
AND	American	93.83%	(76/81)	70.83%	(34/48)	35.71%	(30/84)
BEV		83.33%	(5/6)	92.31%	(12/13)	90.91%	(20/22)
EMI		31.82%	(7/22)	53.33%	(32/60)	66.67%	(18/27)
HEA		15.79%	(6/38)	3.33%	(1/30)	0.00%	(0/46)
JAC		12.84%	(14/109)	3.48%	(4/115)	0.00%	(0/198)
JEF		72.55%	(37/51)	72.73%	(72/99)	43.64%	(24/55)
KAT		85.90%	(67/78)	60.15%	(80/133)	50.88%	(58/114)
KRI		95.45%	(63/66)	77.61%	(52/67)	81.82%	(54/66)
MAT		79.27%	(65/82)	27.27%	(24/88)	18.18%	(20/110)
MEL		97.78%	(44/45)	34.00%	(17/50)	14.02%	(15/107)
MIC		33.33%	(5/15)	38.10%	(8/21)	28.57%	(6/21)
ROB		77.78%	(21/27)	18.42%	(7/38)	23.53%	(4/17)
SAM		97.78%	(44/45)	89.02%	(73/82)	54.24%	(32/59)
SEA		75.00%	(12/16)	19.70%	(13/66)	8.89%	(4/45)
CHE	Chinese	38.46%	(15/39)	13.89%	(5/36)	4.17%	(1/24)
HAI		78.57%	(22/28)	46.88%	(15/32)	79.31%	(23/29)
HUA		93.48%	(43/46)	70.37%	(95/135)	36.71%	(29/79)
HUI		40.91%	(27/66)	28.04%	(53/189)	51.90%	(41/79)
LEI		88.24%	(15/17)	42.37%	(25/59)	22.81%	(13/57)
LUO		73.08%	(38/52)	79.25%	(42/53)	53.45%	(31/58)
MEN		61.90%	(26/42)	66.67%	(32/48)	53.95%	(41/76)
RON		82.05%	(32/39)	48.39%	(30/62)	63.49%	(40/63)
SHU		1.59%	(1/63)	0.00%	(0/70)	0.96%	(1/104)
WEN		72.00%	(36/50)	75.00%	(39/52)	58.82%	(20/34)
WUY		79.22%	(61/77)	73.05%	(103/141)	74.49%	(73/98)
XIE		65.45%	(36/55)	62.96%	(85/135)	55.88%	(57/102)
YAP		83.02%	(44/53)	72.97%	(54/74)	81.08%	(30/37)
YAX		22.00%	(11/50)	16.98%	(18/106)	26.87%	(18/67)
YIN		49.23%	(32/65)	42.65%	(29/68)	32.26%	(20/62)
Mean (Standard deviation)		64.88% (28.23)		48.27% (27.46)		41.83% (27.03)

presents the percentages of standard realisation of the negative ne (ne retention) produced by the 29 learners at three time points as well as the number of occurrences. The table shows the results sorted by nationality, then alphabetically.

Similar to the optional liaison, there was a notable overall decrease in the rates of realisation of the standard variant of ne negation between the time points. In contrast to the optional liaison, where the reduction mainly occurred between T2 and T3, the decline in the ne realisation was evident as early as T2. There was a statistically significant decrease of 16.61% (Wilcoxon, z = 3.525, p = 0.001, with large effect size, r = 0.65) between T1 and T2, followed by a significant decrease of 6.44% (Wilcoxon, z = 1.979, p = 0.04, with medium effect size, r = 0.37) between T2 and T3.

No significant differences were found between the American and Chinese learners in terms of the rates of realisation of the two sociolinguistic variables (Gautier, 2016).

4.2. Personal Networks Analysis

In this study, we use personal (or egocentric) network analysis, which focuses on the social ties surrounding a single individual, in contrast to complete network analysis, which examines all the connections among members of a defined group. The following section outlines the indices used to define the networks in question. The structural and composition variables were selected based on previous work on social network typologies (Bidart et al., 2011; Brandes et al., 2010; Lubbers et al., 2007).

