Introduction
Since the foundation of the heritage language (HL) education field in the United States, the development of HL students’ writing skills has been a major goal (e.g., Valdés, Reference Valdés1995). Along these lines, scholars have examined issues pertaining to HL students’ written texts in classroom contexts, particularly with Spanish-English HL writers (e.g., Colombi, Reference Colombi, Colombi and Alarcón1997, Reference Colombi, Montrul and Polinsky2021; Elola, Reference Elola2018; Torres, Reference Torres and Pascual y Cabo2016). Studies have shown that college-aged HL students report how writing is the most difficult skill for them in the HL to the extent that it can produce anxiety. Thus, HL students report a major desire to improve their writing skills to perform tasks such as translating documents for their family members among others (e.g., Callahan, Reference Callahan2010; Carreira & Kagan, Reference Carreira and Kagan2011; Hedgcock & Lefkowitz, Reference Hedgcock, Lefkowitz and Manchón2011). Most studies, however, have primarily focused on examining individual HL writing through product-based indicators including lexical diversity, accuracy, and syntactic complexity measures, among others (e.g., Bowles & Bello-Uriarte, Reference Bowles, Bello-Uriarte, Sato and Loewen2019). To date, less is known in the field of HL students’ writing processes (e.g., planning, linguistic encoding, editing) while composing written texts across the heritage and societal languages (see Elola & Mikulski, Reference Elola and Mikulski2013; Mikulski & Elola, Reference Mikulski and Elola2011, for exceptions).
To further investigate HL students’ writing processes would be pedagogically and empirically relevant. First, from the perspective of HL education, understanding Spanish-English HL students’ writing processes is crucial for developing effective writing instructional strategies that align with HL students’ writing goals and their development of a written register in the HL that moves beyond a default reliance on their oral communication skills (Colombi, Reference Colombi, Montrul and Polinsky2021; Elola, Reference Elola2018). Such evidence can also complement findings from HL students’ self-reports on specific writing areas of difficulty they encounter when writing in the HL (Torres et al., Reference Torres, Arrastia-Chisholm and Tackett2020). Second, from an empirical perspective, due to their early bilingual experience that differs from late second language learners, HL students’ writing processes can provide additional insights into the claims of theoretical writing models (e.g., Kellogg, Reference Kellogg, Levy and Ransdell1996) in accounting for the cognitive processes underlying writers’ behaviors in general. Third, it is equally empirically relevant to note that although previous research has documented the relationship between Spanish-English HL students’ writing proficiency and biographical data, such as early schooling in the HL (Gatti & O’Neill, Reference Gatti and O’Neill2017), the role of individual differences, in particular, cognitive variables like working memory, remains unexplored with HL students’ writing processes. This is an important gap given the role working memory has been purported and documented to play in first language writing (e.g., Kellogg et al., Reference Kellogg, Whiteford, Turner, Cahill and Mertens2013) and adult second language writing (see review in Kormos’s and Li’s contribution to this special issue). Additionally, although scholars have attributed HL students’ writing difficulties in the HL to their language dominance in the societal language (Mikulski & Elola, Reference Mikulski and Elola2011), previous studies have not provided evidence to support this claim. This is especially due to how HL bilinguals can widely differ with respect to their language dominance, thereby contributing to variation in linguistic performance such as the production of phonetic features in the HL (e.g., Amengual, Reference Amengual2016). Considering these issues, then, the exploration of individual differences such as working memory and language dominance can provide insight into whether and how HL writers differ from one another in tackling writing tasks.
Therefore, to address these goals, the current study examined the writing processes of a group of college-aged Spanish-English HL writers, who were enrolled in a tailored writing course for HL bilinguals, in executing writing tasks in the heritage (Spanish) and societal (English) languages by integrating keystroke-logging and think-aloud data. The study is framed in the theoretical foundations of Kellogg’s cognitive writing model (Kellogg, Reference Kellogg, Levy and Ransdell1996), the phonological and executive working memory framework (Wen, Reference Wen2016), and language dominance as a multidimensional construct (Birdsong, Reference Birdsong, Silva-Corvalán and Treffers-Daller2016).
Background
HL bilinguals’ writing behaviors
Previous studies have reported that Spanish-English HL bilinguals have expressed a need to develop their writing skills in the HL when enrolling in formal Spanish courses at the university level (e.g., Hedgcock & Lefkowitz, Reference Hedgcock, Lefkowitz and Manchón2011). This has prompted research efforts in classroom settings to document HL students’ writing behaviors when composing texts in Spanish (the HL). Using think alouds (TAs) as a method to tap into Spanish-English HL writers’ thought processes while composing texts in Spanish, Schwartz (Reference Schwartz, Roca and Colombi2003) found that her three participants relied more on their intuitions (i.e., what sounded right to them) when writing in the HL (Spanish) and they also used translation strategies from English to Spanish to compose their ideas in Spanish. In a follow-up study with five Spanish-English HL writers, Schwartz (Reference Schwartz, Ortiz López and Lacorte2005) reported that the more fluent HL writers in Spanish produced more lexical errors and employed overall more writing strategies except when editing surface linguistic constructions (e.g., orthography). In contrast, the less fluent HL writers employed more strategies during the editing of surface constructions and overall produced more verb tense errors.
Mikulski and Elola (Reference Mikulski and Elola2011) and Elola and Mikulski (Reference Elola and Mikulski2013) were the first studies to examine the writing processes a group of 12 Spanish-English HL bilinguals, who were enrolled in an intermediate Spanish language course, through their pausing and revision behaviors. Mikulski and Elola’s (Reference Mikulski and Elola2011) results revealed that their HL writers’ pauses between sentences were significantly longer when writing in Spanish (their HL), suggesting that HL writers spent more time planning on the formulation and organization of their ideas when composing texts in Spanish. Elola and Mikulski reported on the revision behaviors of the group in the 2011 study. Their findings, which included participants’ revisions for meaning and surface linguistic constructions, showed no significant differences between the revisions of texts in English and Spanish. Put differently, their participants exhibited similar editing processes across both the heritage (Spanish) and societal (English) languages.
