Perceived L2 oral fluency, working memory, foreign language enjoyment and foreign language anxiety: identifying distinct patterns of relationship

Abstract

L2 oral fluency is a core aspect of general oral proficiency. However, it represents a distinct challenge for L2 speakers as it requires a real-time efficient allocation of cognitive resources. Specifically, working memory (WM) plays a crucial role in speech production. These cognitive resources are limited and vary from one speaker to another. Additionally, they have long been shown to interact with foreign language anxiety and foreign language enjoyment, which interact with various aspects of L2 oral production. To our knowledge, no previous studies have investigated the relationship that might exist between perceived L2 oral fluency; i.e. listeners’ judgments about L2 speakers’ fluency, WM, foreign language anxiety and foreign language enjoyment. To fill this gap, 78 ESL French-speaking adults were subjected to a picture-based narration task. A flowchart scheme was used to measure perceived L2 oral fluency. WM was measured using a numerical span test, and foreign language anxiety and foreign language enjoyment were measured through a questionnaire. Results show a distinct pattern of interaction between the variables. More specifically, foreign language anxiety significantly predicted perceived L2 oral fluency among participants with a weaker WM, whereas foreign language enjoyment significantly predicted L2 oral fluency among participants with a stronger WM capacity.

1. Introduction

Oral fluency, considered a core characteristic of second language (L2) oral production (de Jong and Perfetti 2011; Préfontaine et al. 2016; Tavakoli et al. 2016), is essential during communicative acts, as it helps maintain interlocutors’ attention (Lennon 2000; Suzuki and Kormos 2020). Perceived L2 oral fluency, that is, listeners’ judgments about L2 speakers’ fluency (Segalowitz 2010), is particularly relevant in that regard. However, oral fluency represents a distinct challenge for L2 speakers whose cognitive resources, specifically working memory (WM), are crucial for L2 oral production (Wen and Li 2019), as they are often overstretched by developing L2 oral skills and limited linguistic knowledge. This challenge can be exacerbated by L2 speakers’ limited available cognitive resources, known to vary among speakers (Kormos 2006). It should be noted that these cognitive resources have long been shown to interact with emotions (Damasio 1994, 2011; Phelps 2005) and, especially, foreign language anxiety and foreign language enjoyment (Dewaele and MacIntyre 2014, 2016). Research results show they both interact with various aspects of L2 oral production. More recently, a growing interest in how emotions moderate such links between cognition and oral production has appeared in research (e.g., Zuniga and Simard 2022). However, to our knowledge, no previous studies have investigated the relationship that might exist between perceived L2 oral fluency, WM, foreign language anxiety and foreign language enjoyment. This is what we set out to examine in the present study.

2. Literature review

2.1. Oral fluency in L2

Since Fillmore’s (1979) proposed conceptualization of fluency, according to which fluent speakers talk at length, with only a few pauses, in semantically dense messages, appropriately in various contexts, and in a creative manner (p. 93), a plethora of research has been conducted to investigate it in L2 (Peltonen 2024). Unfortunately, despite the important number of studies conducted, there is still, to this day, no consensus on its meaning (Huhta et al. 2019; Suzuki and Kormos 2020).

To better understand how fluency is examined in L2, it is possible to group the definitions found in the literature into two categories (Kormos and Dénes 2004). On the one hand, some definitions highlight temporal phenomena, such as pauses and repairs (e.g., Skehan et al. 2016), closely related to what Lennon (1990, 2000) named narrow definitions of fluency. An example of such a narrow definition of fluency is the “rapid, smooth, accurate, lucid, and efficient translation of thought or communicative intention under the temporal constraints of online processing” (Lennon 2000: 26). On the other hand, the definitions found in the literature highlight more general spoken language competencies or proficiency, which is related to what Lennon (1990, 2000) called the broad definition of fluency. An example of such a definition is that of Meisel (1987), for whom fluency is seen as “the communicative acceptability of the speech act or communicative fit” (p. 86).

With these multiple definitions comes a variety of operationalizations. Confronted with this diversity of outlooks on fluency, Segalowitz (2010) worked on identifying reliable indicators that highlight the efficiency with which “the speaker can mobilize and temporally integrate, in a nearly simultaneous way, the underlying processes of planning and assembling an utterance to perform a communicatively acceptable speech act” (p. 47). It resulted in his proposing three broad dimensions in which fluency should be studied: cognitive, utterance and perceived fluency. Cognitive fluency is related to “the efficiency of operation of the underlying processes responsible for the production of utterances” (p. 52). Utterance fluency corresponds to “the features of utterances that reflect the speaker’s cognitive fluency” (p. 52), including pauses, hesitation and repairs. These are also referred to as the temporal dimension of fluency (Tavakoli and Skehan 2005), which can be grouped into speed of speech (i.e., speed of delivery) and flow of speech (i.e., pause behaviours, self-repairs) (Skehan 2014).