4.2.1. Structural Aspects of Personal Networks

This research draws on the mathematical analysis of social networks usually conducted in sociology. To determine network structure, many analysts use adjacency matrices to calculate structural indices, such as density and centrality measures. From now on, we will call the learner whose personal network is being analysed ego, and we will call each of this learner’s contacts alter.

• Size and number of ties: an essential criterion is the size of the personal network, i.e., the total number of alters cited by ego. The number of links between the alters is also a basic indicator of how alters are connected within the social network.
• Density: The simplest structural parameter is density. It represents the proportion of existing relationships to the number of possible relationships across the network. In a non-directed network, the number of possible links corresponds to n (n − 1)/2 (with n corresponding to the number of points of the network, i.e., to the number of alters). Density is generally expressed as a proportion. The higher the density, the stronger the cohesion within the alters’ networks.
• Measures of centrality: Centrality is defined by several measures that are complementary to density. In social network analysis, researchers have observed that some alters play a ‘more important’ role than others. Some people may have many contacts within the network while others have very few. Centrality creates the link between the general structure of the network and the specific position of each of its members (Degenne & Forsé, 2004). In this analysis, the most used centrality measures have been taken into account (Bidart et al., 2011; Borgatti et al., 2013), namely betweenness, closeness, and eigenvector centrality. Betweenness (Freeman, 1979) is —“the extent to which a particular point lies ‘between’ the various other points in the graph” (Scott, 2000, p. 86), while closeness (Freeman, 1979) is the proximity of each individual to the others in the network, and eigenvector centrality is the degree to which an individual is connected to highly connected individuals (Bonacich, 1972).
• Isolates: when an alter is not connected to any other network member. This indicator makes it possible to determine whether the network is more or less interconnected.

Density and measures of centrality were calculated using the UCINET (Borgatti et al., 2002), which is a software package for analysing social network data.

4.2.2. Compositional and Interactional Aspects of Personal Networks

Following Lubbers et al. (2010) and Brandes et al. (2008), the network composition was described using four classes of alters based on the types of international relations.

• Originals: alters from the same country as ego, living in the country of origin.
• National peers: alters from the same country as ego, living in France.
• Hosts: alters who come from and live in France.
• Transnationals: alters from other countries, living in France.

For each class, the percentage of alters present in the network was calculated. We also measured the strength of ties based on the frequency of interaction, the number of shared activities, and the length of the relationship.

For the interactional aspects of the personal networks, the two indicators considered were the language mostly used in conversations between ego and each alter and the amount of interaction time reported in the logbook.

4.3. Development of Sociolinguistic Competence and Personal Networks

Previous analyses have shown that the learners’ overall use of the optional liaison and the negative ne decreases over the SA. However, this reduction was not uniform, as some learners increased their use of standard variants or maintained similar rates across time points. In addition, the learners’ social networks showed significant differences in terms of composition, structure, and interaction (Gautier, 2019). Some have social networks primarily centred on French speakers, while others are mostly connected to speakers of their L1. Some have small networks, others larger. Some networks are made up of a large proportion of national peers, and others of a large proportion of hosts. Our aim now is to understand to what extent the personal networks developed during the stay abroad are related to changes in the learners’ sociolinguistic competences.

To analyse the changes in the use of sociolinguistic variables in relation to the development of personal networks, we adopted a two-step approach. First, we summarised the various dimensions describing changes in network indicators by performing a principal component analysis (PCA) on the differences between T1 and T3 (e.g., the difference between the percentage of alter hosts in T3 and that in T1). Based on a correlation matrix, the PCA reduces large sets of variables into a smaller number of factors, each representing a cluster of correlated variables. To gain a deeper understanding of network dynamics, we conducted separate PCAs on the differences in network indicators across structural, compositional, and interactional aspects (see Section 3.2). In our analysis, the extracted factors (e.g., Factor 1 and Factor 2 in

Table 3. PCA of changes in structural aspects of personal networks between T1 and T3.