To date, these are the only studies that have investigated Spanish-English HL participants’ writing processes, which included HL writers’ verbalizations of their thought processes while composing texts as well as their pausing and revision behaviors. To further advance the field’s knowledge of HL bilinguals’ writing processes, a number of issues need to be addressed. First, with the goal of making eventual generalizations about Spanish-English HL bilinguals’ writing processes, it is critical to gather more data given that these studies reported on small sample sizes. Second, as Mikulski and Elola (Reference Mikulski and Elola2011) pointed out, a combination of both quantitative (e.g., keystroke logging) and qualitative (e.g., TAs) data are needed to elucidate HL bilinguals’ writing processes to clearly address the source problem (e.g., text organization, surface linguistic structures) for HL writers’ pausing and revision behaviors when composing texts in the heritage vis-à-vis the societal language. Along these lines, such methodological consideration of combining keystroke-logging and TA methods has been addressed in second language writing research to gain more valid insights into writing processes (see Révész & Michel, Reference Révész and Michel2019). Both methods also complement each other by addressing the limitations of only using one of these methods. Keystroke-logging data, on one hand, cannot capture writers’ direct cognitive processes, but it is unobtrusive in recording writers’ keystrokes and mouse movements, which does not interrupt the writing process. On the other hand, although TAs can provide information on writers’ direct cognitive processes, the obtrusive nature of thinking aloud while writing can alter a writer’s composing process leading to divergent outcomes. Third, given the well documented inter- and intravariability that HL bilinguals exhibit when producing the HL (for reviews, see Aalberse et al., Reference Aalberse, Backus and Muysken2019; Montrul, Reference Montrul2016a), the field can benefit from a comprehensive view of how HL writers’ individual differences (e.g., working memory, language dominance) can contribute to affordances and/or challenges during the writing process.
Kellogg’s cognitive writing model
Mikulski and Elola (Reference Mikulski and Elola2011) and Elola and Mikulski (Reference Elola and Mikulski2013) framed their studies within Kellogg’s (Reference Kellogg, Levy and Ransdell1996) cognitive writing model. Due to its considerable relevance and influence in first language (L1) and second language (L2) writing studies, the current study also adopts Kellogg’s (Reference Kellogg, Levy and Ransdell1996) model to advance our knowledge of HL bilinguals’ writing processes. Therefore, the findings of this study can expand on the hypotheses purported in Kellogg’s (Reference Kellogg, Levy and Ransdell1996) model to an understudied population of writers whose HL is at least one of their L1s, which has most likely undergone significant dynamic and adaptive changes due to its sociolinguistic status as a minority language in contact with a societal language. Empirical support for Kellogg’s (Reference Kellogg, Levy and Ransdell1996) model with a HL population is not clear with the studies from Mikulski and Elola due to methodological shortcomings—namely, a lack of qualitative data to explain HL writers’ pausing and revision behaviors to clearly demonstrate whether and how these behaviors align with Kellogg’s model. Furthermore, the two studies did not include a measure of working memory (WM), which is an integral component of Kellogg’s model.
Kellogg’s (Reference Kellogg, Levy and Ransdell1996) cognitive writing model made theoretical claims for the integration of WM in supporting six cognitive processes that writers undergo when composing a text. In the model, Kellogg described three main cognitive phases—formulation, monitoring, and execution—along with two processes under each phase. Importantly, these processes ought to be viewed as interactive and recursive. Formulation involves a planning process for the generation and organization of ideas that subsequently are encoded into linguistic forms in writing known as the translating process. The execution phase deals with the motor skills required to accomplish the writing task. Thus, a programming process refers to the preparation of the output from the translating phase for the corresponding motor system, which can be handwriting or typing. The executing process entails the actual muscle movements associated with the action of handwriting or typing. Last, the purpose of the monitoring phase is for writers to ensure that their written text matches their writing intent or goal. Therefore, writers engage in reading and editing processes to adjust their written texts accordingly. To understand these writing processes, prior writing studies have used keystroke-logging and TA data to examine writers’ pausing and revision behaviors (e.g., Manchón et al., Reference Manchón, Roca de Larios, Murphy and Manchón2009; Sullivan & Lindgren, Reference Sullivan and Lindgren2007). Pausing can provide insights into writers’ allocation of attentional resources during writing processes to determine whether writers are accessing conceptual (i.e., planning) or linguistic resources (i.e., translating), whereas revisions are a signal of writers’ monitoring processes during which editing is made to the written text (e.g., Lindgren & Sullivan, Reference Lindgren, Sullivan, Sullivan and Lindgren2006; Schilperoord, Reference Schilperoord, Rijlaarsdam, van den Bergh and Couzijn1996).
These six cognitive processes that are associated with writing have also been useful to capture L1 and L2 writers’ behaviors. In the case of L1 writing, most studies have focused on these cognitive processes to distinguish between skilled and unskilled L1 writers among school age children (e.g., McCutchen, Reference McCutchen1996; Olive, Reference Olive and Berninger2012). Likewise, L2 studies have applied these processes to investigate adult L2 writers’ underlying cognitive processes when composing texts in the L2 (e.g., Révész et al., Reference Révész, Michel and Lee2019). For example, some findings have shown that Spanish-English L2 writers allocate similar composing time to planning and translating processes in both L1 (Spanish) and L2 (English) writing. However, the time allocated to translating processes in the L2 seemingly decreases with higher proficiency in the L2, which allows these L2 writers to direct attentional resources to other writing processes (e.g., Manchón & Roca de Larios, Reference Manchón and Roca de Larios2007; de Larios et al., Reference de Larios, Marín and Murphy2001).
Both L1 and L2 writing are complex cognitive tasks given that writers need to hold useful information (e.g., writing goal, language use), which is often retrieved from long-term memory, while also using such information to address content and rhetorical issues. Given these writing dynamics, Kellogg’s (Reference Kellogg, Levy and Ransdell1996) model also frames how WM is involved in writers’ processes of planning, translating, and editing written texts. By integrating Baddeley’s (Reference Baddeley1986) model of WM, Kellogg’s model specified the WM components associated with each writing process. The central executive component of WM, which regulates attentional control resources, is virtually associated with every process except for executing motor skills. Put differently, the other five processes will place demands on the central executive component of WM. The model postulates that the planning process also places cognitive demands on the visuospatial sketchpad (e.g., the mental visualization of the planning of ideas), whereas the translating and reading processes place demands on the phonological loop, which refers to the individual’s inner speech that temporarily stores words, clauses, and sentences to be used during the executing phase.