Finally, perceived fluency refers to “the inferences listeners make about speakers’ cognitive fluency based on their perceptions of their utterance fluency” (Segalowitz 2010: 48). Although perceived fluency is considered a subjective type of fluency assessment, it offers a window into utterance fluency (i.e., objective temporal characteristics; Segalowitz 2016), with which it has been found to correlate (see Suzuki et al. 2021), as well as with objective measures of general language proficiency (e.g., Bosker et al. 2012; Derwing et al. 2004; Préfontaine et al. 2016). Consequently, as perceived fluency, based on listeners’ judgments, is a reliable indicator of general language proficiency, it is considered to play a crucial role in classroom language assessment (see Suzuki et al. 2021).

Despite its potential for classroom language assessment, oral fluency is still a neglected aspect of the L2 classroom (e.g., Rossiter et al. 2010; Tavakoli and Hunter 2018). More specifically, perceived fluency is particularly relevant for practitioners since it is based on listeners’ judgments, which represent what L2 speakers experience in real life, all the while tapping into some more objective characteristics of utterance fluency (Suzuki et al. 2021). Thus, investigating perceived fluency presents valuable social and scientific implications.

Since fluency is mainly investigated as a “central aspect of a more general oral proficiency” (Peltonen 2024: 2; see also de Jong 2016; Foster 2013) and considered critical in understanding it (Kahng 2018), a speech production model such as the one put forth by Levelt (1989, 1999) is a valuable basis for its exploration (Foster 2020; see also Kormos 2006; Segalowitz 2010). It should be noted that the model was initially formulated for first language (L1) oral production but adaptions for L2 oral production were put forth by authors such as de Bot (1992) and Kormos (2006). Nevertheless, it is frequently used in its original form in L2 research (de Bot and Batyi 2022; Kormos 2006). In Levelt’s model (1989, 1999), speech production occurs in three main phases, succinctly presented in what follows. Firstly, during the conceptualizer phase, the speakers produce propositional forms corresponding to their communicative intentions. This results in a preverbal message that is transferred to the formulator for morphosyntactic processing. The product of this morphosyntactic processing is a phonological form fit for articulation, i.e. the last phase of the model. Levelt proposes that at each of the three stages of production, a self-regulator (self-monitor) checks the conformity of the message in relation, for example, to the original communicative intentions or linguistic precision. For fluent oral production to occur, the cognitive processes involved in each phase must be coordinated efficiently and effortlessly (Levelt 1989).

2.2. L2 oral fluency and cognitive resources: the role of WM

During oral production, the three phases in Levelt’s model co-occur and operate in parallel, making the process incremental (Kempen and Hoenkamp 1987). Concretely, this means that a segment of preverbal planning in the conceptualizer may become the object of articulation before the entire message is completed. This can happen smoothly only by recruiting WM, a limited-capacity system that temporarily stores and manipulates information (Baddeley 2012). More specifically, WM is necessary for keeping information used in the chain of speech available for subsequent phases by temporarily storing the different generated parts of the message (Kormos 2006; Temple 1997; see Simard 2022 for a discussion of the role of WM in Levelt’s model). This theoretical explanation of the role of WM in oral production is supported by research, as it has generally been found to be associated with various aspects of L2 oral production (see Linck et al. 2014) and L2 oral fluency in particular (e.g., Brown-Nielson and Dekeyser 2019; Gagné et al. 2022; Kormos and Sáfár 2008; Linck et al. 2014).

Since most language knowledge and processes necessary for oral production are usually highly automated in L1 adults, oral production appears effortless. L2 users’ experience may fundamentally be different. Unlike L1 users, the fact that L2 users must mobilize linguistic knowledge mastered to varying degrees places a heavy demand on their cognitive resources during oral production (see Tavakoli and Wright 2020). As these cognitive resources (e.g., WM) are limited and vary from speaker to speaker (Kormos 2006), variation in fluency may be observed (see Kormos and Sáfár 2008). Another source of variation might come from the fact that cognitive resources are amenable to the emotions experienced by L2 users during oral production, which is known to be an emotion-inducing act (see Pavlenko 2005).

2.3. Cognitive resources and emotions: foreign language anxiety and enjoyment

Cognition and emotion collaborate in shaping human experiences (Gray et al. 2002; Pessoa 2008). More specifically, “emotions affect both the focus and the process of cognition” (Fredrickson and Cohn 2008: 784) by allocating available resources to relevant aspects of the task being performed (Pessoa 2009). Emotions refer here to the preconscious physiological responses triggered automatically by reactions to internal and external stimuli (Damasio 2012; Phelps 2005) and can be experienced positively (i.e., pleasant emotion characterized as positive valence) or negatively (i.e., unpleasant emotion characterized as negative valence), and with varying intensity (i.e., the amplitude of the experienced arousal) (Russell 2003, 2009).

