Actual and Self-Perceived Linguistic Proﬁciency Gains in French during Study Abroad

: The present study focuses on 33 British and Irish students, including non-language special-ists and language specialists, who spent their study abroad (SA) period in Francophone countries. Their proﬁciency in French ranged from lower independent (B1) to advanced level (C2). The analysis of quantitative data collected at the start, in the middle, and at the end of the SA period through an online questionnaire showed that both actual proﬁciency and self-reported proﬁciency increased signiﬁcantly after SA. A closer look at self-reported proﬁciency in the four skills showed a signiﬁcant linear increase in speaking and listening, while scores for reading and writing only increased signiﬁcantly after the mid-way point in the SA period. The same pattern emerged for grammar and vocabulary. Only pronunciation showed no signiﬁcant change over the SA period. Linking the amount of change in actual proﬁciency between the start and the end of the SA period to participants’ descriptions of their experience revealed that progress was not always linked to overall positivity of the experience but rather to the development of a strong local French social network. Actual and self-reported proﬁciency scores were signiﬁcantly correlated. Participants with lower initial actual proﬁciency were found to have made the biggest gain during SA.


Introduction
As the world has become more mobile and interconnected, the phenomenon of studying or working abroad during higher education has become increasingly common. Students have enthusiastically taken up the opportunity to go to another country: in the US, the Department of State reported that 325,339 US students studied abroad in the year 2015/16, constituting around 10% of enrolled students in the US 1 . In Europe, the European Union's Erasmus+ programme has made study abroad (SA) increasingly accessible across the EU, which has led to a significant increase in participation. At its conception in 1987, 3244 students participated. In 2017/18, this figure stood at 325,000 students (European Commission 2018, p. 34). Around half of UK university students who studied abroad did so through Erasmus (https://www.bbc.co.uk/news/education-47293927).
Not all Study Abroad (SA) has a foreign language element, but the present study will focus on SA where improving linguistic proficiency in a target language is the primary aim. The term "study abroad" is used throughout as an umbrella term for all activities undertaken by students while abroad, including internships and teaching assistantships.
While SA is commonly perceived to be a panacea for language students (Watzinger-Tharp 2014), research findings present a more nuanced picture. Research on SA in the context of second language acquisition began in the 1960s. Its initial focus was linguistic gain (Coleman 2009). Evidence of gain turned out to be less clear-cut than expected, with some studies reporting positive effects (Howard 2005;Segalowitz and Freed 2004;Regan et al. 2009), while others found limited or no effect (DeKeyser 1991;Isabelli-García 2006) compared to classroom-based language learning in the home institution. This led Collentine and Freed (2004) to conclude that there is "no evidence that one context of learning is superior to another for all students, at all levels of language learning, and for all language skills" (Collentine and Freed 2004, p. 164). Their conclusion reflects a shift in approach away from viewing SA as a single "monolithic construct" (Devlin 2013, p. 200), where linguistic progress is expected for all participants across the four skills and in all aspects of language (ranging from phonology to pragmatics), simply by virtue of being in a SA context.
One of the ongoing debates in SA research is the optimal timing for students to complete their SA. Do SA students need to have reached some proficiency threshold in order to benefit maximally from their SA (DeKeyser 2014;Hessel 2017;Lafford and Collentine 2006)?
Another question relates to the rate of development of different aspects of language. There is growing evidence that there is an imbalance between how phonology, morphology, syntax, lexis, and pragmatics are affected by SA (Jensen and Howard 2014;Mitchell et al. 2017;Serrano et al. 2012).
One surprising finding in SA studies is the large amount of individual variation in proficiency outcomes in SA (Gass 2017;Jensen and Howard 2014;Kinginger 2008Kinginger , 2011. Possible reasons are complex interactions between various learner-internal and learner-external independent variables, and pure chance. For example, Kinginger's (2008) study of American students studying abroad in France took place at the time of the Iraq war, which was strongly opposed by the French and which forced the students who were often apolitical to defend the United States' geopolitical decisions. It led several students to avoid interactions with French people, restricting their opportunities to use French beyond mere service encounters. This was detrimental to their linguistic gains because, as Briggs Baffoe-Djan and Zhou (2021) point out, linguistic gain during SA depends as much on quantity as quality of language contact.
The current study responds to an ongoing need to measure the effectiveness of SA, both in terms of change in self-perceived and actual target language proficiency. It also answers the call issued by Mitchell et al. (2017) for more studies on target languages other than English, with French, for example, being "comparatively little studied" (Mitchell et al. 2017, p. 73).
The study looks at a relatively homogeneous group of 33 Anglophone students studying French as part of their degree at a British or Irish university, who filled out three successive online questionnaires in 2018-2019, at the start, in the middle, and at the end of their SA in a Francophone country.