Working memory in L1 writing
In their evaluation of the predictions of Kellogg’s (Reference Kellogg, Levy and Ransdell1996) model on WM and the cognitive processes involved in L1 writing over 17 years, Kellogg et al. (Reference Kellogg, Whiteford, Turner, Cahill and Mertens2013) concluded that the central executive is the primary WM component involved in the different writing processes, with supportive roles from the phonological loop for linguistic processes in the translating phase and from the visuospatial sketchpad for generating ideas in the planning phase. Though not conclusive, as other areas need to be addressed such as the link between editing and WM, these findings support the usefulness of Kellogg’s model in shedding some light on the demands L1 writing places on WM. More specifically, Kellogg’s et al.’s (Reference Kellogg, Whiteford, Turner, Cahill and Mertens2013) research synthesis reported on weak to moderate relationships between WM and L1 writing among younger students due to necessary attentional resources invested in their motor skills (e.g., handwriting), which, in turn, limits their focus to planning, translating, and editing. Studies with older students with developed motor skills, on the other hand, have demonstrated stronger relationships between WM and writing processes involving formulation and monitoring. Importantly, these writing studies contribute to the broader claim that individual differences in WM are related to variation in L1 development and processing among typically developing children and adolescents (e.g., Adams & Gathercole, Reference Adams and Gathercole2000; Gathercole & Baddeley, Reference Gathercole and Baddeley1993; Wen, Reference Wen2016).
Working memory in L2 writing
Similarly, the field of L2 acquisition has made claims on the links between WM and L2 development, performance, and processing (for recent reviews, see Jackson, Reference Jackson2020; McCormick & Sanz, Reference McCormick, Sanz, Schweiter and Wen2022; Serafini, Reference Serafini, Kersten and Winsler2023; Tsai et al., Reference Tsai, Au, Jaeggi, Granena, Jackson and Yilmaz2016; Wen, Reference Wen2016; Wen & Jackson, Reference Wen, Jackson, Li, Hiver and Papi2022; Wen & Skehan, Reference Wen and Skehan2021; see also Kormos’s and Li’s contributions to this special issue). These recent accounts on the nexus between WM and L2 development have included six major claims. First, WM capacity is limited, leading to trade-off effects in L2 learning. Second, WM is a portal to L2 knowledge stored in long-term memory. Third, WM is the primary construct in L2 language learning aptitude. Fourth, WM capacity varies among L2 learners indicating individual differences in the ability to combine the temporary processing and storing of L2 linguistic information. Fifth, phonological and executive control WM are the main subcomponents involved, albeit differentially, in L2 learning and processing. Sixth, WM conspires in different aspects of L2 development, which include but are not limited to L2 input processing, instructed L2 learning, and effects of corrective feedback. According to findings from meta-analytic studies (Li, Reference Li and Gurzynski-Weiss2017; Linck et al., Reference Linck, Osthus, Koeth and Bunting2014; Watanabe & Bergsleithner, Reference Watanabe, Bergsleithner, Madden-Wood and Ueki2006), WM is a significant contributor to L2 processing and outcomes though such association is minor. For example, Linck et al. (Reference Linck, Osthus, Koeth and Bunting2014), which included the largest number of studies (k = 79), found an overall effect size of .255, with executive control WM being more strongly related to L2 outcomes.
Additionally, WM has been associated with skills such as listening, speaking, reading, and writing in the L2 though fewer studies have examined WM and adult L2 writing performance (see Wen, Reference Wen2016). These few studies have mostly investigated the relationship between WM and L2 writing performance, as measured by complexity, accuracy, lexis, and fluency measures as well as by L2 writers’ use of specific linguistic constructions such as gender agreement. The findings have revealed significant strong to moderate correlations between WM and syntactic complexity and fluency, but mixed findings with respect to accuracy and lexical performance (Bergsleithner, Reference Bergsleithner2010; Kormos & Sáfár, Reference Kormos and Sáfár2008; Lu, Reference Lu, Wen, Mota and McNeill2015; Révész et al., Reference Révész, Michel and Lee2017; Yi & Luo, Reference Yi and Luo2013). Concerning morphosyntax in L2 Spanish writing, in particular, Zalbidea’s (Reference Zalbidea2017) study found a significant negative correlation between WM and Spanish gender agreement errors in the writing of participants who completed a complex writing task. However, Zalbidea and Sanz (Reference Zalbidea and Sanz2020) reported that WM was more strongly associated with accurate morphosyntactic performance in their speaking condition relative to the written one; however, visuospatial WM contributed to written accuracy for a less salient Spanish linguistic structure. In a study examining the relationship between WM measures and L2 writing processes, Révész et al. (Reference Révész, Michel and Lee2017) found the following significant correlations: (a) visuospatial span and less frequent pauses to gaze at task instructions, (b) operation span measuring updating ability and fewer pauses between paragraphs, and (c) color–shape task measuring task-switching abilities and longer pauses between sentences. More recently, in a study with English-Spanish L2 writers, Vallejos (Reference Vallejos2020) found a significant relationship between visuospatial WM and pause frequency between sentences during L1 (English) writing as well as pauses frequency within words during L2 (Spanish) writing. Further, a significant relationship also emerged between operation-span WM and pause frequency between sentences in participants’ L2 Spanish writing. Overall, these findings indicate that different WM components are seemingly related to L2 writing products and processes. However, based on these few studies on the role of WM in L2 writing, it is evident that no conclusions can be drawn, so more research is needed to better understand when and how WM contributes to L2 writing.
To extend the investigation of WM and writing processes to a population of HL writers, the current study adopts Wen’s (Reference Wen2016) phonological and executive WM framework, which consists of four levels that include long-term memory that stores a multilingual individual’s knowledge of all their languages (e.g., mental lexicon, grammatical knowledge), WM components (i.e., phonological WM, executive control WM), cognitive mechanisms associated with each WM component (e.g., updating, switching), and the specific measures for each WM component (e.g., operation-span tasks). This WM framework aligns, in part, with Kellogg’s (Reference Kellogg, Levy and Ransdell1996) cognitive writing model in that it integrates the role of WM components such as phonological and executive control WM. That is, the phonological and executive control WM components in Wen’s (Reference Wen2016) framework allows researchers to test Kellogg’s predictions with respect to the cognitive load writing places on WM resources, especially the role of executive control WM. Furthermore, a major benefit of Wen’s (Reference Wen2016) framework is including the activation of linguistic knowledge stored in long-term memory, which allows for the examination of the potential synergy between HL writers’ long-term memory and WM. Of equal relevance is that this activation of linguistic knowledge can address HL writers’ access to their stored linguistic resources that would allow for the execution of writing processes (e.g., linguistic encoding, editing), as captured in Kellogg’s writing model. Given the prominent role of executive control WM in writing and L2 outcomes (e.g., Kellogg et al., Reference Kellogg, Whiteford, Turner, Cahill and Mertens2013; Linck et al., Reference Linck, Osthus, Koeth and Bunting2014) and the goal of the current study to explore WM in HL writing processes, this study examines the contribution of executive control WM to HL writers’ pausing and revision behaviors across both of their languages, as measured by an advanced operation span WM task.