According to Fredrickson’s Broaden-and-Build Theory (e.g., 2001, 2003, 2006), negative emotions, such as anxiety, encourage avoidance behaviour and narrow cognitive resources and the range of thoughts and actions that come to mind. This is echoed by Eysenck and Calvo’s (1992) Processing Efficiency Theory, which proposes that, specifically, anxiety burdens WM in a way that impairs its efficiency and, in turn, impacts learning outcomes (Pekrun 2014). With regard to the L2 domain, negative emotions have been the subject of a significant number of studies since the 1970s (e.g., see MacIntyre 2017 for an overview, and Botes et al. 2020 for a meta-analysis). The results of these studies generally support the Broaden-and-Build Theory by demonstrating that high-intensity negative emotions interfere with oral production. By far, the most studied negative emotion has been foreign language anxiety, defined as “the worry and negative emotional reaction aroused when learning or using a second language” (MacIntyre 1999: 27; see also the seminal work by Horwitz et al. 1986) and its negative impact on various aspects of L2 language learning and use has been extensively described (see Gkonou et al. 2017).

In contrast, positive emotions, such as enjoyment, broaden those cognitive resources, encouraging engagement, creativity and exploration (e.g., Fredrickson and Branigan 2005). According to Pekrun (2014), enjoyment is believed to enhance the efficiency of cognitive resources, resulting in increased learning. In the L2 domain, although positive emotions as a whole have been less studied (Bielak 2022), enjoyment, defined as “the sense of satisfaction and reward generated from the activity and/or the outcome of the activity” (Ainley and Hidi 2014: 206), “capturing interacting dimensions of challenge and perceived ability that reflect the human drive for success in the face of difficult tasks” (Dewaele and MacIntyre 2016: 216), is gaining momentum (Botes et al. 2022).

It should be noted that although foreign language anxiety and foreign language enjoyment can be considered either as a state occurring specifically during an L2 task or as a trait, which exists more generally throughout various L2 experiences and in a relatively stable manner across time (see Endler and Kocovski 2001; MacIntyre and Gardner 1989), foreign language anxiety is generally considered more of a trait and foreign language enjoyment more of a state (Dewaele 2013; Dewaele and MacIntyre 2019; Pan and Zhang 2021). Indeed, foreign language anxiety seems to depend on internal factors, as evidenced by its strong association with personality traits (e.g., Dewaele and Dewaele 2020; Dewaele and MacIntyre 2019; Li et al. 2021), whereas foreign language enjoyment seems to depend more on external factors, as evidenced by its relationship with the various personality characteristics of the individuals with whom the L2 speakers interact (e.g., their language teacher; see Dewaele and MacIntyre 2019; Dewaele et al. 2018). Whether positive or negative, it is believed that emotions that are experienced as states during, for instance, L2 oral production may eventually turn into “learned” traits, which will dictate the type of emotional reaction and, consequently, lead to expectations that influence future task performances (MacIntyre and Gardner 1989; Pekrun 2006).

Nevertheless, foreign language anxiety and foreign language enjoyment may very well be experienced simultaneously (Dewaele and MacIntyre 2014; MacIntyre and Gregersen 2022) and, thus, should not be considered the two ends of a continuum (see Dewaele and MacIntyre 2014). Additionally, despite foreign language anxiety and foreign language enjoyment being negatively correlated (Dewaele and Alfawzan 2018; Dewaele and MacIntyre 2014), foreign language enjoyment has been found to outweigh foreign language anxiety’s adverse effects (e.g., Dewaele and Alfawzan 2018). This is supported by Boudreau et al. (2018), according to whom positive emotions counteract negative emotions during L2 oral production. However, as Bielak (2022) put it, more studies investigating both foreign language anxiety and foreign language enjoyment in relation to L2 performances, including oral fluency, are needed. Therefore, in the present study, we investigated them in relation to perceived L2 oral fluency.

2.4. Perceived L2 oral fluency, foreign language anxiety, foreign language enjoyment and WM

Numerous studies have examined the links between L2 oral fluency and emotions. On the one hand, the studies that specifically investigated foreign language anxiety and perceived L2 oral fluency consistently reveal negative correlations between these two variables (e.g., MacIntyre and Gardner 1994; Simard et al. 2023). For instance, in MacIntyre and Gardner (1994), foreign language anxiety was measured using a three-part scale targeting what the authors called input anxiety (i.e., “apprehension experienced when taking in information in the second language”, p. 289), processing anxiety (i.e., “apprehension experienced when learning and thinking in the second language”, p. 289) and output anxiety, (i.e., apprehension experienced when speaking or writing in the second language, p. 289). For perceived L2 oral fluency, a judge was asked to rate the degree of fluidity in the participants’ oral production (self-description). All three parts of the anxiety scale were negatively correlated with perceived L2 oral fluency. The same type of results was also observed in Simard et al. (2023) in which foreign language anxiety was measured using Dewaele and MacIntyre’s (2014) scale, and for perceived L2 oral fluency, three judges were asked to rate participants’ oral production (picture-based narration) using a flowchart.

On the other hand, to our knowledge, only one study looked at foreign language enjoyment during oral L2 fluency. In his study, Bielak (2022) investigated how foreign language anxiety and foreign language enjoyment are related to L2 oral utterance fluency among ESL speakers. Foreign language anxiety was measured using an adapted version of the Foreign Language Classroom Anxiety Scale (Horwitz et al. 1986) and foreign language enjoyment through the foreign language enjoyment scale (Dewaele and MacIntyre 2014). Both emotions were assessed as states, as the instruments were administered during an experimental task. The task consisted of small discussion groups to collaboratively decide what to take to survive on a desert island. This was followed by the recording of monologues by each participant to explain their group’s decision. Utterance fluency was measured through breakdowns (e.g., pauses) and speed (e.g., phonation time ratios) coded using the monologues. On the one hand, results revealed a weak relationship between state foreign language anxiety and foreign language enjoyment during both experimental tasks. On the other hand, they lend support to the Broaden-and-Build Theory regarding their relationship with L2 oral fluency.