Early SA Research
The investigation of the role of SA in second language acquisition began by seeking to understand "the overall efficacy" of SA as a language-learning context (Collentine 2009, p. 219). The first study to do so was Carroll's (1967) study of 2782 American language students majoring in French, German, Italian, Russian, and Spanish at 203 universities in the USA. He found that students with a lower initial proficiency level made greater proficiency gains than students with higher initial proficiency and that students' spending time in the target-language-speaking country, compared to staying in the USA, was one of the best predictors of target-language proficiency gains. Further studies also found that SA was linked to proficiency gains: Brecht et al. (1993Brecht et al. ( , 1995 conducted a longitudinal study of 658 American students in Russia, where they investigated a series of groups over several years at a number of Russian-language institutions. However, in contrast with Carroll, they found that higher scores on pre-departure reading and grammar tests predicted greater proficiency gains in most skills, such as speaking, listening, and reading. This led them to suggest that there might be a proficiency threshold that learners must reach to benefit most from SA. It was also reported that individual differences, such as reading aptitude, predicted proficiency gains 2 . Barbara Freed, in her seminal 1995(Freed 1995 overview of the field, underlined the need for more investigation of issues such as individual differences and for research to be conducted with more methodological rigour. This led to the expansion of the field of SA and the adoption of a wide variety of theoretical and methodological approaches, such as a new emphasis on gathering more qualitative and ethnographic data about participants (Dewey et al. 2013;Kinginger 2008Kinginger , 2011Moratinos-Johnston et al. 2021).

SA and Language Proficiency
Segalowitz and  study of 40 native English learners of Spanish in a SA versus at home (AH) context for one semester found that the SA cohort made significant gains in global oral proficiency, as measured by an Oral Proficiency Interview (OPI), whereas the AH cohort did not. They also noted "the importance of the dynamic interactions that exist among oral, cognitive, and contextual variables" (Segalowitz and Freed 2004, p. 173), which may help to explain why there is so much individual variation in learning outcomes. SA has varying effects on different dimensions of language oral and written proficiency (Edmonds and Gudmestad 2018;Milton and Meara 1995). Pérez-Vidal and Juan-Garau (2011) found that SA led to improved use of formulas in oral performance and increased lexical complexity in writing. Moreover, the rate of gains varies across domains. Serrano et al.'s (2012) study of the oral proficiency of 24 Spanish students before, during, and after their SA in the UK found that significant gains were made in fluency and lexical diversity early on and that significant accuracy gains also emerged at the end of the year.
Fewer studies have looked at reading and written proficiency, and their results have been more variable (Pérez-Vidal and Barquin 2014; Pérez-Vidal and Juan-Garau 2009). Some studies did not find that SA had a positive effect on reading and writing. For example, DeKeyser (1990DeKeyser ( , 1991 found that the AH students in fact performed as well or better over the six-month period compared to the SA group on a grammar test, possibly because the onset level of proficiency in these students was below the "functional" level (DeKeyser 2014). Kinginger (2008) measured the reading, listening, and grammatical comprehension of 23 American students before and after spending a year abroad in France. She found that 13 participants did not make any significant progress, with seven of these students actually scoring lower on reading proficiency after the SA period. However, other studies do report improvements in reading and writing performance after SA. In a series of studies of 18 Irish Anglophone students of French, Howard (2005) found that French past tense usage in the SA group was significantly more accurate than the AH group, producing the correct form in more than 90% of cases, compared to 75-84% of cases in the AH group. Similarly, Juan-Garau et al. (2014) conducted a study of 57 trilingual (Spanish-Catalan-English) university students as part of the Study Abroad and Language Acquisition (SALA) Project and found that students' gains in general written proficiency were significantly greater over the SA period compared with the period at university. Overall, results do generally point to a positive effect of SA on written fluency and syntactic and lexical complexity. Mitchell et al.'s (2017) study of 56 British students' development in terms of general proficiency, fluency, accuracy, and complexity after completing their SA in Spanishspeaking countries (n = 27) and in France (n = 29) revealed striking differences between dimensions in oral and written speech (the LANGSNAP project). Lexical complexity and oral fluency increased fastest between Time 1 and Time 2, rising more slowly until Time 3. The authors were surprised to find very little improvement in accuracy after SA.

