2.1. Sources of Foreign Language Enjoyment (FLE) and Foreign Language Classroom Anxiety (FLCA)

A study by Dewaele and MacIntyre (2014) was among the first to defend a more holistic perspective on FL classroom emotions. It introduced the concept of FLE, defined as a complex positively valenced emotion resulting from a combination of challenge and perceived ability that allows the tackling of difficult tasks, to complement a measure of FLCA, a complex negatively valenced emotion defined as “a distinct complex of self-perceptions, beliefs, feelings, and behaviors related to classroom language learning arising from the uniqueness of the language learning process” (Horwitz et al., 1986, p. 128). The main research question in Dewaele and MacIntyre (2014) was whether FLE and FLCA were opposite poles of the same dimension or whether they were in fact independent dimensions. The second research question was identifying sources of variation in FLE and FLCA linked to sociobiographical and contextual variables. The authors combined a 21-item FLE scale and a shortened, 8-item version of Horwitz et al.’s (1986) original 33-item FLCA scale in an online questionnaire with closed questions and one open question about enjoyable episodes in the FL class. A total of 1742 FL learners from all over the world contributed. A significant but moderate negative correlation between FLE and FLCA suggested that FLE and FLCA are separate dimensions and that it is possible to experience both simultaneously. This was confirmed by Dewaele and MacIntyre’s (2014) analysis of the qualitative feedback of participants. Levels of FLE were found to increase significantly with the overall number of languages known, perceived proficiency, relative standing in the FL class, age, gender and education level, while FLCA showed an inverse pattern. After running a Principal Component Analysis on the original corpus that revealed separate dimensions for Personal and Social Enjoyment, Dewaele and MacIntyre (2016) concluded that it is not just FLE that powers progress but that FLCA has a role to play in FL acquisition. In other words, FLE and FLCA are like the right and left feet of learners and a balance needs to be found between both. It implies that “the goal is not to eliminate FLCA any more than a runner would wish to eliminate one of her feet (even the aching, sore one)” (Dewaele & MacIntyre, 2016, pp. 233–234).

A number of studies have investigated the relationships between FLE and FLCA, and the sources of variation for both emotions in single FL contexts, with a focus on the role of the teacher. The first study to do so was Dewaele et al. (2018), who analysed data on FLE and FLCA from 189 pupils (aged 12 to 18) in London schools who had English as a first language (L1) and were studying French or German as a FL. FLE was significantly linked to positive attitudes toward the FL, positive attitudes toward the teacher, frequent use of the FL by the teacher, more time spent on speaking during classes, a higher relative standing in the peer group and being more advanced in the FL. In contrast, FLCA was linked to negative attitudes towards the FL, lower relative standing in the peer group and being less advanced in the FL but was not linked in any way to the teacher or teacher behaviour. Dewaele and Dewaele (2020) used a subsample of this corpus to investigate to what extent learners’ FLE and FLCA vary at a single point in time when facing two different teachers for the same FL. Participants were 40 pupils who had one main teacher and a second teacher for the same FL. FLCA was found to be constant with both teachers, but pupils reported significantly higher FLE with the main teacher. It thus seems that FLCA is more stable while FLE is more teacher-specific. Further studies confirmed the important role of the teacher on FLE among 592 learners of Turkish as a FL in Kazakhstan (Dewaele, Özdemir et al., 2019), 564 Chinese English as foreign language (EFL) learners in China (Jiang & Dewaele, 2019) and 750 FL learners from around the world (Dewaele & MacIntyre, 2019).