As for the relationship between L2 oral fluency, emotion and cognition, the only study that investigated this relationship examined foreign language anxiety measured through Dewaele and MacIntyre’s (2014) scale, attention control measured through the Trail Making Test and a flow of speech aspect of L2 oral fluency, namely, self-repairs (i.e., Zuniga and Simard 2022). The results highlight the mediating role of attention control in the relationship between foreign language anxiety and self-repairs, with lower attentional control accounting for the adverse effects of foreign language anxiety. Speakers with higher attentional control seemed resistant to those adverse effects. Relatedly, previous studies have shown how cognitive resources, specifically WM, contribute to L2 oral fluency (e.g., Brown-Nielson and Dekeyser 2019). However, none have yet explored links between foreign language anxiety, foreign language enjoyment, WM and perceived L2 oral fluency. More specifically, no studies have examined how WM capacity interacts differentially with emotion and perceived L2 oral fluency, as observed in Zuniga and Simard’s (2022) investigation of the relationship between attentional control, foreign language anxiety and self-repairs.

Therefore, our research objective was to examine the relationship that might exist between foreign language anxiety and foreign language enjoyment, WM and perceived L2 oral fluency, by answering the following questions:

• –Is there a relationship between foreign language anxiety, foreign language enjoyment, WM and perceived L2 oral fluency?
• –If so, is this relationship distinct according to WM capacity?

3. Methodology

To answer our research questions, a cross-sectional quantitative study was conducted in which 78 French-speaking English L2 learners completed a picture narration task, an emotion questionnaire, and a WM task.

3.1. Participants

In total, 78 adults from a French-speaking Canadian university participated in the study. Of these 78 participants, 17 identified as male and 61 as female and most of them were enrolled in an education programme (however, none came from a language education or from a linguistics programme). They had a mean age of 25.35 (SD = 7.05; range = 18–48). The participants overall self-assessed using an adaptation of the Common European Framework of Reference grid (Council of Europe 2001) as intermediate (B1).

3.2. Data collection instruments

3.2.1. Picture-based narration task

The participants completed a picture-based narration task in their L2 to obtain oral production data to measure perceived L2 oral fluency. Such tasks allow for eliciting relatively realistic language in a somewhat controlled manner (Rossiter et al. 2008). We selected A Boy, a Dog, and a Frog (Mayer 1967), composed of 25 detailed illustrations, as it presents a linear storyline and a clear climax and resolution. Two minutes of planning time were given to the participants in order to increase language production (e.g., Foster and Skehan 1996; Yuan and Ellis 2003). During pre-task planning, the participants were instructed to write down notes to help them produce their narration overall storyline. They were informed that while creating their narration, they could not use their notes (see Lennon 1990).

3.2.2. Working memory task

The Highest-Number Task (Oakhill et al. 2011), a complex numerical span test, was used to measure our participants’ WM. To carry out the task, the participants had to identify and recall the highest numbers they saw on a computer screen and heard in French, their L1. Sets of three numbers were presented at a time, each composed of one number below 10 and two between 10 and 19. The participant had to identify verbally in French and recall the highest number from each set that appeared on the computer screen. The sets were organized into 15 main groups (four series of two sets, four series of three sets, four series of four sets and three series of five sets). A question mark appeared on the screen after each series, at which point the participant had to say out loud in the sequence of their presentation the recalled highest numbers.

3.2.3. Emotion questionnaire

Dewaele and Macintyre’s (2014) questionnaire measured the participants’ emotional traits among our participants. We chose to treat foreign language anxiety and foreign language enjoyment as trait emotions, even though foreign language anxiety is considered more of a trait than foreign language enjoyment, as we needed a standard measure for both. The original questionnaire was slightly adapted to consider our participants’ L1 and that they were not necessarily enrolled in language classes at the time of the study but were L2 users. We selected, adapted, and translated eight items into French to measure foreign language enjoyment and eight to measure foreign language anxiety, resulting in 16 items. Participants indicated their level of agreement or disagreement with each statement using a five-point Likert scale. High internal consistency scores were obtained for foreign language anxiety (α = 0.91) and foreign language enjoyment (α = 0.87) measures.

3.3. Procedure

The participants were met individually on the university campus. They first provided consent and completed a sociodemographic questionnaire in which they provided information regarding their age, gender, education. They were also asked to self-assess their L2 English proficiency (CEFER; Council of Europe 2001). Next, since we measured emotional traits, the questionnaire (foreign language anxiety and foreign language enjoyment) was administered before performing the cognitive and L2 oral production task. Then, the researcher explained the procedure for the WM task which the participants carried out. Finally, after the procedures for the picture-based narration task were presented, the participants planned their oral production for 2 min before narrating their story. The narrations, which lasted, on average, 3 min, were recorded and transcribed.