Language dominance
In addition to examining HL writers’ individual differences in WM, this study includes the potential contribution of language dominance to the variation observed in HL writing. The current study characterizes language dominance as a multidimensional construct that refers to a bilingual’s ability to access with easiness one language over the other due to a constellation of factors such as language proficiency, age of acquisition, frequency of language use across different domains at a given point, and language attitudes (Birdsong, Reference Birdsong, Silva-Corvalán and Treffers-Daller2016). It is important to highlight that proficiency is a dimension of language dominance, but it is not the same construct. Proficiency specifically refers to a bilingual’s linguistic ability related to grammatical knowledge such as vocabulary and morphosyntax (Montrul, Reference Montrul, Silva-Corvalán and Treffers-Daller2016b). Of equal relevance is that language dominance can shift in a bilingual’s lifetime, which is the typical case of bilingual children who learn a L1 in the home that is different from the one used in the school and spoken in the society at large (Peña et al., Reference Peña, Bedore, Torres and Francis2021). Considering language dominance as a multidimensional construct, as measured by the Bilingual Language Profile questionnaire (Birdsong et al., Reference Birdsong, Gertken and Amengual2012), empirical studies have demonstrated that language dominance predicts HL bilinguals’ production of phonetic features that align to monolingual norms as well as their processing of emotional words in the societal language (Amengual, Reference Amengual2016; Vargas Fuentes et al., Reference Vargas Fuentes, Kroll and Torres2021). However, with respect to HL writing, very little is known about the role of language dominance. And, although Mikulski and Elola (Reference Mikulski and Elola2011) attributed their findings to their participants’ language dominance in English, they do not provide an independent measure of language dominance to support such a claim. Therefore, the current study also addresses this issue by measuring HL writers’ language dominance of Spanish and English with the Bilingual Language Profile questionnaire used in previous studies.
Research questions
By integrating keystroke-logging and TA data, the current study seeks to expand our understanding of individual HL learners’ writing processes by examining their pausing and revision behaviors and their underlying cognitive processes while composing written texts in the heritage and societal languages. Additionally, given the lack of studies investigating the role of individual differences among HL writers, this study explored whether and how WM and language dominance contributed to HL writers’ pausing and revision behaviors. Therefore, the following research questions (RQs) were considered:
RQ#1. Do HL writers’ pausing behaviors differ between Spanish and English? What are HL writers’ underlying cognitive processes when pausing during the completion of a writing task?
RQ#2. Do HL writers’ revision behaviors differ between Spanish and English? What are HL writers’ underlying cognitive processes when revising during the completion of a writing task?
RQ#3. What are the contributions of WM and language dominance to HL writers’ pausing and revision behaviors during the completion of writing tasks in the heritage and societal languages?
Methods
Participants
The current study tested 61 Spanish-English heritage bilinguals who were enrolled in a university-level Spanish hybrid writing course tailored for heritage speakers in Southern California at the time of the study. The writing instruction of the course is focused on writing tasks (e.g., write a letter to your Latinx students’ parents/caregivers about using Spanish in the household) and argumentative essays on different Latinx experiences in Southern California in conjunction with the provision of explicit instruction and corrective feedback on writing conventions, strategies, and language use according to individual students’ writing needs. Participants’ enrollment in the class was determined by a local department placement test designed for heritage speakers.
Participants completed the Bilingual Language Profile questionnaire, which is available in the IRIS digital repository website (Birdsong et al., Reference Birdsong, Gertken and Amengual2012), to estimate their language dominance index score and to gather information about their language background history. Based on participants’ responses on the questionnaire, a dominance index score was also calculated that could range from -218 to 218. A positive score indicates English dominance, whereas a negative score indicates that a participant is dominant in Spanish. Scores that are close to 0 indicate more balanced bilingualism, which means that a participant is not dominant in either language. Importantly, this should not be confused with being at monolingual native levels in both languages (Birdsong, Reference Birdsong, Silva-Corvalán and Treffers-Daller2016). Results demonstrated an average language dominance index among the participants of 35.22 (SD = 44.96, Min. = -99.43, Max. = 198.50). Due to a number of Spanish dominant participants in the current sample, Table 1 below reports on separate demographic data for Spanish and English dominant participants.
Table 1. Participant demographic information
Note: Standard deviations are in parentheses.
a Language use is based on the average percentage of time during the week participants use English and Spanish with friends and family as well as at school and/or work.
b Self-rated proficiency is based on a scale of 1 to 6, with 6 indicating very well.
Based on the demographic information in Table 1, both Spanish- and English-dominant participants learned Spanish from birth but Spanish-dominant participants, on average, reported learning English during their elementary school years. Spanish-dominant participants spent more years studying in Spanish and both groups of participants use a higher percentage of English during the week. With respect to self-rated proficiency in English, Spanish-dominant participants differed from their English-dominant peers with lower self-ratings for speaking and writing in English. In contrast, English-dominant participants reported lower self-ratings for speaking and reading in Spanish. Both groups, however, reported comparable and the lowest self-ratings for writing in Spanish.
Materials
Writing tasks
The topics for the writing tasks were adopted from a previous writing study (Roca de Larios et al., Reference de Larios, Marín and Murphy2001). Each participant completed two writing tasks counterbalanced by language (Spanish and English) and writing prompt. The writing task goal for each version consisted of participants contributing a three- to four-paragraph essay for a special issue of the local campus newspaper on the topic of educational success. For one prompt, participants responded to whether success in education was more influenced by a student’s home life or the quality and effectiveness of the educational program. The second prompt was concerned with whether school failure is due mostly to the lack of responsibility and commitment among teachers or students’ aptitude, effort, and motivation (see Appendix Supporting Information 1 online for all four versions of the writing prompts per language). Both writing task versions required participants to produce an argumentative essay with a similar topic to minimize genre and topic effects. Prior to their writing, participants had 3 min to brainstorm and to plan out their response to the prompt in any language. The researcher also answered any questions the participant had about the prompt. Students completed the writing tasks by using the keystroke-logging software, InputLog 7.0 (Leijten & van Waes, Reference Leijten and Van Waes2013), which uses the program Microsoft word for word processing. Participants were not timed during their completion of the writing tasks because the study was not concerned with fluency measures and we wanted to avoid raising participants’ anxiety levels. Furthermore, this aligns with participants’ experience in the writing course given that they are not timed to complete their writing assignments. Therefore, they were finished once they completed at least three paragraphs, as per writing task instructions. The use of an online dictionary or thesaurus was not allowed during the writing process to examine participants’ use of their own linguistic resources (or lack thereof) in composing essays in both languages, with the goal of better understanding HL writers’ needs.