	Factor 1	Factor 2
Changes in density	0.920	−0.036
Changes in the size (number of ties)	0.740	−0.112
Changes in closeness centrality	0.717	0.557
Changes in betweenness centrality	0.195	0.863
Changes in eigenvector centrality	0.608	0.038
Changes in the number of isolates	0.251	−0.537
Variance	39.64%	22.63%
Cumulated variance	39.64%	62.27%

for structural indicators) represent dimensions capturing the changes in the learners’ network indicators. In a PCA, the contribution of each variable and each individual to the extracted factors is also calculated (Dancey & Reidy, 2007). In our analysis, individual contributions reflect patterns of change in each learner’s social network between T1 and T3.

In the second step, we performed Spearman’s correlations between the learners’ contributions to each factor—representing changes in their network—and variations in their use of sociolinguistic variables between T1 and T3 (see

Table 4. Correlations between changes in structural aspects of networks and changes in the usage of the two sociolinguistic variables.

	Factor 1	Factor 2
Changes in the percentage of the optional liaison realised	0.388 * (29)	−0.226 (29)
Changes in the percentage of the negative ne realised	0.251 (29)	−0.251 (29)

). Changes in sociolinguistic usage were quantified for each of the two variables as the difference between realisation rates at T3 and T1. The difference was computed as T3−T1, ensuring that a negative value indicates a decrease in standard variants, while a positive value signifies an increase.

By examining the correlation between changes in factors representing network indicators and changes in sociolinguistic usage, we can identify which transformations in personal networks are associated with the development of sociolinguistic competence. This approach places greater emphasis on the variation over time rather than on the differences in scores between individuals (Gautier, 2016).

4.3.1. Changes in Structural Aspects of Personal Networks and in Sociolinguistic Competence

The Table 3 presents the results of the PCA on the changes in the structural aspects of personal networks. It reveals two factors explaining 62.27% of the variance. The values corresponding to the variables with the highest contributions to each factor are highlighted in bold. This presentation facilitates the identification of the most influential variables in the construction of the principal components.

The primary variables contributing to the first factor include increases in density, number of ties, closeness centrality, and eigenvector centrality. This factor accounts for 39.64% of the total variance. The four variables are positively correlated. The second factor is mainly determined by changes in closeness centrality, betweenness centrality, and the number of isolates. The negative relationship observed with the number of isolates indicates that, between T1 and T3, a lower number of isolates in the learners’ networks correlates with higher closeness and betweenness centralities. This component explains 22.63% of the variance.

Factor 1 suggests that networks become more interconnected between T1 and T3. Our network analysis indicates that high density often results from strong relationships with national peers. We explored this trend further by correlating the change in density between T3 and T1 with the change in the proportions of various relationship types (hosts, transnationals, national peers, and originals). Only the correlation between changes in density and the proportion of national peers is significant, showing that a denser network corresponds to a higher proportion of national peers in the students’ networks between T1 and T3 (Rho = 0.433, p = 0.01).

Previous studies have shown that a high level of interconnection between the members of a network favours the retention of the same linguistic variants between individuals (Milroy, 1987, 2002). This factor should therefore be linked to an increase in the realisation of standard variants. Conversely, looser networks have been associated with innovation and linguistic change (Milroy, 1987, 2002). We hypothesise that changes in network structure have an impact on changes in the use of sociolinguistic variables. At T1, we know that the learners, as a whole, have high rates of use of standard variants. Table 4 presents the results of the correlations between the factors derived from the PCA representing changes in the structural indicators and changes in usage of the two sociolinguistic variables between T1 and T3. Statistically significant results are indicated in bold for clarity. Asterisks are used to indicate the level of statistical significance associated with each correlation coefficient, allowing a rapid assessment of the reliability of the relationships observed.

Factor 2 does not show a statistically significant correlation with the changes in the use of the sociolinguistic variables between T1 and T3, but Factor 1 is significantly related to optional liaison use (Rho = 0.388. p < 0.05). This result indicates that as density, number of ties, closeness centrality, and eigenvector centrality increase between T1 and T3, the usage of the optional liaison also increases.

Thus, the interconnection of networks with national peers appears to be associated with the use of standard variants. This hypothesis holds only for the optional liaison.

4.3.2. Changes in Compositional Aspects of Personal Networks and in Sociolinguistic Competence

The composition of the personal networks refers to the four types of relationships that ego has with alters, as previously described (see Section 4.2.2).