3.4. Data coding

3.4.1. Fluency measure

To measure perceived L2 oral fluency, we used a flowchart adapted from Turner and Upshur (1996) (see Simard et al. 2023; Appendix A). More specifically, we used the version created and piloted by Simard et al. (2023). This type of flowchart contains “ordered set of binary questions relating to boundaries between levels on the performance being evaluated” (Upshur and Turner 1995: 3) and has been used successfully in past studies to assess oral production (e.g., Simard et al. 2011; Turner 2001; White and Turner 2005). As we detailed in our previous study (Simard et al. 2023: 9), the questions contained in the flowchart are related to three characteristics of fluency identified in previous research: perceptions of fluidity (Koponen and Riggenbach 2000), lexical knowledge, including conventional expressions (Derwing 2017), and intonation (van Gelderen 1994; Wennerstrom 2000) and lead to six different levels of fluency (from 1 to 6), for a score out of six. Three experienced teachers answered the yes/no questions, for a possible maximum score of six per judge, after receiving training with recordings from ESL speakers presenting the same characteristics as our participants and performing the same narration task. As recommended, the entire oral production was used to “maximize the ecological validity of the findings” (Suzuki and Kormos 2020: 151; but also, Prefontaine et al. 2016). A strong inter-rater coefficient of reliability scores was obtained (ICC = 0.977). The scores out of six attributed by each judge were combined for a maximum score of 18 per oral production (Upshur and Turner 1995).

3.4.2. Working memory

For the WM task (Oakhill et al. 2011), we adopted the “all-or-nothing” coding procedure resulting in a maximum of 15 points, with one point being awarded for each correct answer (i.e., correct recall and correct order).

3.4.3. Emotion questionnaire

Negative items on the questionnaire were first reverse-coded. Then, the Likert scale responses for the foreign language anxiety and foreign language enjoyment constructs were added, divided by the maximum possible score and multiplied by 100 to obtain a ratio.

3.5. Data analyses

First, we verified the normality of our data’s distribution. To do so, we calculated the skewness ratios by dividing the skewness values by their standard error. The distributions were considered to be normal when the ratios fell between −2 and +2 (Larson-Hall 2010). The observation of the histograms summarizing the data distribution confirmed those results. Next, we conducted correlations to answer our first research question. This allowed us to verify whether there were any relationships between the dependent (i.e., fluency) and independent variables (i.e., WM, foreign language anxiety, and foreign language enjoyment). It also informed us regarding any potential multi-collinearity problems within the data (Larson-Hall 2010). Pearson r results were presented, except in cases of non-normally distributed data for which the Spearman rho was used. Linear regression analyses were conducted to determine the type of relationship that might exist between the variables. Prior to conducting the analyses, we verified that our data respected the assumptions associated with regressions. In that respect, the error distribution was verified by inspecting the P–P plot of standardized residuals. The normality of the error distribution was confirmed, as no significant deviation from the line was observed. Next, the homogeneity of variance was verified by observing scatterplots of the studentized residuals against the predicted value of the standardized residuals (Larson-Hall 2010). The homogeneity of the variance was confirmed by the randomly scattered data. Finally, the presence of possible outliers in the data was verified by inspecting the standard residual values. They all fell between −3.0 and +3.0, which ruled out the presence of outliers (Larson-Hall 2010).

Finally, to answer our second research question, we created two groups of participants based on their WM capacity using k-means cluster analysis – a higher and a lower WM capacity group. Then, we conducted correlation and linear regression analyses to verify whether emotion and perceived L2 oral fluency would interact differently according to the WM capacity group. The confidence interval was set at 95 %.

4. Results

This section presents the descriptive statistics for WM, foreign language anxiety, foreign language enjoyment and L2 oral fluency and the results of the inferential statistics that allowed us to answer our research questions.

4.1. Descriptive statistics: WM, foreign language anxiety, foreign language enjoyment and perceived L2 oral fluency

Table 1 presents the means, standard deviations, and skewness values for the WM, foreign language anxiety, foreign language enjoyment, and perceived L2 oral fluency data.

Table 1:

Descriptive statistics: WM, foreign language anxiety, foreign language enjoyment, and perceived L2 oral fluency.

Variables (N = 78)	Means	SD	Skewness (SE = 0.272)
WM (max. = 12)	8.18	1.89	−0.932
FLA (max. = 100)	53.31	21.02	0.190
FLE (max. = 100)	77.18	11.14	−0.360
PL2OF (max. = 18)	10.37	3.59	0.092

FLA, foreign language anxiety; FLE, foreign language enjoyment; PL2OF, perceived L2 oral fluency; SD, standard deviation; SE, standard error.

As can be seen in Table 1, our participants scored an average of 8.18 (SD = 1.89) for WM. They obtained an average score of 53.31 (SD = 21.02) for the foreign language anxiety and 77.18 (SD = 11.14) for the foreign language enjoyment. Finally, they obtained an average score of 10.37 (SD = 3.59) for fluency, which placed them at an intermediate level and confirmed their overall self-assessment.