2.2. The role of the specificity of the FL and the meso- and macro-context on FLE, FLCA and attitudes/motivation

Learners’ classroom emotions and language attitudes are influenced by contextual factors within and beyond the classroom. De Smet et al. (2018) collected data from 896 primary- and secondary-level pupils in French-speaking Belgium in traditional schools and in schools that used Content and language integrated learning (CLIL) where part of the curriculum was taught in Dutch or English. The researchers found significantly higher levels of FLE and lower levels of FLCA for English than for Dutch, a language associated with the Flemish community with whom Belgian Francophones have a tense relationship and a language perceived as less useful than English despite the fact that Dutch is often required for jobs in Belgium. A follow-up study by the same authors showed similar patterns for language attitudes and motivation, with significantly more positive attitudes towards the target languages in the CLIL schools compared to the non-CLIL schools and higher perceived attractiveness and easiness of English compared to Dutch at the start of secondary education (De Smet et al., 2019).

2.3. The role of emotions and motivation on FL performance

Dewaele and Alfawzan (2018) compared the link between classroom emotions and FL performance among 189 FL learners in the United Kingdom (UK) and 152 EFL learners and users in Saudi Arabia. They found that overall FLE had a stronger positive effect on FL performance than the negative effect of FLCA. Sources of FLCA in the Saudi sample included bad pedagogical practices and humiliating comments by teachers – sometimes even physical abuse.

Further confirmation of the role of positive emotions on L2 performance emerged in Jin and Zhang (2018), who collected data from 320 Chinese EFL high school pupils. The authors found that enjoyment of English learning had a direct positive effect on pupils’ English achievement while enjoyment of teacher and student support only had an indirect effect. The authors argued that a range of internal and external variables influenced learners’ FLE and that learners’ authentic liking of FL learning was an important source of FLE.

Li et al. (2020) considered the relationship between FLE, FLCA and proficiency among 1718 Chinese EFL high school students. They found that FLCA had a stronger negative effect on self-perceived EFL proficiency (explaining twice as much variance) than the significant positive effect of FLE. The relationship between FLE, FLCA and proficiency was found to be different among students with low, medium and high levels of proficiency: FLE had a stronger effect than FLCA in the low proficiency group, but FLCA had a stronger effect in the medium and high proficiency groups. Interviews with 64 students indicated that poor results in English tests and fear of criticism by the teachers for weak performance were the most frequent sources for FLCA. Good test results, higher relative standing in the group, and teacher praise were frequently mentioned sources for FLE.

Li (2020) considered the role of Trait Emotional Intelligence (defined as a lower order personality trait reflecting a constellation of emotion-related self-perceptions and dispositions) on FLE and on both self-perceived and actual English proficiency of 1307 Chinese high school pupils. A mediation analysis revealed that Trait Emotional Intelligence and FLE were positively linked to both self-perceived and actual English proficiency.

FLE and FLCA were combined with motivation in Saito et al. (2018) and the three variables were found to have a significant effect on comprehensibility of 108 Japanese high school EFL pupils. Adopting a longitudinal design, pupils’ progress over three months was firstly linked with high FLE and low FLCA and secondly with motivation. Higher levels of FLE and a stronger Ideal L2 Self predicted the amount of English practice, which had a positive impact on comprehensibility in English.

Using a comparable cross-sectional design and focusing on students’ second FL, Zhang et al. (2020) found that their 335 Chinese university students’ instrumental and integrative motivation positively influenced their second FL self-rated proficiency (mostly Japanese, French and German). Further analysis revealed that the relationship between motivation and self-rated FL proficiency was mediated by FLE. The authors did not compare the second FL with English, the first FL.

Botes, Dewaele, et al. (2020b) reversed the perspective on the relationship between classroom emotions and self-perceived FL proficiency, by looking at how proficiency influences FLE and FLCA among a sample of 1622 FL learners extracted from the database created by Dewaele and MacIntyre (2014). A higher level of perceived FL proficiency and knowledge of more languages were linked to a significantly stronger FLE and lower FLCA. The authors pointed out that the causal pathway could be multidirectional, where proficiency and multilingualism on the one hand, and FLE and FLCA on the other hand, could be both a cause and an effect.

In an attempt to gain a broader view on the average effects of FLCA on FL achievement, Botes, Greiff and Dewaele (2020) carried out a meta-analysis on 67 studies. A moderate negative correlation was found between FLCA and all categories of academic achievement (r = –.39). Several studies have also reported a moderately positive correlation between FLE and academic FL achievement (Dewaele & Alfawzan, 2018; Li, 2020).