Think alouds
A subset of 16 participants was instructed to TA while completing their writing tasks in both languages to tap into their direct cognitive processes. Participants watched a video with instructions to think aloud while writing, and they were allowed to think aloud in the language of their choice including code switching. To ensure that participants were clear on thinking aloud, the researcher modeled thinking aloud while composing an essay with a format similar to the one in the video. The researcher answered any questions the participant had about thinking aloud while completing their writing tasks. The participants practiced thinking aloud during their 3-min brainstorming session. During the writing session, if participants stopped thinking aloud at any point, a research assistant showed them a sign to remind them to think aloud. Participants’ verbalizations were recorded with Audacity 2.0.5 and a digital recorder. A concern with TAs as a research methodology, however, is the reactivity it can induce in the experimental results (e.g., Bowles, Reference Bowles2010; Révész & Michel, Reference Révész and Michel2019). Reactivity occurs when the act of completing two tasks—writing and verbalizing their thoughts—affects the very processes that verbalizations are supposed to unveil. With HL bilinguals, previous writing studies have included TAs in the study design (Schwartz, Reference Schwartz, Roca and Colombi2003; Yanguas & Lado, Reference Yanguas and Lado2012). Yanguas and Lado (Reference Yanguas and Lado2012) found positive reactivity in their study in that HL writers assigned to the TA condition showed significantly superior performance on accuracy and fluency measures.
Therefore, given the potential issue of reactivity when employing TA methodology, and in line with recommendations from previous studies (e.g., Leow & Morgan-Short, Reference Leow and Morgan-Short2004), a comparison between participants who engaged in TAs and participants who did not (see Appendix Supporting Information 2 online for a table with the descriptive statistics for each group by language) was run on the different pausing and revision scores (dependent measures). Two separate multivariate analyses of variance for Spanish and English writing behaviors were run, with pausing and revision scores as within-subject variables and TA or non-TA groups as between-subject factors. No significant differences emerged between the TA and non-TA groups for writing behaviors for the Spanish writing task, F(5, 62) = 1.53, p = .19, partial η2 = .04, or the English writing task, F(5, 62) = 2.02, p = .09, partial η2 = .07). The nonsignificant findings and modest effect sizes indicate that in the current study thinking aloud during task completion did not induce reactivity; therefore, the results of the TA group were also included in the current study’s analysis.
Advanced Ospan test
To estimate individual differences in WM, participants completed an advanced operation-span (Ospan) test for a college-aged population developed by Engle’s attention and working memory lab (Draheim et al., Reference Draheim, Harrison, Embreston and Engle2017; Unsworth et al., Reference Unsworth, Heitz, Schrock and Engle2005), which was administered in E-prime 3.0. The advanced Ospan is a complex span task that has both storage and processing components. Participants were instructed to solve simple math problems by indicating whether a proposed answer on the screen (e.g., “7” or “5”) was a true or false solution to a math problem (e.g., “(7/7) + 6 =”) while also remembering a letter (e.g., “R”) that was presented at the end of each math trial. After a few trials, participants selected the letters in the order presented in the set of trials (see Figure 1 for a visual representation). Participants also received feedback on the number of letters they recalled accurately and the number of math errors in each set of trials. Standard operation-span tasks typically have set sizes of three to seven trials, but this range seems to be inadequate to discriminate between high or even average ability participants, especially among college-aged students. The advanced Ospan included sets of eight and nine trials, and it better discriminates WM in a college-aged population (Draheim et al., Reference Draheim, Harrison, Embreston and Engle2017). To ensure participants’ attention to the math problems, the math trials had a participant-adaptive response deadline of 2.5 standard deviations above the mean reaction time on the practice trials. Further, participants whose math scores were below 85% accuracy were not included in the final analysis. Participants’ advanced Ospan scores were determined by the total number of letters recalled in their accurate serial position.
Figure 1. Advanced Ospan working memory test.
Data collection procedures
Participants were recruited from Spanish writing courses designed for heritage bilinguals. Each participant completed four sessions in a computer and research laboratory upon consenting to participate in the current study. During the first session, participants went to a computer lab as a class to get more information on the study and to complete the Bilingual Language Profile questionnaire if they agreed to participate in the study. In the second session, participants completed the advanced Ospan WM test. For sessions three and four, participants completed a writing task per session, wherein the writing tasks were counterbalanced by language—Spanish or English—and topic to avoid ordering effects. Participants received extra credit points for their participation in the study.
Data analysis procedures
Keystroke-logging data analysis
Participants’ pausing and revision behaviors were recorded with the program InputLog while they were completing their writing tasks for each language. InputLog processes the key-logging data and provides summary reports with participants’ average pausing and revision times. With respect to pausing behavior, in line with previous writing research (e.g., Révész et al., Reference Révész, Michel and Lee2019), a pause threshold was set at 2 s in InputLog. The average pausing times were also extracted based on the textual pause location, including within words, between words, and between sentences. A hierarchy exists with pause times ranging from long pauses to shorter ones, wherein longer pauses are associated with sentence boundaries and shorter pauses with between and within word boundaries. Previous research has shown that pauses between sentences typically indicate writers’ engagement in the planning of content, whereas pauses between and within words point out to the translating (or linguistic encoding) of ideas (e.g., Révész et al., Reference Révész, Michel and Lee2019; Schilperoord, Reference Schilperoord, Rijlaarsdam, van den Bergh and Couzijn1996). For the analysis of revision data, the average times of r-bursts and number of word revisions were extracted from the InputLog’s summary report. R-bursts refer to timed series of writing events in which participants made overall revisions to their texts, as characterized by deletions and insertions to modify the text. Each r-burst was bounded by the time the participant began and ended the deletion or insertion of characters in previously produced written text. Thus, participants’ average r-burst times were analyzed per writing task. A second revision analysis consisted of the specific number of word revisions participants made when writing in Spanish and English. The analysis of revision data signals the writer’s editing process, during which modifications are made to the written text ranging from content and organization to lexical and grammatical accuracy (Lindgren & Sullivan, Reference Lindgren, Sullivan, Sullivan and Lindgren2006).