It is also possible to observe through the skewness ratios, all of them being between −2 and +2 (Larson-Hall 2010: 78), that perceived L2 oral fluency, foreign language anxiety and foreign language enjoyment were normally distributed. Therefore, parametric statistical testing was used. The WM data were, on the other hand, negatively skewed, indicating that most participants demonstrated relatively strong WM, though without reaching a ceiling effect (mean = 8.18/12; 68 %).

4.2. Answer to the first research question

Next, we verified the relationship between our variables to answer our first research question. Table 2 presents the correlations between WM, foreign language anxiety, foreign language enjoyment and perceived L2 oral fluency.

Table 2:

Correlation matrix for WM, foreign language anxiety, foreign language enjoyment, and perceived L2 oral fluency.

	WM (p)	FLA (p)	PL2OF (p)
WM	–	−0.054 (n.s.)	0.227* (0.02)
FLA	−0.054 (n.s.)	–	−0.432*** (<0.001)
FLE	0.132 (n.s.)	−0.527*** (<0.001)	0.424*** (<0.001)

FLA, foreign language anxiety; PL2OF, perceived L2 oral fluency; n.s., not significant; *p < 0.05; **p < 0.01; ***p < 0.001.

As can be seen in Table 2, perceived L2 oral fluency is unsurprisingly positively correlated with both WM (r(78) = 0.227; p = 0.02; r² = 0.05) and foreign language enjoyment (r(78) = 0.424; p < 0.001; r² = 0.18), and negatively correlated with foreign language anxiety (r(78) = −0.432; p < 0.001; r² = 0.19). These results indicate, on the one hand, that higher scores on the WM task were associated with higher perceived oral fluency scores. Furthermore, anxious participants displayed less fluency on the narration task, while those who reported more foreign language enjoyment revealed increased fluency. On the other hand, WM did not correlate significantly with either foreign language anxiety or foreign language enjoyment. As for foreign language anxiety and foreign language enjoyment, as expected, they are negatively correlated (r(78) = −0.527; p < 0.001; r² = 0.28).

These results show that the dependent (i.e., perceived L2 oral fluency) and independent variables (i.e., WM, foreign language anxiety and foreign language enjoyment) are related. Therefore, we verified the character of this relationship between fluency and the independent variables by conducting regression analyses.

Table 3 presents the results of the multiple regression for perceived L2 oral fluency.

Table 3:

Multiple regression: examining the predictors of perceived L2 oral fluency.

Model	Total R	R 2	R 2 adjusted	WM (β)	FLA (β)	FLE (β)
1	0.539	0.290	0.244	n.s.	−0.493***	n.s.

FLA, foreign language anxiety; FLE, foreign language enjoyment; n.s., not significant; ***p < 0.001.

As can be seen in Table 3, the regression model is significant and explains 25 % of the variation (F(3,78) = 24.471, p < 0.001, R = 0.493, R² = 0.243; R²_adjusted = 0.234), with only foreign language anxiety significantly predicting fluency (β = −0.493, t(1) = −4.95, p < 0.001). Increases in foreign language anxiety appears to predict significant reductions in L2 oral fluency.

4.3. Answer to the second research question

Since the cognitive resources available for L2 oral production, such as in WM, are limited and vary among individuals (Kormos 2006; see also Dewaele 2002), we verified whether differences in our participants’ WM would lead to the observation of distinct relationships between the independent variables and L2 oral fluency. Thus, we created two groups of participants using cluster analysis on our WM data to explore further the relationship between our cognitive variable, WM, emotions, and L2 oral fluency. The analysis resulted in a weaker group (min. = 1, max. = 8) and a stronger group (min. = 9, max. = 12). Table 4 presents the descriptive statistics for each variable according to the two groups.

Table 4:

Descriptive statistics: weaker and stronger WM groups.

	Weaker WM group (n = 38)		Stronger WM group (n = 40)
	M (SD)	Skewness SE = 0.374	M (SD)	Skewness SE = 0.383
WM (max. = 12)	6.78 (1.51)	1.993	9.66 (0.847)	1.022
FLA (max. = 100)	53.97 (21.35)	0.203	52.60 (20.92)	0.150
FLE (max. = 100)	75.95 (11.10)	1.79	78.47 (11.19)	0.583
PL2OF (max. = 18)	9.63 (3.57)	0.406	11.16 (3.48)	0.212

FLA, foreign language anxiety; FLE, foreign language enjoyment; PL2OF, perceived L2 oral fluency; SD, standard deviation; SE, standard error.

As can be seen in Table 4, data are normally distributed, except for WM in each group and foreign language enjoyment in the Weaker group. Additionally, the two groups are statistically different regarding WM and perceived L2 oral fluency, with the Stronger WM Group obtaining higher scores on the WM (U = 0.000; Z = −7.732; p < 0.001) and perceived L2 oral fluency (U = 0.562; Z = −1.988; p = 0.047) tasks.

Then, we looked at the relationship between the variables within each WM group. Table 5 presents the correlations for the Weaker and Stronger WM groups.