SA and Individual Differences
In addition to indications that SA affects various aspects of linguistic knowledge differently, it is clear that there is also immense individual variation in terms of language gains (Grey et al. 2015). It has been suggested that "study abroad intensifies individual differences in achievement" (Kinginger 2011, p. 58) because these differences affect the amount of input and interaction that a learner will have and will also affect the way in which input is processed (Sanz 2014). Personality plays a role, as highlighted in Arvidsson et al. (2018), as well as the arbitrary nature of the SA experience (Klapper and Rees 2012).
One crucial individual factor is students' proficiency level before they embark on SA. Lafford and Collentine (2006) suggest that initial proficiency may affect gains made abroad; having target language proficiency above a certain threshold may be necessary for measurable gains to be made. Intermediate learners, such as those in DeKeyser (1991) and Collentine (2004), may not yet have attained a level where they could move beyond basic communication. They may not yet have had a chance to automatize their existing declarative knowledge (DeKeyser 2014) and may thus have struggled in fast-flowing exchanges, especially if these happened in stressful circumstances with a lot of background noise, for example. Advanced learners have generally been found to make significant proficiency gains while abroad. Edmonds and Gudmestad (2018) looked at 20 advanced learners of French and assessed their use of correct gender marking over a 21-month period, which included a year abroad. They found a significant improvement between pre-stay and in-stay testing and concluded that the learners' high global and grammatical-gender proficiency pre-departure was key in allowing them to progress rapidly during their SA. Similarly, Leonard and Shea (2017) focused on 39 English L1 users learning Spanish in Argentina over a 3-month period. They found that those with higher levels of linguistic knowledge and faster processing speed in Spanish before departure made the greatest gains in accuracy and syntactic and lexical complexity during SA.
In contrast, Rees and Klapper (2007) reported of the 57 British foreign language students of German who spent their SA in Germany that those with higher initial proficiency level made fewer gains compared to those with lower initial proficiency levels. Similarly, Hessel's (2017) longitudinal, mixed-methods study of 143 German university students who spent a semester abroad as part of the ERASMUS programme revealed that initial L2 proficiency was the strongest predictor of proficiency gain. Those with lower initial proficiency made greater gains. Considering the contradictory results of Edmonds and Gudmestad (2018) on the one hand and Rees and Klapper (2007) and Hessel (2017) on the other hand, the question about which learners may benefit the most from SA remains to be investigated.
There are also still a number of areas in SA research that need further investigation: numerous studies have investigated English L2 learners, but the work on French L2 development remains relatively limited (Mitchell et al. 2017). Further study of L2 skills other than oral fluency is also needed (Grey et al. 2015). A number of both dependent and independent variables also deserve further study, such as the effect of SA on advanced L2 learners and the development of both actual and self-perceived proficiency in a number of domains. The present study will attempt to contribute to our understanding of the complex and dynamic development of these variables in a SA context.
The following research questions were formulated: (1) How did British and Irish students' actual and self-reported French proficiency change after their SA in France? In other words, what are the French L2 proficiency gains during SA through objective and subjective measures? (2) Is the change in self-reported French proficiency after a year in France linked to initial level of proficiency? In other words, what is the interaction or association between the objective and the subjective measures of gains? (3) Can the amount of change be linked to participants' reports on their SA experience?
In other words, can French L2 proficiency gains during SA be linked to reports on the SA experience?
(4) Does the relationship between actual and self-reported proficiency change between the start and end of the SA period? In other words, does the gap between objective and subjective measures narrow over time?

Methodology
In order to capture the complexity of SA outcomes, this study will adopt a longitudinal, mixed-methods approach, which Creswell (2015) defines as an approach where "the investigator gathers both quantitative (closed-ended) and qualitative (open-ended) data, integrates the two, and then draws interpretations based on the combined strengths of both sets of data to understand research problems" (Creswell 2015, p. 2). More specifically, it will be a convergent parallel design (cf. Creswell and Clark 2011, p. 70), in which quantitative and (some) qualitative data are collected in parallel. Therefore, in addition to the quantitative results, a sample of participant voices describing their SA experience will be included to allow a more nuanced interpretation of statistical findings. Dörnyei (2007) points out that "longitudinal research serves two primary purposes: to describe patterns of change, and to explain causal relationships" (Dörnyei 2007, p. 78). He expresses surprise at the scarcity of longitudinal studies in the field of Applied Linguistics, arguing that the use of mixed longitudinal designs is theoretically warranted, as they are "inherently concerned both with the micro and macro-levels of development and change (for example, individual growth and community change)" (Dörnyei 2007, p. 88). The method received ethical approval from the first author's Research Ethics Committee.