Table 5. PCA of changes in compositional aspects of personal networks between T1 and T3.

	Factor 1	Factor 2	Factor 3
Changes in % of national peers	0.826	−0.050	0.480
Changes in % of hosts	0.363	0.142	−0.060
Changes in % of original	0.012	0.013	−0.927
Changes in % of transnationals	−0.904	−0.090	0.254
Changes in frequency of interaction	0.096	−0.855	0.273
Changes in length of the relationship	0.218	0.251	−0.286
Changes in number of shared activities	0.183	0.793	0.286
Variance	24.58%	20.78%	19.93%
Cumulated variance	24.58%	45.37%	65.30%

presents the results of the PCA of changes in the compositional indicators of the personal networks. It reveals three factors explaining 65.30% of the variance.

The first factor can be summarised as follows: as the number of national peers in the network increases, the number of transnationals decreases. This factor explains 25.58% of the variance. The second factor is characterised by a decrease in the average of the frequency of interaction and an increase in the average number of shared activities with an alter. It explains 20.78% of the variance. Finally, the third factor is focused on a single variable: the change in the percentage of the originals (alters from the same country as ego, living in the country of origin). It explains 19.93% of the variance. We hypothesise that the learners whose network composition is more oriented towards national peers during their stay (Factor 1) will tend to increase their use of standard variants.

Table 6. Correlations between changes in compositional aspects of networks and changes in the usage of the two sociolinguistic variables.

	Factor 1	Factor 2	Factor 3
Changes in the percentage of the optional liaison realised	0.416 * (29)	0.025 (29)	0.043 (29)
Changes in the percentage of the negative ne realised	0.425 * (29)	−0.059 (29)	0.046 (29)

presents the results of the correlations between the factors derived from the PCA representing changes in the compositional indicators and changes in usage of the two sociolinguistic variables between T1 and T3.

Factors 2 and 3 are not significantly correlated with the difference in the rate of realisation of the standard variants between T1 and T3. Factor 1 exhibits significant positive correlations with the rates of realisation of ne and the optional liaison (for the optional liaison, Rho = 0.416, p < 0.05 and for ne, Rho = 0.425, p < 0.05). Our hypothesis is therefore confirmed: the more networks are oriented towards individuals from the same country of origin, the more learners use standard variants.

4.3.3. Changes in Interactional Aspects of Personal Networks and in Sociolinguistic Competence

In this section, we explore the possible correlations between changes in interactional aspects of the social network and changes in sociolinguistic usage.

Interactional aspects are represented by the percentage of L1 speakers (English or Mandarin speakers), L2 speakers (French speakers), and L1/L2 speakers (those who can use both languages) within the learners’ personal network. These aspects are also defined by the number of hours per week the learners spend using the L1, the L2, or a combination of both (L1/L2).

Table 7. PCA of changes in interactional aspects of personal networks between T1 and T3.

	Factor 1	Factor 2
Changes in % of L1 speakers	−0.898	−0.272
Changes in % of L2 speakers	0.855	−0.355
Changes in % of L1/L2 speakers	0.064	0.978
Changes in interactional time in L1 (number of hours/week)	−0.482	−0.162
Changes in interactional time in L2 (number of hours/week)	0.892	−0.036
Variance	51.36%	23.67%
Cumulated variance	51.36%	75.04%

presents the results of the PCA on the changes in the interactional dimensions of personal networks between T1 and T3. It reveals two factors that explain 75.04% of the variance.

This first factor explains 51.36% of the variance. It suggests that during their stay, as the learners’ networks shift towards L2 speakers, their conversation time in L2 increases, while the number of L1 speakers decreases. The second factor, which explains 23.67% of the variance, is specifically determined by the proportion of L1 and L2 speakers engaged in the same conversation. Our general hypothesis posits that learners who increase both the number and length of their conversations with L2 speakers between T1 and T3 (Factor 1) will likely decrease their use of standard variants during this period.

Table 8. Correlations between changes in interactional aspects of networks and changes in the usage of the two sociolinguistic variables.