Table 5:

Correlation matrix for WM, foreign language anxiety, foreign language enjoyment, and perceived L2 oral fluency.

	Weaker WM group (n = 38)			Stronger WM group (n = 40)
	WM (p)	FLA (p)	PL2OF (p)	WM (p)	FLA (p)	PL2OF (p)
WM	–	−0.06 (n.s.)	0.10 (n.s.)	–	−0.11 (n.s.)	0.03 (n.s.)
FLA	−0.06 (n.s.)	–	−0.57*** (<0.001)	−0.11 (n.s.)	–	−0.42** (0.008)
FLE	0.36 (n.s.)	−0.44** (0.004)	0.21 (n.s.)	0.25 (n.s.)	−0.63*** (<0.001)	0.57*** (<0.001)

FLA, foreign language anxiety; FLE, foreign language enjoyment; PL2OF, perceived L2 oral fluency; n.s., not significant; **p < 0.01; ***p < 0.001.

As can be seen in Table 5, in the Weaker WM group, fluency is only negatively correlated with foreign language anxiety (r(38) = 0.57; p < 0.001; r² = 0.33), whereas it is correlated with both foreign language anxiety (r(40) = −0.42; p = 0.008; r² = 0.18) and foreign language enjoyment (r(40) = 0.57; p < 0.001; r² = 0.33) in the Stronger WM group. WM did not significantly correlate with any of the other variables, and as expected, foreign language anxiety and foreign language enjoyment did correlate for both the Weaker WM (r(38) = 0.44; p = 0.004; r² = 0.19) and Stronger WM groups (r(40) = 0.42; p < 0.001; r² = 0.18). Again, the more foreign language anxiety the participants reported, the lower the scores they obtained for oral fluency, and that is true for both groups of participants. As for foreign language enjoyment, the more the participants reported it, the higher the score on the fluency task, but only in the Stronger WM Group.

Since the results show that fluency, foreign language anxiety, and foreign language enjoyment are related, we verified whether they exist in a linear relationship using multiple regressions. Again, we first confirmed that the data respected the statistical assumptions for regression. Table 6 presents the results from the regression analysis for the Weaker WM Group. Only foreign language anxiety was entered in the model as it was the only variable correlated with perceived L2 oral fluency.

Table 6:

Simple regression: Predicting perceived L2 oral fluency among weaker WM participants.

Model	Total R	R 2	R 2 adjusted	FLA (β)
1	0.568	0.322	0.304	−0.493***

FLA, foreign language anxiety; ***p < 0.001.

As can be seen in Table 6, the regression model is significant and explains 32 % of the variation (F(1,38) = 18.068, p < 0.001, R = 0.568; R² = 0.322; R²_adjusted = 0.304) with foreign language anxiety significantly predicting fluency (β = −0.568, t(1) = −4.251, p < 0.001).

A multiple regression was conducted for the stronger WM group to verify whether foreign language anxiety and foreign language enjoyment predict perceived L2 oral fluency. The results are presented in Table 7.

Table 7:

Multiple regression: predicting perceived L2 oral fluency among stronger WM participants.

Model	Total R	R 2	R 2 adjusted	FLA (β)	FLE (β)
1	0.581	0.377	0.319	n.s.	0.581***

FLA, foreign language anxiety; FLE, foreign language enjoyment; n.s., not significant; ***p < 0.001.

As can be seen in Table 7, the regression model is significant and explains 37 % of the variation (F(2,40) = 18.327, p < 0.001, R = 0.581; R² = 0.327; R²_adjusted = 0.319) with this time around only foreign language enjoyment significantly predicting fluency (β = 0.581, t(1) = −4.281, p < 0.001).

5. Discussion

Our study set out to investigate the relationship between WM, foreign language anxiety, foreign language enjoyment and perceived L2 oral fluency, which, to our knowledge, had never been done before. The two research questions that guided our study were: Is there a relationship between foreign language anxiety, foreign language enjoyment, WM and perceived L2 oral fluency? If so, is this relationship distinct according to WM capacity?

To answer our research questions, 78 French-speaking ESL adults completed a picture-based narration task, the highest number WM task (Oakhill et al. 2011) and a trait emotion questionnaire assessing foreign language anxiety and foreign language enjoyment (Dewaele and MacIntyre 2014).

Generally, our results confirm the coexistence of positive and negative emotions among our participants, as both were significantly and negatively correlated (MacIntyre and Gregersen 2022). Additionally, our participants’ foreign language anxiety average, similar to the 55 (2.75/5) for 1746 participants reported in Dewaele and MacIntyre (2014), indicates that they experienced moderate negative emotion intensity. As for foreign language enjoyment, they reported fairly high-intensity positive emotions, which aligns with previous studies that showed higher average foreign language enjoyment scores than foreign language anxiety (e.g., Dewaele and Alfawzan 2018; Dewaele and MacIntyre 2014; Dewaele et al. 2018). Regarding WM, it only correlated with L2 oral fluency when we considered the results for the entire group of participants (N = 78). This absence of correlation between our cognitive and emotional measures may seem to counter the claim put forth in the Broaden-and-Built Theory (see Eysenck 1979 for a discussion on the absence of a relationship between WM and general trait anxiety). However, past results (e.g., Zuniga and Simard 2022) showed an indirect relationship between a cognitive measure (i.e., attention shift) and an emotion measure (i.e., foreign language anxiety). This indirect relationship is further discussed below.

