1 Introduction

Writing in a second language (L2) is a cognitively demanding task that requires the coordination of various processes, such as planning, formulating, and revising (Hayes & Flower, 1980; Kellogg, 1996). L2 writers face additional challenges due to their limited linguistic resources, which increases the cognitive load of managing multiple tasks simultaneously (Xu & Qi, 2017). Engaging in L2 writing necessitates not only the retrieval and application of linguistic knowledge but also the deployment of cognitive and motor functions coordinated by the central executive component of working memory (Galbraith & Vedder, 2019). This coordination is imperative for the effective transcription of thoughts and ideas into coherent text, making L2 writing a dynamic problem-solving activity that extends beyond mere language use.

The cognitive processes in L2 writing, including lexical retrieval, language formulation, and the use of problem-solving strategies, are complex and demanding due to the limited resources available to L2 writers (Barkaoui, 2019). This complexity highlights the need to understand the cognitive mechanisms that are crucial during the planning and revision phases of text production. Additionally, writing is not a linear activity but involves bursts of language production punctuated by pauses. These bursts reflect the cognitive load on the writer and the strategies used to manage it, while pauses indicate moments of cognitive processing and strategy application (Alves & Limpo, 2015).

Research on the relationship between pause behaviors and text quality among young EFL learners is limited. These learners, young and non-native English speakers, face unique challenges compared to adults or native speakers, and are an under-researched group in writing process studies (Garcés-Manzanera, 2021; Criado et al., 2022). They deal with the dual tasks of mastering language while developing writing skills, with their performance influenced by cognitive load and linguistic proficiency (Bereiter & Scardamalia, 1987; Sweller, 1988). There is a particular lack of understanding regarding how pause bursts (P-bursts) and revision bursts (R-bursts) affect text quality in digital writing contexts for this demographic. This study aims to fill this gap by examining the connections between language bursts, pause patterns, and text quality, while also considering the impact of writing proficiency levels on these elements in young EFL learners from Spain.

The implications of this study are twofold. Firstly, it enhances academic understanding of the cognitive processes in L2 writing, shedding light on the dynamics of language bursts and pauses in young EFL learners. Secondly, the findings offer practical insights for L2 writing instruction and assessment, providing educators, researchers, and learners with tools to address the challenges of L2 writing (Tavassoli et al.,  2022a, 2022b).

2 Writing processes in L2

In the recent decades, the exploration of L2 writing processes has been approached from multiple angles, one of which has been the observation of how writing unfolds in an online manner (see Lindgren & Sullivan, 2019, for a review). Thus, a vibrant strand of studies has emerged as a result of the interest in adding more empirical evidence to how pausing behavior occurs in the framework of these writing processes (see Barkaoui, 2019; Garcés-Manzanera, 2023; Michel et al., 2020). Although the conceptualization of pausing and revision behavior will be specified further in this paper, such elements derive from the models of writing which have been theorized in the last decades. The relevance of these models lies in the necessity of exploring the components of writing processes as well as the underlying cognitive operations and other behaviors involved. In this respect, Hayes and Flower's (1980) influential model of writing has been taken as reference for most of the body of research concerned with writing processes. In their model, three components are identified: (1) the task environment; (2) the writer's long-term memory, and (3) the writing process as a whole. Nevertheless, Hayes (2012) further refined the model by introducing components such as the control level, general process level, resource level, and a revised task environment. In this framework, cognitive mechanisms and external factors play crucial roles, with pauses being particularly significant. Considering writing's dual nature as both recursive and linear (Galbraith, 2009), these components depend on each other. Additionally, L2 writers' engagement in various processes, especially those related to content organization, frequently involves pauses, particularly in higher-order processes. For instance, when an L2 writer retrieves lexical elements from their long-term memory, this action is often followed by a shift in attention to translation or formulation processes for integrating these elements into the text. Research involving both adult L2 writers (e.g., Barkaoui, 2019; Révész et al., 2019) and children (e.g., Garcés-Manzanera, 2021; Criado et al., 2022) indicates that pausing may reflect planning activities, as evidenced by the frequent occurrence of pauses.