Design
Initial background information and information on participants' departure and return dates was collected well before departure. It determined the dates when participants were sent an invitation to participate in three further anonymous online questionnaires via email over the course of their SA. Participants completed the LexTALE test and the self-rated proficiency items before their departure (Time 1), they repeated the self-rated proficiency items when they were halfway through their stay in the Francophone country (Time 2), and they completed both the LexTALE and the self-rated proficiency items again after they had completed their stay in the Francophone country and had returned to their home country (Time 3) (see Table 1).

Participants
The participants were 33 British and Irish university students from nine different research-intensive universities, who were completing a compulsory year abroad as part of their language studies. To preserve their anonymity, students were asked to create a unique identifier combining their initials and their university and to provide their email address so that they could be contacted later. Most were female (26 females, 7 males); aged 20 (mean age = 20.09); and most had English as their first language (n = 31). The majority were studying French as a joint major, often with another language such as Spanish (n = 6), Italian (n = 4), German (n = 1), or with another subject such as Linguistics (n = 6), Business (n = 1), History (n = 1), Geography (n = 1), or Philosophy (n = 1).
Students were recruited via a personal email sent through the first author's Year Abroad office and through an open call on the mailing list of the Association for French Language Studies. Participants had been studying French for at least five years, and the mean length of study of French prior to departure was eleven and a half years. All spent between four and twelve months of their year abroad in a Francophone country (see Table 2). As a consequence, "Time 2" ranged from 2 months into SA to 6 months. There was a fairly even spread in the activities undertaken by participants: ten studied, thirteen completed internships, eight worked as teaching assistants, and two both studied and completed an internship. Most (n = 30) went to France-other destinations were Switzerland (n = 1), French Guyana (n = 1), and New Caledonia (n = 1).

Questionnaires
In order to have an actual proficiency measure, LexTALE tests were administered to participants before their departure and upon their return from their SA (Time 1 and Time 3). The LexTALE test for French was developed by Brysbaert (2013), who followed the procedure used in the original LexTALE test for English (Lemhöfer and Broersma 2012). LexTALE is a visual lexical decision task for intermediate and advanced language learners. It has been shown to be a good predictor of vocabulary knowledge and to give a good indication of general linguistic proficiency (Lemhöfer and Broersma 2012), when compared with the longer, more thorough, Quick Placement Test and the Test of English for International Communication (TOEIC). Lemhöfer and Broersma (2012) equate scores below 0.59 on LexTALE as corresponding to the lower independent users and lower-level descriptors of the Common European Framework (B1 and lower). LexTALE scores between 0.60 and 0.80 correspond to upper independent users (B2), scores between 0.80 and 0.90 correspond to lower advanced (C1) users, and scores above 0.90 correspond to upper advanced (C2) users. In the LexTALE French test, participants are presented with 56 French words of varying difficulty and 28 French-looking non-words, in a random order, and have to identify the real words (see Appendix A). The test has been found not to be at ceiling level even for L1 users (Brysbaert 2013) and so was deemed appropriate for use in this study. The same test was used both times, as this would allow direct comparison of scores. Given the significant time gap between the completion of each test (see Table 2), there was little risk of a training effect. Brysbaert's (2013) instrument is popular among researchers who work on French L2 (Wetzel et al. 2020). The initial mean LexTALE score was 0.675.
To strengthen the reliability of the actual proficiency measure and to obtain a more granular view of participants' skills in French, we also added a set of self-reported proficiency items. Participants were asked to assess their own proficiency in reading, writing, speaking, listening, grammar, vocabulary, and pronunciation, on a scale from 1 to 10 at Times 1, 2, and 3. Scores on the various dimensions were analysed individually and an average global score of self-reported proficiency was calculated. Internal consistency of the seven items was excellent (Cronbach alpha = 0.873 at Time 1, Cronbach alpha = 0.874 at Time 2, and Cronbach alpha = 0.886 at Time 3).
At Time 3, participants were also asked to write between 200 and 250 words in French about their SA experience. They were invited to reflect on whether it had lived up to their expectations, mention some highlights and lowlights, to report on things about Francophone culture that surprised them, and to describe how they felt about going back to full-time study at their home university. This yielded a corpus of 13,855 words.