In relation specifically to the first research question, our results from the correlation analyses corroborate those of previous emotion research, which indicate that high-anxiety speakers tend to experience diminished speech performances (e.g., Phillips 1992; Tsang 2022; Woodrow 2006; Zuniga and Simard 2022) and an increase among participants reporting high states of enjoyment (e.g., Botes et al. 2022). That said, despite the similar magnitude in the correlations between perceived L2 oral fluency, foreign language anxiety (−0.432) and foreign language enjoyment (0.424), only foreign language anxiety significantly predicted fluency among the 78 participants taken together. In this sense, the results show that foreign language enjoyment did not have the expected counter effect on foreign language anxiety (e.g., Dewaele and Alfawzan 2018).

However, the analyses conducted to answer our second research question show that WM, foreign language anxiety and foreign language enjoyment manifest a complex relationship with perceived L2 oral fluency. Effectively, we observed distinct patterns of interaction when we examined the participants according to their WM scores. The analyses conducted on the WM groups (i.e., Weaker and Stronger groups) illustrate the mediating role WM plays in the emotional valence and intensity experienced by our L2 participants, offering support for the Broaden-and-Build Theory (Fredrickson 2001). In the Weaker WM group data, only foreign language anxiety significantly negatively correlated with perceived L2 oral fluency. In contrast, for the Stronger WM group, both foreign language anxiety and foreign language enjoyment significantly correlated with perceived L2 oral fluency. Thus, while the Broaden-and-Build Theory (Fredrickson 2001) predicts that negative emotions, such as anxiety, constrain cognitive resources, and positive emotions, such as enjoyment, expand them, our findings indirectly support this theory through the differential patterns observed in the two WM groups, as was the case in Zuniga and Simard’s (2022) study.

The regression analyses conducted on our data showed that foreign language anxiety predicted fluency in the Weaker WM group, and interestingly, only foreign language enjoyment predicted L2 oral fluency in the Stronger WM group. This specific result regarding the Stronger WM group seems to support Pekrun’s (2006) claim that positive emotions counteract negative ones, albeit higher intensity positive emotions. These results led us to hypothesize that L2 speakers with more robust WM experience easier processing, resulting in more fluency and positive emotions. Those state emotions solidify with experience into traits (MacIntyre and Gardner 1989). The opposite exists for L2 users with weaker WM.

Some recommendations can be formulated from these results. First, a closer look at individual variation must be taken when investigating the interaction between cognitive and emotional variables. Recall that no WM, foreign language anxiety and foreign language enjoyment relationship could be observed when all the participants’ results were analyzed together. Zuniga and Simard (2022) obtained this same type of result. Only when looking at the participants from an individual variation standpoint, i.e., weaker and stronger WM, were we able to observe how the variables interacted with oral fluency, albeit indirectly. From a practical viewpoint, the flowchart we used to assess subjective perceived fluency offers an indirect window into WM, as our participants presenting the strongest WM capacity also obtained the highest scores on the fluency measure and appeared to be the ones experiencing more foreign language enjoyment. Therefore, the flowchart could be used by teachers to understand their students’ speaking profiles better and provide them with support for experiencing positive emotions during L2 oral production. Such interventions should aim to lighten the cognitive demands of oral tasks by using, for instance, pre-tasking, online planning, and repetition of the same task.

In terms of future directions, it would be interesting to verify whether the same relationship types would be observed when measuring emotions as states (see Bielak 2022) and during different speaking tasks (see Préfontaine and Kormos 2015), monologic and dialogic (between-participant tasks) leading to different oral fluency performances (see Tavakoli 2016). Additionally, it would be interesting to compare the subjective measure we used to an objective assessment of fluency (utterance fluency) to test their interconnectivity further. More specifically, investigating the relationship between cognition, anxiety, enjoyment, and utterance fluency would allow for completing the portrait we offered in the present study.

6. Conclusions

Oral fluency is fundamentally important for different reasons, particularly in facilitating successful communication. According to past research, L2 oral fluency is modulated by cognitive and affective variables. Our study contributes new insights in that regard, as it explored the relationship between both positive and negative emotions, WM, and perceived L2 oral fluency. The results revealed distinct patterns that exist between them, with stronger WM being associated with positive emotions and better perceived L2 oral fluency, and weaker WM being associated with negative emotions and reduced fluency. More research in needed to investigate these connections across different contexts, with varying combinations of languages and proficiency levels. As could be seen throughout our manuscript, we adopt the position that cognition and emotion interact while contributing to the human experience. This fundamental premise guided our study, and we hope we successfully contributed to the collective effort to better understand the complex relationship between cognition and emotion at play during L2 oral production.

Appendix A.

Perceived L2 oral Fluency Flowchart (Simard et al. 2023: 21)