Bereiter and Scardamalia's (1987) model of writing, distinct from Hayes (2012) and Hayes and Flower (1980), has been influential for child L2 learners. It includes three knowledge processing approaches: knowledge-telling, knowledge-transforming, and knowledge-crafting, with an emphasis on knowledge-telling for this study. This approach is suited to lower proficiency child L2 writers who often write directly from memory, typically ignoring structured text formats. While pauses are less emphasized in this model, their importance is apparent in young EFL writers who generally write in continuous bursts. Interruptions in their writing, when they occur, usually indicate motor skill challenges or the cognitive effort needed to recall words (Kellogg et al., 2013; Schilperoord, 1996). Despite being less featured, pauses are crucial in the writing process, especially due to their frequent occurrence.

3 Pausing behavior in L2 writing

Pauses are pivotal in the writing process, accounting for up to three-quarters of the time dedicated to the task (Alamargot et al., 2007). Their significance is underscored by their measurable and observable characteristics. The ability to track pauses enables researchers to explore the underlying reasons based on their frequency, location, and duration. Crucially, studies have highlighted that pauses are indicative of specific cognitive operations, potentially mirroring the intricacies of writing processes (Barkaoui, 2019; Garcés et al., 2023). Research on pauses has revealed that underlying cognitive processes in writing draw parallels with speech production research (Spelman Miller, 2006), associating hesitations, false starts, and silent pauses with specific activities such as planning or lexical retrieval.

More specifically, in L2 writing, pauses serve as the primary observable unit, offering insights into pre-writing or online writing tasks such as planning content or language, alongside other operations: revisions, lexical searches, or formulation (Wengelin, 2006). The occurrence of pauses can signal the coordination among various processes, highlighting writing as a sophisticated task that involves managing multiple tasks simultaneously. This coordination, and the need to transition smoothly between different processes, can place additional cognitive demands on L2 writers, necessitating pauses to manage these transitions effectively (Xu & Qi, 2017).

Cognitive models of writing, notably those by Hayes and Flower (1980) and Kellogg's (1996) model, highlight the significance of pauses within the writing process. These frameworks propose that the entirety of writing activities relies on the central executive's ability to facilitate transitions between different processes. Pauses are intricately linked with key stages such as planning, formulation, and revision. In the context of L2 writing, pauses can be deliberate, representing the writer's strategic choice to shift processes, or they might stem from constraints in cognitive capacity (Alamargot et al., 2007; Kellogg et al., 2013).

Pauses in writing represent intervals where no written output is produced (Garcés-Manzanera, 2021). Such pauses do not necessarily imply an absence of cognitive processes. Among less proficient writers, like young EFL learners, pauses can result from limited cognitive capacity or resource overload. Additionally, these young writers might not have developed the motor skills required for fluent writing (Olive & Cislaru, 2015).

Research into pausing behavior in both L1 and L2 writing contexts has often been based on specific hypotheses to make sense of the data pauses provide (Barkaoui, 2019). Alamargot et al. (2007) outlined four key assumptions for examining pausing behavior: (a) pause duration, which reflects the challenges L2 writers may face and the extent of the disruption; (b) pause location, important for showing where in the text the disruption occurs, offering clues to its cause; (c) pause behavior, linked to the context of the pause or subsequent actions, and (d) graphomotor execution, indicating the central executive's overload and the difficulty in tackling complex tasks.

Previous studies on L1 and L2 writing have explored the significance of pauses, revealing that they often indicate cognitive processes like planning, especially when occurring at sentence boundaries in adult L2 learners' writing (Barkaoui, 2019; Medimorec & Risko, 2017; Révész et al., 2017). The importance of a pause extends beyond its location to include its duration, with longer pauses potentially reflecting cognitive or motor difficulties, unlike shorter pauses (Van Waes & Leijten, 2015).

The operationalization of pauses in writing research faces challenges, notably the lack of a universally accepted pause threshold. Studies vary these thresholds, commonly between 1000 to 2000 ms (ms), to suit specific research goals rather than adhere to a standard for all populations or proficiency levels. The 2000-ms threshold is widely used, as longer pauses are thought to indicate higher-level cognitive processes such as planning and revising, which are particularly significant in L2 writing research (Alves et al., 2008; Barkaoui, 2019; Tiryakioglu et al., 2019).