2.4. Enjoyment and classroom anxiety in L1 and FL classes

Resnik and Dewaele (2020) is the first study to have investigated enjoyment and classroom anxiety not just in the same learners’ FL but also in their L1. The L1 German participants (N = 768) were secondary- and tertiary-level students studying English as an FL. Significantly higher levels of anxiety and enjoyment were found in the English classes than in the German classes (small effect size). Additionally, enjoyment and anxiety were negatively correlated in both the L1 and FL contexts. Levels of enjoyment and anxiety in the L1 were positively linked with FLE and FLCA (small to medium effect sizes). Participants’ feedback suggested that the increased FLE and FLCA might be linked to a more engaging didactic approach in English classes. The authors speculated that the FL class might also be perceived as being both more exciting and challenging than the L1 class.

Dewaele and Proietti Ergun (to appear) investigated the link between enjoyment, attitudes/motivation and anxiety in Turkish L1 and in Italian FL of 110 pupils in an immersion school in Istanbul. Statistical analyses revealed similar levels of levels of attitudes/motivation in Turkish and Italian but anxiety levels were significantly higher in Italian. Levels of enjoyment were marginally higher in Italian than in Turkish. A positive relationship was found between levels of enjoyment in both languages. Anxiety was negatively correlated with course marks in both languages.

2.5. Attitudes, motivation and classroom emotions

Gardner (1985) developed the socio-educational model to describe the role of motivation in SLA. Although newer models have been introduced such as the L2 Self model by Dörnyei (2009), Gardner’s model is “still relevant” (Dörnyei, 2019, p. xxi). The socio-educational model is particularly useful for investigating attitudes and motivation while taking into account the influence of the meso-context (classroom, school) and the macro-context (society). It consists of four broad dimensions, namely integrativeness, attitudes towards the learning situation, language anxiety and language attitudes and motivation, as well as 12 sub-dimensions. Reflecting back on the development of the model, Gardner (2019) pointed out that he wanted “to clarify the underlying process linking affective variables to language achievement” (p. 6). The aim of the Attitudes/Motivation Test Battery (AMTB) is to gather information on “the student’s affective reaction to the classroom environment, the cultural influences on the learner’s reaction to acquiring attributes of the cultural community, anxiety reactions when called upon to use the language, and the effort, persistence and satisfaction associated with the process” (p. 11). The sub-dimensions include interest in FL, parental encouragement, motivational intensity, FL class anxiety, evaluation of teacher, attitude toward learning the FL, desire to learn the FL, attitudes toward speakers of the FL, integrative orientation, instrumental orientation, course evaluation and FL use anxiety.

MacIntyre et al. (2019) investigated the relationship between 11 dimensions of the AMTB and 20 learner emotions (10 positive and 10 negative ones). The study was based on two learner samples (157 Chinese EFL learners studying in China and 750 learners from around the world). Despite significant differences in AMTB scale means and means for individual emotions between the two samples, consistent correlations emerged between AMTB scales and individual emotions, with stronger positive correlations between AMTB scales and positive emotions compared to the link between AMTB scales and negative emotions. The authors concluded that “learners’ emotions are correlated with attitudes toward the FL learning situation, integrativeness and motivation” (p. 76).

2.6. Summary

What this literature review has revealed is that recent research on classroom emotions has unearthed a wide variety of learner-internal and learner-external variables that affect FLE and FLCA in unique ways depending on learners’ age, proficiency and context. Moreover, they interact in complex and dynamic ways and can have both direct and indirect effects on FL performance and learner motivation. FLE and FLCA are unique and independent dimensions that can sometimes work in tandem and affect performance in the FL (Dewaele & Li, 2020). The most obvious gap is the scarcity of research on the relationships between classroom emotions, attitudes/motivation and performance in the same learners’ multiple FLs. Such research is crucial because we need a clearer understanding of the role of fleeting emotions combined with longer-term psychosocial constructs on learners’ progress and FL performance in multiple FLs. It can allow us to establish whether sources of individual differences that had previously been identified for a learner’s single FL (typically English), and which were implicitly assumed to be universal, do in fact apply to all other FLs of that learner.