Think-aloud analysis
The TA data consisted of a total of 360 pausing and revision episodes of which 234 episodes emerged during participants’ completion of the Spanish writing task and 126 episodes emerged during the English version. The pausing and revision episodes were transcribed for coding purposes according to textual location (i.e., within words, between words, between sentences) and emerging categories (e.g., spelling, generating ideas, organization, lexical retrieval). The emerging categories were then classified into planning, translating, and editing phases based on Kellogg’s (Reference Kellogg, Levy and Ransdell1996) model. An independent coder was trained to code 15% of the TA episodes. The independent coder and researcher coded the same 15% of TA episodes and then discussed any discrepancies until reaching agreement. An analysis of interrater reliability yielded K = .96, which was calculated after consensus between both coders. The researcher coded the remaining TA episodes. Tables 2 and 3 below display the coding categories and examples that emerged from the TA data for both pausing and revision episodes.
Table 2. Think-aloud (TA) coding categories for pauses
Note: Nonitalicized words indicate students typing their essays. Italicized words indicate students’ pauses from typing. English translations are provided in brackets for Spanish examples.
Table 3. Think-aloud (TA) coding categories for revisions
Note: Nonitalicized words indicate students typing their essays. Italicized words indicate students’ verbalizations of their revisions. English translations are provided in brackets for Spanish examples.
Statistical modeling
The quantitative data of the current were analyzed using mixed-effect models from the lme4 R-package in Rstudio (version 2022.02.3+492), as recommended by previous second language research examining individual differences (Linck, Reference Linck, Granena, Jackson and Yilmaz2016). Baseline models were first created that consisted of a dependent variable (e.g., pausing time, r-burst time, number of word revisions) and independent variables (e.g., language, pause location). The baseline models were then compared with alternative models that included the predictor variables—WM, language dominance—using a likelihood-ratio test through the ANOVA function in the ImerTest R-package to determine which statistical models were a better fit for the data. The quantitative results in the next section were based on the best-fitting statistical models (see Appendix Supporting Information 3 online for model fit statistics).
Results
Descriptive statistics
Beginning with participants’ overall performance on the writing tasks, their average time spent on the Spanish writing task (M = 39:18, SD = .010, Min. = 22:47, Max. = 60:18) was slightly longer than the English writing task (M = 34:01, SD = .011, Min. = 13:43, Max. = 58:09). Further, on average, participants made more pauses when writing in Spanish (M = 1,929.46, SD = 593.70, Min. = 749, Max. = 3,212) than in English (M = 1,740.44, SD = 661.80, Min. = 420, Max. = 3,260). Likewise, participants had more revision episodes in Spanish (M = 414.59, SD = 240.70, Min. = 63, Max. = 1,181) than in English (M = 359.91, SD = 201.59, Min. = 85, Max. = 839).
Table 4 below shows the descriptive statistics (mean, standard deviations) for participants’ (N = 61) pausing and revision behaviors. The results showed that for pausing times participants paused more within words and between words in Spanish than in English. However, on average, participants paused slightly more between sentences in English than in Spanish. With respect to revisions, participants’ average r-burst times were higher in Spanish than in English. Likewise, with the number of word revisions, the average was slightly higher in Spanish than in English. Figures 2, 3, and 4 provide visual representations of the descriptive results.
Table 4. Descriptive statistics for pauses and revisions by language
Note: (s) = seconds;
a = mean;
b = standard deviation;
c = minimum score;
d = maximum score.
Figure 2. Pause times (s) by language and location.
Figure 3. R-burst times (s) by languages.
Figure 4. Number of word revisions by languages.
Research question 1: Pausing behaviors
RQ#1 asked whether Spanish-English heritage writers differed in their pausing behavior in Spanish and English as well as HL writers’ underlying cognitive processes when pausing during the completion of writing tasks in Spanish and English. The pausing times were further analyzed by textual location—within word, between words, between sentences—to determine whether interactions emerged between language and location. The findings first revealed that writing in Spanish or English did not significantly contribute to HL writers’ overall pausing behaviors (b = -0.04, SE = 0.23, t = -0.19, p = .85, 95% CI [-0.50, 0.41]). However, the results did show that location of pauses within words (b = -1.74, SE = 0.23, t = -7.50, p < .01, 95% CI [-2.19, -1.28]) and between words (b = -1.16, SE = 0.23, t = -5.02, p < .01, 95% CI [-1.62, -0.70]) significantly contributed to smaller pausing times in comparison with pauses between sentences. Importantly, these results are independent of language. Put differently, these findings indicate that the location of pauses within and between words made a significant contribution to smaller pausing times across languages.
In examining participants’ underlying cognitive processes for pausing during their writing tasks, Tables 5 and 6 summarize participants’ TA comments by language and pause location. According to Table 5, most of the TA episodes during the Spanish writing task were related to the translation writing process (n = 132/166) and these occurred most frequently between words (57%). However, all TA comments between sentences were related to generating ideas and the organization of the essay. More specific to coding categories, participants’ pauses between words were mostly related to lexical retrieval issues (50%). In other words, participants paused most frequently between words to think about how to say words in Spanish (e.g., “how do you say provide?” in Table 2). Following lexical retrieval issues, participants often paused within words to think about the spelling of words (19%). The next largest percentage of TA comments was between sentences to generate ideas for the essay (12%). Other categories accounted for less than 10% of the TA episodes, which included organization, lexical choice, accent placement, language form, and punctuation. In contrast, most TA episodes during the English writing task, according to Table 6, were related to planning writing processes (n = 52/66). Like the Spanish writing task, most of the TA episodes for the English writing task occurred during participants’ pauses between words (50%). However, most of the pauses between words were due to generating ideas (34%), which means that participants paused more frequently to think about the content of their English essays. The next largest categories occurred between sentences and were related to generating ideas (22%) and the organization (19%) of the essay. The TA episodes related to participants’ translation writing process were less than 10%, which included lexical choice, spelling, and punctuation. No TA comments emerged in relation to lexical retrieval issues.