4 Revision behavior in L2 writing

As outlined in the previous section, pausing behavior is closely linked to another essential macro-writing process: revision. Influential writing models by Flower and Hayes (1980) and Bereiter and Scardamalia (1987), or Faigley and Witte (1981) emphasize revision as a central component of text production. Revision is commonly defined as the process in which writers re-evaluate and potentially modify their texts based on critical assessment and self-reflection (Barkaoui, 2007). The revision process in L2 writing is complex, involving a range of cognitive activities. According to Stevenson et al. (2006), L2 text revision activates several mechanisms: visually reading the text, which requires visual skills, and integrating this with the planning process, wherein the writer revises ideas before translating them into written form. The revision process involves modifying text aspects like meaning, grammar, or lexis, especially for L2 writers who often focus more on linguistic changes than content (Révész et al., 2019; Stevenson et al., 2006). Revisions are categorized into internal (inferred from pauses) and external (visible changes in text). Within these, there are precontextual revisions at the point of writing and contextual revisions occurring within the surrounding text (Lindgren & Sullivan, 2006). Our study focuses on micro-contextual elements such as characters typed per revision burst. Given that low-proficient and child L2 writers often revise lower-level linguistic units due to limited cognitive resources, focusing on these micro-contextual aspects provides valuable insights into their revision behaviors (Garcés-Manzanera, 2021; Barkaoui, 2016; Révész et al., 2019; Whalen & Ménard, 1995). Furthermore, the digital environment has been noted to enhance the frequency and scope of revisions, allowing for more detailed analysis of these processes (Li, 2018; Van Waes & Schellens, 2003).

5 Keystroke logging and bursts in L2 writing

The introduction of computers has significantly changed how we communicate in writing (Leijten & Van Waes, 2005). As more activities require computing skills for text production, computers are now prevalent in L2 classrooms, helping students produce texts more efficiently. This advancement also aids L2 researchers in studying writers' pausing and revision patterns without intrusion, allowing for natural observation of the writing process.

Keystroke logging programs are now widely used in writing research to examine the details of computer-based writing, tracking keystrokes, cursor movements, and edits to study cognitive processes during writing (Leijten & Van Waes, 2013). These programs can precisely measure pauses between keystrokes, defined as inter-keystroke intervals (IKIs), and categorize pause durations and locations (Garcés-Manzanera, 2023; Chukharev-Hudilainen et al., 2019; Van Waes & Schellens, 2003).

Moreover, keystroke logging helps analyze revisions by recording detailed logs of deletions, insertions, and replacements, providing insights into the cognitive aspects of these revisions (Lindgren & Sullivan, 2006; Stevenson et al., 2006; Van Waes & Leijten, 2015). Additionally, this method identifies "bursts" of uninterrupted writing, which are key indicators of cognitive load and writing fluency. P-bursts and R-bursts distinguish between different types of writing flows, aiding in understanding the cognitive dynamics of writing (Kaufer et al., 1986; Chenoweth & Hayes, 2001; Breuer, 2017).

Researchers should consider whether to analyze P-bursts and R-bursts together or separately, as they represent different aspects of the writing process and cognitive capacity (Galbraith & Baaijen, 2019). Previous studies have suggested that P-bursts, compared to R-bursts, provide a more accurate measure of writing fluency in adult L2 writers (Baaijen et al., 2014).

6 Text quality

Research has extensively investigated the link between text quality and writing mechanisms, a key aspect of L2 writing research (Medimorec & Risko, 2017). Studies suggest that less fluent writers often use more pauses, possibly to improve their text quality, indicating a correlation between writing fluency and text quality (Alves et al., 2008). Various researchers, including van Weijen (2009) and Tillema (2012), have noted a connection between cognitive processes during different writing stages and text quality. However, Ganem-Gutierrez & Gilmore (2018) found that L2 proficiency was the primary factor influencing essay quality, challenging earlier findings by Roca de Larios et al. (2008) and Tillema (2012) that associated variations in cognitive activities with text quality.