4.1. Context and participants

Respondents were 110 pupils enrolled at the selective fee-paying Liceo Scientifico Italiano, Istanbul, Turkey (www.liceoitaliano.net). Pupils have 40 hours of teaching per week including Turkish (L1) classes, Italian (L2) classes and English (L3) classes. During the first two years at the school, pupils follow the Italian curriculum for math, science, philosophy and Latin and the Turkish curriculum for other classes. For grade III and IV pupils choose either the double program (which is prestigious but requires high course marks in Italian) or the Turkish national curriculum. Staff use their L1, meaning Italian teachers use Italian, Turkish teachers use Turkish and American teachers use English. The administrative staff uses Turkish. An overview of the amount of teaching hours per week for each language is included in the Appendix. Multilingualism is highly appreciated in Turkey and there is social pressure for children to acquire FLs, English in particular (Akalın & Zengin, 2007).

Participants’ age ranged from 14 to 18, with a mean age of 15.6 (SD = 0.98). Female pupils (n = 65) outnumbered male pupils (n = 45). This reflects the typical gender ratio of pupils enrolled in language classes in Turkey and it is a common pattern in web-based research on emotion in SLA (Dewaele, 2018). All participants had Turkish nationality and Turkish as an L1, with Italian and English as FLs. Overall, the sample consisted of 76 trilinguals, 31 quadrilinguals, and 3 pentalinguals.¹ Many pupils also knew Latin. A comparison of self-perceived proficiency in Italian and English measured on a 5-point Likert scale (ranging from minimal to maximal) revealed that participants felt significantly less proficient in Italian than in English (Italian: M = 2.88, SD = 1.14, English M = 3.76, SD = 1.04, t(104) = –8.43, p < 0.0001). Course marks for both FLs were calculated on the basis of three to four written tests and a score on performance, which include one project per semester, participation in class activities and homework.

4.2. Procedure

Ethics approval was obtained from the first author’s institution. Data were gathered from April to September 2019. The sampling strategy used in this study can be described as convenience sampling (Dörnyei, 2007). The headmaster gave his consent after which parents were contacted about the study on emotional variables in the FL classes and were given the choice to opt out (none did). Parents then signed a written consent form. Two teachers helped the second author, who had a child in the school, with the data collection. Pupils were asked for their individual consent at the beginning of the online survey. Participants wrote their names down on the questionnaire in order to link the data with language course marks. No names were included in the present study. Questionnaires that were not entirely completed were discarded (n = 25).

4.3. Instruments

The original questionnaire in English was translated into Turkish by a professional translator who had Turkish as an L1, and the translation was double-checked by two English-Turkish multilinguals. The first set of questions focused on participants’ socio-biographical background and their language-learning history, such as the languages known, age and gender. This was followed by items on pupils’ FLE, FLCA, attitudes and motivation in Italian and English classes.

4.4. Data analysis

Although one-sample Kolmogorov-Smirnov tests revealed that FLE, FLCA, attitudes/motivation and course marks were not normally distributed (KS values ranging between .07 and .16, p < .001), the calculation of Q-Q plots suggests that they follow a normal distribution reasonably well except for the extreme tails (see Figure 1). We thus opted for the more powerful parametric statistics. This was particularly important as it allowed us to use multiple regression analyses in order to identify the predictors of course marks. The statistical analyses were carried out using SPSS version 23. These included t-tests, Pearson correlation analyses and follow-up multiple regression analyses.

5.1. Differences in FLE, FLCA and attitudes/motivation values in Italian and English classes

A series of paired t-tests with Bonferroni correction (p < 0.0125) revealed that FLCA was significantly higher in Italian than in English (medium effect size, cf. Plonsky & Oswald, 2014)³ and that course marks were significantly lower in Italian (large effect size, cf. Plonsky & Oswald, 2014). No significant differences existed between both languages for FLE and attitudes/motivation (see Table 2).