Table 5. Results of think-aloud (TA) comments for Spanish writing pausing behaviors by location and category
Table 6. Results of think-aloud (TA) comments for English writing pausing behaviors by location and category
Research question 2: Revision behaviors
RQ#2 was concerned with whether Spanish-English HL writers differed in their revision behaviors in Spanish and English as well as HL writers’ underlying cognitive processes when making revisions during the completion of Spanish and English writing tasks. To address this research question, participants’ r-burst times and number of word revisions were considered in separate statistical analyses. With respect to potential differences in revision behaviors between languages, the findings revealed that writing in Spanish or English did not make a significant contribution to participants’ r-burst times (b = 1.44, SE = 0.91, t = 1.58, p = .12, 95% CI [-0.37, 3.24]) or number of word revisions (b = 2.39, SE = 32.21, t = 0.07, p = .94, 95% CI [-61.74, 66.53]).
To address participants’ underlying cognitive processes for revision episodes during their writing tasks, Tables 7 and 8 provide a summary of participants’ TA comments by revision location. For the Spanish writing task, Table 7 shows that most revision comments were related to participants’ editing of linguistic encoding issues and these occurred mostly within words (51%). Regarding the categories that emerged during revisions, most revisions were due to spelling issues (26%). The next largest category (22%) was related to lexical choice issues followed by language forms (20%). Then, comments related to generating ideas accounted for 19% of the TA episodes, which emerged in pauses between words and between sentences. Smaller categories included organization (7%) and accent placement (4%). Table 8 summarizes the TA episodes that emerged during participants’ completion of the English writing task. Unlike for the Spanish writing task, the largest percentage of TA comments in the English writing task occurred between words (46%). Most of the TA comments were related to generating ideas during revisions between words and between sentences (39%). The next two largest categories of TA episodes were spelling issues (21%) and lexical choice (20%). TA comments related to text organization followed, with 10% of revisions between words and between sentences. Finally, two small categories emerged with respect to revisions of language form (3%) and punctuation (1%).
Table 7. Summary of think-aloud (TA) comments for Spanish writing revision behaviors by category and location
Table 8. Summary of think-aloud (TA) comments for English writing revision behaviors by location and category
Research question 3: WM and language dominance
RQ#3 asked whether participants’ individual differences along WM and language dominance contributed to their pausing and revision behaviors when completing writing tasks in the heritage and societal languages. First, participants’ average WM score was 79.80 (SD = 16.97, Min. = 35, Max. = 110), and the average score for language dominance in the current sample was 35.22 (SD = 44.96, Min. = -99.43, Max. = 198.50). The findings revealed that individual differences in WM significantly contributed to pausing times within words (b = 0.02, SE = 0.01, t = 2.28, p = .02, 95% CI [0.003, 0.041]), whereas a marginal significant contribution was found for pausing between words (b = 0.02, SE = 0.01, t = 1.79, p = .07, 95% CI [-0.001, 0.036]). Importantly, these findings are independent of language, as WM did not significantly contribute to pausing times in either Spanish or English texts (b = 0.00, SE = 0.01, t = -0.22, p = .83, 95% CI [-0.02, 0.02]). Furthermore, individual differences in WM did not significantly contribute to r-burst times (b = 0.00, SE = 0.04, t = 0.00, p = 1.00, 95% CI [-0.08, 0.08] or number of word revisions (b = -1.21, SE = 1.82, t = -0.66, p = .51, 95% CI [-6.05, 2.06]). With respect to language dominance, language dominance did not significantly contribute to pause times (b = 0.00, SE = 0.00, t = 0.37, p = .71, 95% CI [-0.006, 0.009]). Likewise, language dominance did not significantly contribute to r-burst times (b = 0.00, SE = 0.01, t = -0.36, p = .72, 95% CI [-0.03, 0.02]. As for word revisions, language dominance marginally contributed to overall number of word revisions (b = -1.34, SE = 0.67, t = -2.01, p = .05, 95% CI [-2.67, -0.01]) independent of language (b = -0.23, SE = 0.72, t = -0.32, p = .75, 95% CI [-1.66, 1.20]).
In summary, the main results revealed that HL writers’ pausing and revision behaviors did not significantly differ when writing in Spanish or English. Pause location within words and between words significantly contributed to smaller pausing times vis-à-vis pauses between sentences, and this occurred independent of language. However, the qualitative data (i.e., TAs) showed that the focus of HL writers’ pausing and revision behaviors between Spanish and English were different. For pausing behaviors, participants were more focused on linguistic encoding issues when writing in Spanish, especially in retrieving lexical items. In contrast, participants focused more on generating and organizing the content of their essays in English. A similar pattern emerged in participants’ revision behaviors. For the Spanish writing task, participants mostly edited words for spelling, lexical items, and language forms. For the English writing task, however, the participants were more focused on editing content though spelling and lexical choice issues were also addressed at rather high percentages. With respect to individual differences, participants with higher WM capacity significantly had slightly longer pause times within words independent of Spanish or English writing tasks. Although no significant results emerged for the role of language dominance and participants’ writing behaviors, a marginally significant trend (p = 0.05) indicated that English-dominant participants made overall fewer word revisions.
Discussion
The first goal of the current study was to expand the field’s understanding of HL bilinguals’ writing processes when completing writing tasks in the heritage and societal languages. The first major finding, according to the keystroke-logging data, was that HL writers’ time spent on pausing and revisions as well as their number of word revisions did not significantly differ across the Spanish (the HL) and English (the societal language) writing tasks. This indicates that HL writers engaged with writing processes to a similar extent across both languages, which aligns with findings from Spanish-English L2 writers (Manchón & Roca de Larios, Reference Manchón and Roca de Larios2007; Roca de Larios et al., Reference de Larios, Marín and Murphy2001). A significant finding that did emerge from the pausing behavior results, however, is that pausing times were significantly smaller within and between words independent of language. Put differently, in general, participants significantly paused for a longer time when the pauses occurred between sentences. According to the TA data, participants were only engaged with planning processes between sentences across both languages, which means that they directed their attentional resources to generating and organizing the content of their argumentative essays in Spanish and English according to the writing task goal. This supports claims that accessing conceptual information during the writing process between sentences takes a longer time, as previously reported for L1 and L2 writers (Révész et al., Reference Révész, Michel and Lee2019; Schilperoord, Reference Schilperoord, Rijlaarsdam, van den Bergh and Couzijn1996).
However, these results do differ from Mikulski and Elola’s (Reference Mikulski and Elola2011) finding that their Spanish-English HL writers paused for longer times between sentences when writing in Spanish, suggesting that they spent more time planning in the HL. One potential explanation for these divergent results is that Mikulski and Elola’s participants were enrolled in a foreign language intermediate Spanish language course that focused on grammar but the participants of the current study were recruited from a tailored writing course for HL speakers. Therefore, the participants in this study may have arrived at the experiment with more writing experience in Spanish given their exposure to explicit instruction on writing conventions and strategies. Furthermore, the current study allowed participants 3 min of brainstorming time before completing the writing task, which may have granted an opportunity for participants to generate and organize some ideas before writing, especially given the effects pretask planning can have on task performance (e.g., Ellis, Reference Ellis2005). Future research ought to shed light on these issues to disentangle factors of writing experience and other variables (e.g., pretask planning) that can affect HL writers’ task performance.