Data Analysis
The quantitative data were imported into SPSS 26. As most variables were not normally distributed (Kolmogorov-Smirnov values ranging from 0.098 to 0.262, p < 0.05), non-parametric statistics were used, including Spearman rank correlation analyses and Friedman tests, rather than repeated measures ANOVAs. Since t-tests tolerate moderate violations of their normality assumption rather well (Rosenkrantz 2008), a paired samples t-test was used to assess the change between two time points. A Friedman test was used to assess change over three time points for self-rated proficiency. Spearman rank correlation analyses were used to investigate the relationship between initial proficiency in French and the extent of the difference in actual proficiency between Time 1 and Time 3.

Quantitative Results for Actual Proficiency
In order to answer the first part of the first research question, we ran a paired samples t-test. It revealed that the mean actual proficiency score for the whole group, as measured by the LexTALE test, increased significantly between Time 1 and Time 3: (t(32) = −5.181, p < 0.0001). Cohen's d was 0.80, which is indicative of a medium effect size (Plonsky and Oswald 2014) 3 (see Figure 1). In other words, students had moved, on average, from the lower range of upper independent users (B2) to the upper range of B2.
Languages 2021, 6, x FOR PEER REVIEW 8 of 18 (Plonsky and Oswald 2014) 4 . In other words, participants with the lowest initial scores made the most significant progress, as is shown in Figure 3.

Quantitative Results for Self-Reported Proficiency
To answer the second part of the first research question, we ran a Friedman test. It also revealed a significant increase in self-reported proficiency between Time 1, 2, and 3 (Chi 2 (32) = 17.9, p < 0.001). More detailed analyses (with paired t-tests) showed a nonsignificant increase between Time 1 and 2 (t = −1.53, p = 0.108, Cohen's d = 0.025) and a significant difference between Time 2 and 3 (t = 4.703, p < 0.001, Cohen's d = 0.431, which represents a small effect size). Figure 4 shows the increase in mean scores. In other words, it is only at the end of SA that students' scores started to increase significantly. An actual proficiency difference score was calculated for each individual by subtracting their LexTALE score at Time 1 from their score at Time 3. The measure thus reflects the amount of change over the SA period. The mean score was 0.085 (SD = 0.09), with a range from −0.04 to 0.38. The relatively high standard deviation suggests that there was a large degree of inter-individual variation. The changes and amount of variation are shown in Figure 2. LexTALE proficiency scores decreased marginally for two students (−0.01 and −0.04), two made no measurable progress, and the remaining 29 made varying amounts of progress.   Figure 2 shows that the biggest increases emerged among those at the lower end of the LexTALE scale. This was confirmed by a Spearman rank correlation analysis that revealed a significant negative relationship between initial LexTALE scores and the difference score (Rho = −0.521, p < 0.0001). This is indicative of a medium to large effect size (Plonsky and Oswald 2014) 4 . In other words, participants with the lowest initial scores made the most significant progress, as is shown in Figure 3.

Quantitative Results for Self-Reported Proficiency
To answer the second part of the first research question, we ran a Friedman test. It also revealed a significant increase in self-reported proficiency between Time 1, 2, and 3 (Chi 2 (32) = 17.9, p < 0.001). More detailed analyses (with paired t-tests) showed a nonsignificant increase between Time 1 and 2 (t = −1.53, p = 0.108, Cohen's d = 0.025) and a significant difference between Time 2 and 3 (t = 4.703, p < 0.001, Cohen's d = 0.431, which represents a small effect size). Figure 4 shows the increase in mean scores. In other words, it is only at the end of SA that students' scores started to increase significantly.
In order to obtain a more detailed understanding of which of the four skills and three linguistic domains of French increased most, we ran a series of Friedman tests (see Table 3). It shows a significant increase for all dependent variables from Time 1 to Time 3 except for pronunciation.   In order to obtain a more detailed understanding of which of the four skills and three linguistic domains of French increased most, we ran a series of Friedman tests (see Table  3). It shows a significant increase for all dependent variables from Time 1 to Time 3 except for pronunciation.   Zooming in on the trends between the start and the end of the SA period, we used paired t-tests to look for differences between Time 1 and Time 2 and again between Time 2 and Time 3. It shows that only oral skills increased significantly both between Time 1 and Time 2 and again between Time 2 and Time 3 (see Table 4 and Figure 5). The other skills and domains did not change significantly between Time and Time 2 but did so between Time 2 and Time 3 (with the exception of pronunciation) (see Table 4 and Figure 6). This suggests that the development is non-linear, with a more pronounced change in the second part of the SA period. The effect sizes can be described as small for speaking between Time and Time 2 and again between Time 2 and Time 3. The effect size of the changes for writing and grammar is also small between Time 2 and Time 3 (Plonsky and Oswald 2014).