The next step in the investigation was to learn whether FLE, FLCA and attitudes/motivation values were correlated across Italian and English. Pearson correlation analyses with Bonferroni correction (p < 0.016) suggest that this is the case for FLE (r = 0.356, p < 0.0001) and attitudes/motivation (r = 0.355, p < 0.0001) but not for FLCA (r = 0.084, p = 0.38). High levels of FLE in Italian were thus linked with higher levels of FLE in English with a similar relationship for attitudes/motivation (medium effect size, cf. Plonsky & Oswald, 2014).⁴

5.2. The relationship between FLE, FLCA and attitudes/motivation within Italian and English

Next we investigated the relationship between FLE, FLCA and attitudes/motivation within the Italian and English classes using Pearson correlation analyses. FLE and attitudes/motivation were strongly positively correlated in Italian (r = 0.526, p < 0.0001), which represents a large effect size. A negative relationship emerged between FLCA and FLE (r = –0.202, p < 0.034) and between FLCA and attitudes/motivation in Italian (r = –0.386, p < 0.0001) (small to medium effect sizes).

Similar patterns emerged for English: FLE and attitudes/motivation were strongly positively correlated (r = 0.612, p < 0.0001), which represents a large effect size. A negative relationship emerged between FLCA and FLE (r = –0.321, p < 0.0001) and between FLCA and attitudes/motivation in Italian (r = –0.453, p < 0.0001) (medium effect sizes).

5.3. The relationship between FLE, FLCA, attitudes/motivation and course marks in Italian and English

An initial series of Pearson correlation analyses revealed that course marks in Italian were non-significantly positively linked with FLE (r = 0.173, p < 0.073) and significantly positively linked with attitudes/motivation (r = 0.386, p < 0.0001). They were significantly negatively correlated with FLCA (r = –0.359, p < 0.0001).

A second series of Pearson correlation analyses revealed that course marks in English were significantly positively linked with FLE (r = 0.307, p < 0.001) and attitudes/motivation (r = 0.340, p < 0.0001). They were significantly negatively correlated with FLCA (r = –0.516, p < 0.0001).

The first of two stepwise multiple linear regression models was conducted in order to investigate the unique effect of the two variables that were significantly linked with course marks in Italian while controlling for the effect of the other. The Durbin-Watson value (1.74), the tolerance (0.80) and the VIF value (1.26) indicated no concern for autocorrelation nor multicollinearity; the normality and residual plots indicated linearity and homoscedasticity. A significant regression equation was found, with the two variables predicting 19% of the variance: adjusted R² = 0.177, F(2, 106) = 12.58, p < 0.0001. This represents a small effect size (Plonsky & Ghanbar, 2018). The strongest positive predictor was attitudes/motivation, explaining 14.9% of the variance (β = .28, p < 0.005) (see Figure 2). FLCA was a significant negative predictor, explaining an additional 4.2% of variance (β = –0.23, p < 0.020) (see Figure 3).

A second stepwise linear regression model was conducted on course marks in English entering FLE, attitudes/motivation and FLCA. The Durbin-Watson value (1.97), the tolerance (0.87) and the VIF value (1.15) indicated no concern for autocorrelation nor multicollinearity; the normality and residual plots indicated linearity and homoscedasticity. A significant regression equation was found, with attitudes/motivation and FLCA predicting 29.3% of the variance: Adjusted R² = 0.280, F(2, 106) = 21.96, p < 0.0001. This represents a small-to-medium effect size. The strongest (negative) predictor was FLCA (β = -0.45, p < 0.0001), explaining 26.7% of variance (see Figure 4). Attitudes/motivation was a weaker positive predictor (β = 0.175, p = 0.049), explaining a further 2.6% of variance (see Figure 5). FLE was excluded from the model.