Recent studies have found that frequent pauses and minimal word production may indicate a negative tendency for writing quality in L2 writing (Mohsen, 2021). Additionally, longer bursts tend to occur with better transcription skills, which in turn can impact the quality of the produced text (Limpo & Alves, 2017), though this study focused on handwriting rather than digital L2 writing.

Furthermore, the quality of L2 written texts has been found to have a negative correlation with the frequencies of immediate and distant revisions, but a positive correlation with certain types of online and end revisions (Xu, 2018). This suggests that focused and extensive revising processes are beneficial for text quality, while frequent, small-scope revisions might be less effective. Additionally, pauses in descriptive essays have been shown to have a negative effect on students' writing performance (Mohsen & Qassem, 2020).

Our study aims to explore the relationship between P-bursts (pause bursts) and R-bursts (revision bursts), and their impact on text quality and other key aspects of the writing process. Previous research has individually examined the effects of pausing behavior and revisions on writing performance, but there is a lack of studies specifically investigating these dynamics in terms of P-bursts and R-bursts. Thus, this study seeks to address this gap, exploring how these bursts relate to writing quality in L2 digital writing contexts.

In this study, we aim to provide a detailed picture about how language bursts and text quality may be correlated, and to verify the existing relationship between the correlated language bursts measures and text quality. Thus, we attempt to answer the following questions:

  • RQ1. To what extent are text quality and text density associated with language bursts?

  • RQ2. Are the duration and length of language bursts related to text quality?

7 Method

7.1 Context and participants

This study involved a sample size of 22 participants who were Elementary school students with an average age range between 10 and 11 years old. The participants' English language proficiency was assessed to be at the initial levels of the Common European Framework of Reference for Languages (CEFRL), specifically within the A1 to A2 threshold. The study took place in a primary educational institution, where the participants were fifth-grade students, which is equivalent to "5º Primaria" in the Spanish educational system.

The participants received English as a Foreign Language (EFL) instruction for 3 h per week. This structured learning environment provided an opportunity to examine the language acquisition process at this foundational stage.

The setting of a primary school classroom, with its regular EFL lessons, offered a naturalistic context for observing the participants' writing behaviors and progress.

Regarding ethical considerations, we obtained signed consent forms from the legal guardians of our young participants, authorizing their involvement in the study. This research was part of a nationally funded project by the Spanish Research Agency (AEI). All procedures received prior approval from the project's bioethical committee, ensuring adherence to strict ethical standards when working with minors.

7.2 Data collection instruments and procedure

7.2.1 Instruments

Writing task

The children participated in a picture description writing activity, previously used in research (Coyle et al., 2018), involving a visual prompt composed of numbered picture frames that depicted a story sequence. This allowed the young EFL writers to follow a set narrative while composing their stories, as shown in Fig. 1 below. They were given 30 min to complete this task without any planning time and no word limit, encouraging them to write freely. This structured yet open-ended activity was designed to simulate a naturalistic writing environment and capture their genuine writing processes and language production.

Fig. 1
figure 1

Six-frame picture story task (based on Coyle et al., 2018)

Full size image

Keystroke logging software

Data collection was conducted using InputLog 8.0 (Leijten & Van Waes, 2013), with a 2000 ms pause threshold chosen based on established norms in previous L2 writing research (Garcés-Manzanera, 2021; Wengelin, 2006). This threshold effectively captures the underlying cognitive processes. InputLog facilitated automated recording of relevant measures of pausing and revision behaviors, further detailed below.

7.2.2 Data collection procedure

The research design employed a single data collection session where child participants individually wrote stories on a computer in Microsoft Word with InputLog keystroke logging software activated. This software captured and stored all keyboard entries, cursor movements, and associated timestamp data.

7.3 Analysis of keystroke logs

Our study concentrated on two primary areas: (1) pausing and revision behavior, and (2) text quality. For the first focus, various measures were automatically categorized using the extensive data processed by InputLog. For text quality, both manual and automatic analyses were performed.