Qualitative Analysis
Investigation of the qualitative data, in the form of reports from individual students after their SA, revealed unique patterns and possible reasons to explain why some students made significant progress while others did not. A varied sample of participant reports is presented here. These particular participants' reports were selected to represent the range of different proficiency levels and a range of changes in proficiency scores. The initial reports were made in French but have been translated into English here for ease of reading.
One student who made significant progress, despite having a high initial LexTALE score, was participant 26. She was a French, Spanish, and Portuguese student at the University of Cambridge and spent eight months studying in Paris. Her very high initial LexTALE score of 0.839 (C1 level) increased to a score of 0.981 (C2) (out of a possible 1.0), the highest score of the group. She reports that "My year abroad was one of the best years of my life" and that she is reluctant to return home and leave behind "all of the interesting people I met in Paris". She clearly integrated very well and made many French friends, which helped her still to make significant improvements to her French proficiency despite her very high initial score.
Another student, participant 10, who had a lower initial LexTALE score, equally made significant progress. She was a French and Philosophy student at Oxford University, had an initial score of 0.616 (the lower end of B2 level), but finished with a score of 0.741 (the higher end of B2 level). She worked as a teaching assistant in French Guyana for eight months. In her discussion of her SA, she mentions that she had previously found her French studies at university to be "painfully discouraging". However, on her SA, she made many Francophone friends and found "a new, positive attitude". While these students improved significantly and reported positive experiences, it was possible to improve without having a positive experience. In order to answer the final research question on the relationship between self-reported proficiency scores and actual proficiency scores, we ran a Spearman rho correlation analysis. A highly significant relationship emerged between participants' self-reported proficiency scores and LexTALE scores at Time 1 (Rho (32) = 0.43, p < 0.0001) and at Time 3 (Rho (32) = 0.56, p < 0.0001), which suggests good reliability of both measures.

Qualitative Analysis
Investigation of the qualitative data, in the form of reports from individual students after their SA, revealed unique patterns and possible reasons to explain why some students made significant progress while others did not. A varied sample of participant reports is presented here. These particular participants' reports were selected to represent the range of different proficiency levels and a range of changes in proficiency scores. The initial reports were made in French but have been translated into English here for ease of reading.
One student who made significant progress, despite having a high initial LexTALE score, was participant 26. She was a French, Spanish, and Portuguese student at the University of Cambridge and spent eight months studying in Paris. Her very high initial LexTALE score of 0.839 (C1 level) increased to a score of 0.981 (C2) (out of a possible 1.0), the highest score of the group. She reports that "My year abroad was one of the best years of my life" and that she is reluctant to return home and leave behind "all of the interesting people I met in Paris". She clearly integrated very well and made many French friends, which helped her still to make significant improvements to her French proficiency despite her very high initial score.
Another student, participant 10, who had a lower initial LexTALE score, equally made significant progress. She was a French and Philosophy student at Oxford University, had an initial score of 0.616 (the lower end of B2 level), but finished with a score of 0.741 (the higher end of B2 level). She worked as a teaching assistant in French Guyana for eight months. In her discussion of her SA, she mentions that she had previously found her French studies at university to be "painfully discouraging". However, on her SA, she made many Francophone friends and found "a new, positive attitude". While these students improved significantly and reported positive experiences, it was possible to improve without having a positive experience.
For example, participant 25, who studied French at Queen's University Belfast and spent ten months at the University of Poitiers, had a LexTALE score that increased from 0.527 (B1 level) to 0.902 (C2 level), despite describing her experience in France as being generally difficult and lonely. It thus seems that a positive experience is not essential for linguistic progress.
Participant 25: My year abroad did not go as expected. It was difficult and I often felt lonely. There were a few fun moments with friends, or when I went to visit some tourist sites . . . Altogether my year abroad wasn't great but I think I did improve my French.
However, not all students made significant progress: participant 31, who studied International Relations with French at the University of Portsmouth and spent 10 months studying in Lyon, made no measurable progress. His initial LexTALE score was fairly low at 0.527 (B1 level), and he had a slightly lower score of 0.518 after the SA period. He reports that "there was not much to do at the university", which suggests that he did not socialise much and may not have used French frequently enough to make progress in proficiency.
Equally, not all participants who had made little or no progress had low initial proficiency scores, as is demonstrated by participant 14. Her first language is Romanian and she studied Law and French at the University of Oxford. She spent 10 months at the Panthéon-Assas University, also referred to as the "Sorbonne Law School". Although she obtained a slightly lower score at Time 3 (C1 level) than at Time 1 (borderline C2 level), her LexTALE score is still one of the highest (Time 1 score: 0.892; Time 2 score: 0.848). She had lived in France for seven years before moving to the UK and has an advanced proficiency level in French. It would thus be wrong to interpret the lower score as a sign of "deterioration" in French proficiency. Her feelings about her SA are ambiguous: Participant 14: I really enjoyed my year abroad. I missed my Oxford friends a lot, but I was able to visit them a lot, and I made a lot of new friends. One of the things I found the hardest was having to revise for exams twice a year. I can't wait to go back to Oxford as there are a few things I've been missing, but equally I don't really want to have my exams next year.