	Factor 1	Factor 2
Changes in the percentage of the optional liaison realised	−0.389 * (29)	0.160 (29)
Changes in the percentage of the negative ne realised	−0.488 * (29)	0.211 (29)

presents the results of the correlations between the factors derived from the PCA representing changes in the interactional indicators and changes in usage of the two sociolinguistic variables between T1 and T3.

Factor 1 is significantly and negatively correlated with changes in the rates of realisation of the standard variants for the two sociolinguistic variables (for LF, Rho= −0.389, p < 0.05 and for ne, Rho = −0.488, p < 0.05). Thus, our hypothesis is confirmed: as interactions with L2 speakers increase during their stay, their conversational time in French also increases, and the number of L1 speakers in their network decreases, leading to a decrease in their use of the standard variants of ne and the optional liaison between T1 and T3.

5. Discussion

Analyses linking social networks to L2 acquisition have often been carried out on a small sample of subjects, making results difficult to generalise through statistical analysis. In our study, we followed 29 learners over a nine-month period, collecting data at three time points to track changes in their personal networks and sociolinguistic competence. From collection to analysis of the network data, we used tools from sociological research, which enabled us to qualify and quantify the learners’ social networks from multiple perspectives. By integrating structural, compositional, and interactional indicators into our analysis, we were able to link changes in these variables to the development of sociolinguistic competence using statistical methods.

Our findings demonstrate significant links between the realisation rates of the two sociolinguistic variables (optional liaison and ne omission) and the three network indicators (structural, compositional, and interactional) between T1 and T3. Specifically, these are as follows:

• Network structure: an increase in network density during the stay was associated with a rise in the use of the standard variant of the optional liaison. Denser networks typically include more national peers.
• Network composition: an increase in the number of national peers coupled with a decrease in the number of transnationals during the stay was linked to an increase in the use of the standard variants for both variables.
• Interactional dynamics: an increase in L2 speakers in the personal network corresponding to more interaction time in the L2 and fewer L1 speakers was linked to a decrease of the standard variants of the negative ne and the optional liaison.

In short, learners who developed increasingly dense networks, particularly with more national peers, tended to use more standard French forms. Conversely, those who expanded their networks to include more speakers of French and increased their interaction in French showed a shift toward non-standard usage. The most probable explanation for these trends is that learners who have greater social interaction with L2 speakers are more likely to be exposed to non-standard sociolinguistic variants. Conversely, it can be assumed that those who increasingly socialise with peers of the same national origin remain influenced by the more standard sociolinguistic usage of their language teachers (between 34% and 48% liaison realisation and between 55% and 68% ne realisation, as reported by Gautier, 2016, p. 299).

These findings answer our research question regarding the impact of social networks during the SA on sociolinguistic usage. They also account for some of the individual variation observed in the development of sociolinguistic competence. Future research directions include looking more closely at the interviews conducted with learners to explore how network changes during the SA were sometimes part of a deliberate effort to get closer to native speakers and increase their L2 input. For instance, during the third observation period, an American learner shared how she actively sought to spend more time with French friends to improve her language skills:

“A month ago, I was often with my American friends, and it was just in class that I was speaking French. Then I remember my host mom saying, ‘I think your French is getting worse, what’s happening?’ So, I decided to change, started calling my French friends more, and made more effort. I’m happier now because I’m putting in more effort outside the classroom.”

(MEL) (our translation)

Similarly, a Chinese learner reflected on how living with native speakers improved her ability to communicate in French:

“I don’t know if it’s arrogant to say this, but I think I do better than others when I talk with French people or other foreigners. I’m not afraid or embarrassed anymore. I think living in a flat-share helped—even though it’s not perfect, I’ve learned how to start conversations with French people or foreigners.”

(YAX) (our translation)

Both learners reported significant changes in their social networks during their SA, with an increase in interactions with French speakers. Their experiences suggest that these network changes may be part of a conscious strategy to improve their L2 proficiency. This research provides valuable insights into the link between social network reconfiguration in the host country and changes in sociolinguistic competence during L2 acquisition. The network analysis could be further refined by examining the quality and evolution of relationships—for instance, by distinguishing between strong and weak ties, or considering the stability of these ties over time. Future studies could particularly focus on how enduring, close relationships with native speakers influence sociolinguistic usage.