In the case of pausing and revision behavior, the measures analyzed were obtained from the pause logging file output and the revision matrix. For pausing, metrics included P-Bursts (stretches of uninterrupted writing followed by pauses), providing insights into writing rhythm and cognitive demands (Breuer, 2017; Torrance & Jeffery, 1999). Additionally, the frequency of P-Bursts per minute was recorded, assessing writing fluency by showing how often writers pause and resume, which is critical for a comprehensive understanding of writing fluency (Criado et al., 2022; Van Waes & Leijten, 2015). The number of characters typed per P-Burst indicates the volume of content produced in each burst, reflecting cognitive and motor interactions in writing. Lastly, the average duration of each P-Burst was measured, revealing the time writers spend in productive phases before pausing, which helps assess writing pace and cognitive load, aiding in the diagnosis of fluency problems and guiding interventions (Speltz & Chukharev-Hudilainen, 2021).

For revision behavior, measures included R-Bursts, mean duration of R-Bursts, and Characters typed per R-Burst. R-Bursts, or revision bursts, occur when writers pause to revise, reflecting their involvement in evaluating and enhancing the text (Galbraith & Baaijen, 2019). The mean duration of R-Bursts indicates the time spent on revising, which reveals the complexity of the revision process and the cognitive effort it demands (Berninger et al., 1996). The number of characters typed per R-Burst provides insights into the scope of changes made, showing the writer’s dedication to improving text clarity, coherence, and quality (Kim et al., 2022).

Text quality was assessed using a five-point global evaluation scheme adapted from Storch (2005), focusing on content, structure, and task fulfillment. This method, also used in previous studies (Lázaro-Ibarrola & Hidalgo, 2021; Hidalgo & Lázaro-Ibarrola, 2020; Lázaro-Ibarrola, 2021), aims to comprehensively evaluate the writing quality. Additionally, the number of words produced was tracked to gauge the extent of each written response. Unlike previous research, this study also included text density as a measure of text quality. For this, a Text Analyzer (https://www.online-utility.org/text/analyzer.jsp) was used to calculate lexical density, which reflects linguistic complexity and is determined by the ratio of lexical items (content words) to the total word count, based on Ure's (1971) formula.

7.4 Statistical analyses

Backwards linear regression was employed to identify the most parsimonious model for predicting the outcome variable (ProcessTime_Pburst_Sec) from the set of predictor variables (Pburst, CharsTyped_Pburst, Rburst, and WordLength_Process). This stepwise approach is advantageous when the goal is to derive the most explanatory model while retaining only the significant predictors. The regression analysis commenced with a full model and iteratively eliminated the least significant predictors until only statistically significant predictors remained. Compared to other regression techniques, backwards regression is preferred when there is an adequate sample size and the objective is to identify the subset of predictors with the strongest relationship to the outcome variable (Plonsky & Ghanbar, 2018). To ensure the validity of the regression analysis, several assumptions were evaluated. Collinearity diagnostics indicated no severe multicollinearity issues, and the residual statistics, along with the residual plots, did not reveal any substantial violations of the assumptions of linearity, homoscedasticity, and normality of residuals, supporting the appropriateness of the backwards linear regression model.

8 Results

8.1 RQ1. To what extent are text quality and text density associated with language bursts?

The first research question examined descriptive data on pausing, revision behaviors, and text quality, alongside their correlations, as shown in Table 1. Key findings include a moderate average text quality score of 2.591 out of 4 and a low lexical density (0.49), suggesting simpler language use. The variability in the writing process is evident from the word count range of 29 to 124 words, with an average of 72.318 words. P-Bursts, indicating productive writing periods, averaged 0.757 bursts per minute with each lasting about 5.470 s. In contrast, R-Bursts, which reflect revision times, occurred less frequently at 0.243 bursts per minute but had a longer mean duration of 18.864 s, highlighting more time devoted to revisions than to initial writing.