Discussion
The first research question asked how students' actual and self-reported proficiency changed after their SA. Both increased significantly. Most students were at the lower end of upper independent users (B2) in French before departing and progressed, on average, to the higher end of upper independent users (B2) in French. The highly significant increase in target language proficiency is similar to that reported by Hessel (2017) for German Erasmus students in the UK. Hessel found that her participants' English proficiency increased significantly during their first three months in the UK, a small effect size (d = 0.57) (Hessel 2017, p. 43). In fact, the effect size in the present study is larger (d = 0.80), possibly because the duration of the SA period was longer.
The global self-reported proficiency scores from Time 1, Time 2, and Time 3 demonstrated similar trends to the actual proficiency scores, with the effect size being larger between Time 2 and Time 3 than between Time 1 and Time 2. This was confirmed by looking at individual items for the four skills and the three domains. Speaking was the only skill to reach the threshold for a small effect size between Time 1 and Time 2 and again between Time 2 and Time 3. The increase for self-reported scores for writing and grammar reached a small effect size between Time 2 and Time 3. This reflects the findings by Mitchell et al. (2017) that progress is not always linear and that different skills and domains may develop at different speeds. The present investigation thus adds to the growing evidence that SA boosts linguistic proficiency in areas other than oral proficiency (Rees and Klapper 2007;Juan-Garau et al. 2014;Mitchell et al. 2017;Hessel 2017;Edmonds and Gudmestad 2018).
There was considerable variation between students in terms of how much progress they made, with some not making any measurable improvement, while others made considerable gains. This is again consistent with the literature, where the amount of variation between individuals has long been identified as one of the main features of SA and for which no complete explanation has yet been found despite a long list of variables that explain some variance (Briggs Baffoe-Djan and Zhou 2021;DeKeyser 2014;Kinginger 2008;Mitchell et al. 2017).
The second research question dealt with the effect of initial proficiency on linguistic gains at the end of the SA period. A highly significant inverse relationship between initial proficiency and actual proficiency difference emerged in this study, where those with lower initial proficiency improved more. This deviates from the findings in Leonard and Shea (2017) but corroborates findings from other studies, such as Hessel (2017), where initial L2 proficiency was found to be the strongest predictor of L2 proficiency gain after the SA period. Edmonds and Gudmestad (2018) reported similar results and concluded that students' high global proficiency before departure was key to their progress. It must be noted that these results do not contradict the idea of a proficiency threshold above which learners are found to make the most progress (Lafford and Collentine 2006). As mentioned above, the students in this study are all "independent" to "advanced" users of French, who are above the proficiency threshold level (DeKeyser 2014).
The third research question focused on the potential of the qualitative data to shed light on potential causes of change in proficiency. It showed that for the majority of participants, SA had been a relatively mixed experience, which often defied previous expectations and where there were significant emotional highs and lows. A common refrain in participants' comments was that the year abroad did not go as planned (cf. Klapper and Rees 2012). For example, many participants reported that their universities had often been closed as a result of strikes, meaning that they therefore had far less exposure to French and French people than they had expected. They may have had the best of intentions, and gone with a very positive attitude, but they were denied the opportunity to build French social networks.
An interesting finding is also that the reported enjoyment, or the lack of it, did not always correlate with how much progress students made. Some students reported having a very positive experience and made significant progress, while others who had a positive experience made little progress. Correspondingly, some students who reported having less positive experiences made significant progress, and some who had less positive experiences made little progress. It would thus appear that other factors have more of an impact on proficiency.
While there was a confirmation of the general pattern that larger proficiency gains corresponded to a more positive SA experience and that there was a link between those with more limited linguistic gains and a negative experience, there were also a number of participants who deviated from this pattern. Participants whose actual proficiency had decreased between Time 1 and Time 3 had not necessarily regressed linguistically but had scores that were close to the ceiling level. Moreover, while those with lower initial proficiency did make the biggest proficiency gains, some participants with a high initial proficiency had also made significant gains by Time 3.
The picture that emerges is that besides the common trend-a general gain in actual and self-perceived proficiency-every participant had had a unique experience and mixed feelings. Some participants enjoyed many aspects of their SA but not necessarily all of them, while others did not enjoy the experience overall but did appreciate certain aspects, such as tourism (which is reminiscent of certain participants in Kinginger's (2008) study). Some participants made many French friends, even boyfriends, while others were isolated or only had Anglophone friends; some did not want to leave France after SA, while others could not wait to go back and re-integrate into campus life at their home university. These findings confirm Kinginger's (2008Kinginger's ( , 2011Devlin's (2013); Mougeon and Rehner's (2015); Mitchell et al.'s (2017), and Briggs Baffoe-Djan and Zhou's (2021) observations that engagement with the host environment-and both quantity and quality of language contact-are key to linguistic progress. In other words, the individual's role is central.
The qualitative data showed that some participants found a way to develop new social networks with Francophones or to establish close relationships with Francophones, while others floundered and fell back on their existing Anglophone networks. It is likely that some were better equipped psychologically to overcome the adversities that they were facing. If experience affects proficiency outcomes, this variability of experience between participants helps to explain why there is so much variation in outcomes between participants. It does confirm Howard's (2021) observation that learners need "to be active participants in contributing to shaping their language contact opportunities and general experiences" (Howard 2021, p. 7). The relationship between actual and self-reported proficiency at the start and end of SA was the focus of the fourth and final research question and was an opportunity to test the reliability of both measures. The correlation value between both measures was highly significant both times but it was even higher at Time 3 than at Time 1. This could be attributed to a more accurate judgment of French proficiency. Before their departure, participants would typically have compared their performance with that of their peers and teachers. After the year abroad, they had accumulated multiple daily opportunities to compare their performance with that of L1 users and foreign language users around them, allowing them to judge their performance more accurately.
The current study is not without limitations. Only 33 out of the initial 54 participants provided data three times. Attrition is a well-known problem in longitudinal studies (Dörnyei 2007) and it is particularly hard to obtain data during SA when participants have a lot on their mind. As a consequence, there is a risk that those who failed to complete the second and third survey may have had a slightly different profile and SA experience from those who did complete the three surveys. In other words, our 33 participants may have a unique profile and may therefore not be representative of the whole cohort.