Table 1 Descriptive statistics for all the measures
Full size table

The results in Table 2 show significant correlations between text quality and several pause and revision measures. Text quality was negatively correlated with pause burst frequency (rho = -0.544, p = 0.009), positively correlated with characters typed during pause bursts (rho = 0.567, p = 0.006), processing time during pause bursts (rho = 0.464, p = 0.030), and revision burst frequency (rho = 0.559, p = 0.007). Text quality was also positively correlated with average word length during processing (rho = 0.475, p = 0.025). However, text quality did not significantly correlate with total text density or measures related to revision bursts.

Table 2 Spearman's rho correlations between text quality and text density and all the measures
Full size table

8.2 RQ2. Are the duration and length of language bursts related to text quality?

The second research question explored the relationship between the duration and extent of language bursts (both revision and pause bursts) and text quality, as determined by rubric scores. Text quality was selected as the dependent variable for its strong correlation with other text metrics like word length and text density, as identified by Spearman Rho correlation analysis. The regression analysis, detailed in the Method section and shown in Table 3, used text quality as the dependent variable with covariates including P-Burst, Characters typed per P-Burst, mean duration per P-Burst, R-Burst, and words in the linear text, aiming to clarify these associations through various models.

Table 3 Model summary of the backward linear regression
Full size table

The analysis began with Model 1, showing a moderate correlation strength (R = 0.678) and explaining 45.9% of the variance in text quality (R2 = 0.459) with an adjusted R2 of 0.290, implying a significant explanation of variance. Through Models 1 to 4, iterative adjustments refined the model by excluding less significant covariates, evidenced by changes in R2, adjusted R2, and RMSE.

Model 5 marked a pivotal refinement, achieving an R value of 0.561 and an R2 of 0.315. Despite a slight decrease in R2 from Model 1, the adjusted R2 of 0.281 indicates enhanced explanatory power considering the fewer predictors. The RMSE for Model 5 increased marginally to 0.854, suggesting robustness despite the reduction in variables. The F-change values and corresponding p-values across the models validate the statistical significance of this refinement, with Model 5 deemed most suitable for illustrating the relationship between language production bursts and text quality.

Table 4 provides an ANOVA summary for the regression models, underscoring the statistical robustness of the findings with F-statistics and p-values. Model 1 started with a moderate explanatory power, indicated by an F-statistic of 2.718 (p = 0.058) and a regression Sum of Squares of 9.791. Over subsequent models, there was a progressive refinement in explanatory power, culminating in Model 5, which lowered the regression Sum of Squares to 6.716 and significantly raised the F-statistic to 9.198 (p = 0.007). This improvement signifies a stronger fit and robust explanation of variance in text quality, confirming Model 5 as the most statistically significant for assessing the impact of language production bursts on text quality. The progression seen from Model 1 to Model 5, marked by increasing F-statistics and decreasing p-values, supports the methodological approach and confirms the suitability of Model 5 for the study's objectives.

Table 4 Summary of analysis of variance for comparative models
Full size table

Table 5 offers a detailed breakdown of the coefficients from the regression analysis for each model, highlighting the variables' impacts on text quality. In Model 5, the data reveals that revision bursts (Rburst) significantly influence text quality, with an unstandardized coefficient of 6.631 and a corresponding standardized effect size. This effect is underscored by a t-value of 3.033 and a p-value of 0.007, emphasizing the strong predictive role of revision bursts. The substantial intercept for Model 5 indicates a baseline quality level, against which the effects of revision bursts are measured.

Table 5 Summary of regression coefficients for multiple models
Full size table

This analysis confirms that revision bursts positively correlate with text quality. Model 5 simplifies the framework while pinpointing crucial factors affecting text quality. According to this model, if a young EFL writer demonstrates two revision bursts during a task, there could be a noticeable improvement in their text quality score—approximately 0.3 on a 1 to 5 scale, factoring in the coefficient of 6.631 for revision bursts. This suggests that even minor increases in revision activity can significantly enhance the writing quality within a 5-point rubric framework.

9 Discussion

The results regarding pausing and revision behavior have shown that young L2 writers are likely to engage in these processes with the intention of improving text quality. This is evidenced by the positive correlation between revision bursts (R-bursts) and text quality, alongside the regression analysis demonstrating the significant predictive power of R-Bursts. This aspect is crucial as it demonstrates that children do not solely rely on a pre-existing knowledge-telling model (Bereiter & Scardamalia, 1987) but also attend to revisions in an attempt to enhance their text quality. Indeed, this shift points out that young learners are not merely recounting learned or visualized information, but are critically engaging with their text to improve its quality. In this sense, Vandermeulen et al.'s (2024) study with upper-secondary EFL students noted that language complexity and task demands significantly impact writing behaviors, including revision activities. Hence, engaging deeply with the text through revisions universally benefits writing quality across different linguistic contexts. Previous research (Garcés-Manzanera, 2022) has corroborated this revision behavior, where children writers devoted more resources to deleting content, suggesting a readjustment of ideas and indicating more pauses. In this sense, revision appears to be a central writing process and mechanism for young L2 writers, who may be concerned about their text quality. Remarkably, R-Bursts indicate that text production occurs between revisions, suggesting that children are transitioning from knowledge-telling towards more sophisticated writing processes, such as knowledge-transforming. Such transformations involve iterative processes between generating and restructuring ideas, which is a crucial step for developing effective writing skills (Bereiter & Scardamalia, 1987). Thus, young EFL learners may be more aware of the need to refine and evaluate their text, requiring deeper engagement with the output (Hayes, 2012). Interestingly, the emphasis on revisions is also observed – although not statistically significant – in the longer mean duration of R-Bursts (M = 18.86) compared to P-Bursts (M = 5.47). The positive impact of R-Bursts in our study suggests that young EFL learners are beginning to employ higher-order cognitive processes. Tian et al. (2024) observed similar behavioral patterns among undergraduate students, where effective planning and structured revision practices led to higher quality texts, highlighting the importance of cognitive strategies in L2 writing proficiency across different age groups. Although Tian et al.’s study involved undergraduate students, the behaviors observed align with those in our young EFL learners, suggesting a developmental consistency in how planning influences writing processes in L2 contexts. The reduced frequency of revisions and extended duration of R-Bursts in our study might similarly reflect effective initial planning that facilitates cognitive management during writing tasks. This indicates that even among younger students, strategic planning before writing can enhance the efficiency of text production, supporting the integration of planning skills into early language education.

This study also aimed to explore the correlations between pause and revision bursts, text quality, and density. The results indicate a negative correlation between P-Burst frequency and text quality (rho = -0.544) among young EFL learners. This relationship can be partially attributed to their limited processing capacity (Van Patten, 2004) and the cognitive load theory (Sweller, 1988), where the demands of writing as a problem-solving activity require young learners to allocate their attention to more challenging aspects. This negative correlation suggests that frequent pauses, potentially indicative of cognitive overload or language processing difficulties, disrupt cognitive management during writing. Such disruptions are aligned with findings from Tian et al. (2024), who observed that effective writing among adult L2 writers involved fewer revisions and more extended continuous writing segments when constructing final claims, indicating a more efficient cognitive management that minimizes disruptions to cognitive processes. Frequent P-Bursts, therefore, may indicate interruptions in these processes, potentially affecting text quality in a negative manner. However, these pauses also serve as necessary cognitive breaks, facilitating resource allocation for content creation and planning (Kellogg, 1996). Conversely, Conversely, R-Bursts show a positive correlation with text quality, implying that while excessive pausing can detract from text quality, systematic revision may enhance it. This interplay of effects places some emphasis on the subtle yet significant impact of cognitive load on the writing quality of young EFL learners. It therefore reinforces the theoretical moedls proposed by Hayes (1996, 2012), which emphasize the challenges these learners face in juggling multiple linguistic tasks and content generation simultaneously.