Conclusions
The present study sought to measure the effect of SA in Francophone countries on both actual and self-reported French proficiency of British and Irish students. The results show that both actual proficiency and self-reported proficiency increased significantly after SA.
Because data for self-reported proficiency had been collected three times, it emerged that the increase was more linear for speaking and listening than it was for reading and writing, with the significant increase occurring after the SA mid-way point. The same pattern was found for grammar and vocabulary. The only domain where no significant change occurred over the SA period was pronunciation. Linking change in actual proficiency between Time 1 and Time 3 to participants' descriptions of their SA revealed a clear disconnect between progress and the overall positivity of the experience. Some students who had struggled actually made good progress. The establishment of a local French social network emerged as an important factor and random events such as university strikes were linked to limited progress.
Participants with lower actual proficiency at Time 1, namely lower and upper independent users (B1 and B2), were found to have made the most significant progress during SA.
To conclude, SA provides students with opportunities to boost their linguistic skills in parallel with a strengthening of their resilience in the face of loneliness and occasional adversity. It seems that the decision to actively take control and shape the SA experience was more likely to lead to linguistic progress than the mere enjoyment of the experience. Institutional Review Board Statement: The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Social Sciences and Humanities Inter-divisional Research Ethics Committee of the University of Oxford (R52027/RE001) and date of approval.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.