As regards the role held by text density in relation to pause and revision bursts, we first need to understand that young EFL learners are at a critical stage in both cognitive and linguistic development. Thus, writing in an L2 presents unique challenges which, as mentioned previously, makes the allocation of cognitive and attentional resources a necessary aspect. Text density did not correlate neither positively nor negatively with any of the language bursts measures. One of the potential explanations might be related to the cognitive load experienced by young EFL learners, who have to navigate the complexities of writing in an L2 and thus the cognitie resources are fully devoted to tasks such as vocabulary retrieval, grammar application and idea organization (Kellogg, 1996; Sweller, 1988). Thus, it seems logical that learners might prioritize the clarity and coherence of their writing over achieving a higher level of text density, which involves more sophisticated vocabulary and complex syntactic resources. This might exceed the cognitive resources young EFL learners have at their disposal. Additionally, we must bear in mind that young EFL learners are at a developmental stage where their linguistic abilities—including lexical and grammatical knowledge—are still evolving (Bereiter & Scardamalia, 1987). Since they focus on knowledge-telling strategies to produce text, their writing effort might be more focused on ensuring that their texts are understandable and coherent.

As mentioned, the absence of correlation between text density and burst measures may suggest that young EFL learners prioritize clarity and coherence over text complexity, possibly due to their developmental stage and cognitive capacities. This is futher supported by Grace Kim's (2022) findings, who found that while transcription and language skills were critical in determining writing bursts and quality, young learners often focus on manageable linguistic outputs to maintain coherence. Similarly, as they were relying on the description of a story based on six sequential pictures, they might have prioritized that such description was simple at the expense of using more varied vocabulary. Sweller's (1988) cognitive load theory equally serves to explain this finding. When faced with the complex task of writing in an L2, young EFL learners may opt to reduce the syntactic and lexical complexity of their output in an attempt to manage their cognitive resources more effectively.

10 Conclusions

This study has endeavored to contribute empirical evidence to the under-researched area of writing processes by children L2 writers, specifically in digital writing. Our first main objective was to provide an overview of language bursts and text quality measures, and to examine the correlations between these measures to identify the most associated variables. Our second main objective focused on observing the relationship between text quality and various pause and revision burst measures.

Several conclusions can be drawn: (1) Children L2 writers tend to pause frequently, which has a negative impact on text quality. While frequent pausing may reflect challenges in language processing, young EFL learners' engagement in pauses and revisions can contribute positively to text quality; (2) The role of revisions is central in L2 writing by children, as it directly influences text quality. This finding indicates that revision is not merely a matter of correcting errors but is integral to refining and improving the entire text, as supported by recent research (see Tian et al., 2024); (3) When children L2 writers produce longer texts throughout the writing process – as reflected in the number of characters typed per pause burst – it is related to higher text quality scores. This indicates more cognitive processing and the retrieval of lexical and grammatical knowledge to improve the text (see Limpo & Alves, 2017); (4) Young EFL writers' pause and revision behaviors may be understood within the framework of writing models (especially Bereiter & Scardamalia's, 1987) and the cognitive load theory (Sweller, 1988). More pauses may indicate either cognitive overload or the strategic allocation of attentional resources to a specific element of writing (see Vandermeulen et al., 2024).

One educational implication of this study is the importance of the revision process, pointing to the need for educational practices that encourage and support this process in writing instruction for young EFL learners.

This study has limitations due to its small sample size (N = 22), which restricts the generalizability of the results. Furthermore, another potential limitation is that the L2 proficiency of the young learners was not assessed in a standardized way since the information was obtained directly from their teacher. This could potentially affect the degree of pausing and revision behavior observed. In the same vein, socio-economic factors could also influence some of the pausing and revision behaviors, as well as the L2 writing performance. For instance, whether or not the learners attend after-school L2 lessons could significantly impact the overall results.

Future research should explore: (1) longitudinal studies tracking the development of pause and revision behaviors over time (Barkaoui, 2019); (2) cross-linguistic comparisons to understand L1 influence on L2 writing processes (Révész et al., 2019); (3) the impact of technology-enhanced writing instruction on these behaviors (Li, 2018); (4) the relationship between cognitive load, working memory capacity, and writing behaviors in young EFL learners (Olive & Cislaru, 2015); (5) multimodal analysis incorporating eye-tracking or think-aloud protocols alongside keystroke logging (Chukharev-Hudilainen et al., 2019); (6) task effects on pause and revision behaviors across different genres of writing (Révész et al., 2017); and (7) the role of metacognitive strategies in writing processes and how they influence pausing and revision (Barkaoui, 2016).