Despite the keystroke-data results indicating no significant differences between English and Spanish writing processes, the TA data showed that the underlying nature of those writing processes was not necessarily the same. The findings indicate that, for the Spanish writing task, participants were pausing overall more frequently to address linguistic encoding issues, whereas during the English writing task, participants focused more on within-task planning issues such as organizing the content of their essays. According to the TA data, the major reason for their pausing, especially between words, was due to lexical retrieval issues (50% of the TA comments). In other words, HL writers in this study seemingly spent most of their cognitive effort in retrieving Spanish lexical items, an issue that did not emerge in any of the TA comments for the English writing task. With respect to revision behaviors, a similar pattern emerged in which participants were monitoring more often surface language constructions and, in particular, spelling and language forms (e.g., verb conjugations) issues with the Spanish writing task (see Elola and Mikulski, Reference Elola and Mikulski2013, for similar results). These findings also confirm previous work on HL writers’ reports on their writing in Spanish, which have included challenges with vocabulary, spelling, verb conjugations, and accent placement (Schwartz, Reference Schwartz, Ortiz López and Lacorte2005; Torres et al., Reference Torres, Arrastia-Chisholm and Tackett2020).
These findings may be interpreted as implying that HL writers shifted the focus of their cognitive processes between planning and translating as a result of whether they were completing the Spanish or English writing task. That is, HL writers adapted their writing processes according to the linguistic demands of the two writing tasks. This shift was due primarily to participants’ difficulty in accessing lexical items in the HL, which supports prior arguments that HL bilinguals’ knowledge of more specialized and abstract vocabulary tends to be limiting (e.g., Fairclough & Garza, Reference Fairclough, Garza and Potowski2018; Montrul, Reference Montrul2016a; Zyzik, Reference Zyzik, Pascual y Cabo and Torres2022). Furthermore, Zyzik (Reference Zyzik, Pascual y Cabo and Torres2022) recently argued that these lexical limitations are particularly apparent in free-production tasks like the one in the current study. Such limitations can also be associated, in part, with differences in proficiency levels, as the participants in the current study self-rated their writing proficiency higher in English. Arguably, then, differences in proficiency levels can determine HL writers’ attentional resources to manage writing demands ranging from activating prior knowledge and writing strategies to generating and organizing content to accessing pertinent linguistic information for the lexical and morphosyntactic encoding of concepts to meet writing task goals. This suggests that HL writers with lower writing proficiency in the HL will invest more attentional resources to linguistic encoding at the cost of generating and organizing content when writing in the HL.
The second goal of this study was to provide additional support for the underlying cognitive processes associated with writing, as proposed by Kellogg’s (Reference Kellogg, Levy and Ransdell1996) cognitive writing model. The findings demonstrated that HL writers, who are early bilinguals, engaged in the same writing processes that have been documented in L1 and L2 writing (Kellogg et al., Reference Kellogg, Whiteford, Turner, Cahill and Mertens2013; Révész et al., Reference Révész, Michel and Lee2019). Additionally, the current study showed that WM, as measured by an advanced operation-span task, significantly contributed to pauses within words for both tasks. This finding supports Kellogg’s (Reference Kellogg, Levy and Ransdell1996) claim that writing places demands on executive control WM during the translating phase. However, contrary to the predictions of the model, WM was not a contributing factor to HL writers’ text revision. This latter finding is in line with the empirical challenge the model faces with respect to WM and editing a text among L1 writing (see Kellogg et al., Reference Kellogg, Whiteford, Turner, Cahill and Mertens2013). Along the same lines, Révész et al. (Reference Révész, Michel and Lee2017) did not find significant correlations between their WM measures and revision behavior among their adult L2 writers. Together, these findings indicate that editing a text does not place heavy cognitive demands on writers’ WM resources. This can be due to writers’ more extensive reliance on long-term memory to edit a previously written text. That is, for writers to notice a mismatch between previously written text and their knowledge of conceptual and linguistic information, such information needs to be readily accessible in long-term memory. Therefore, this reliance on long-term memory resources, which are not stored in WM, for retrieval of information during the monitoring process can ease the burden on WM. As McCutchen (Reference McCutchen2000) pointed out with children L1 writers, skilled writers’ writing knowledge allows them to capitalize on their long-term memory resources moving beyond the constraints of WM.
The third goal of this study was to examine whether HL writers’ individual differences along WM and language dominance contributed to the variation in writing behaviors. The findings revealed that HL writers with higher executive WM had significantly longer pause times within words. Recall that, according to the TA data, participants’ comments on pauses within words were all related to translating issues such as spelling, diacritic placement, and language forms. Importantly, these findings were independent of the writing task, which suggests that these HL writers accessed WM resources for pauses within words when writing in Spanish and English. This may be because participants were only engaged with translating processes during pauses within words for both writing tasks. These findings on WM support the claim that accessing linguistic resources for encoding purposes during writing can place the most cognitive demands on L1 writers (e.g., McCutchen, Reference McCutchen1996; Kellogg et al., Reference Kellogg, Whiteford, Turner, Cahill and Mertens2013). In contrast, the findings slightly differ from Révész et al. (Reference Révész, Michel and Lee2017) and Vallejos (Reference Vallejos2020) given that WM correlated with pauses between sentences and between paragraphs among their L2 writers during which participants addressed lexical retrieval and grammatical issues. However, no conclusions can be drawn given the vast differences among these studies with respect to participant background, WM measures, and writing task type. More research is needed to shed light on variables that can potentially modulate the effects of WM on writing processes.
The findings of the current study imply that for orthographic and morphosyntactic encoding episodes, HL writers with higher executive control WM were able to sustain the temporary manipulation and storage of linguistic information for a slightly longer time to figure out these linguistic encoding issues. Based on the phonological/executive WM framework (Wen, Reference Wen2016), we begin with the premise that HL writers activated their linguistic knowledge in long-term memory with such activation, in turn, overlapping with executive WM to address orthographic and morphosyntactic forms. These results support Wen’s (Reference Wen2016) framework that linguistic information overlaps with executive control WM and long-term memory given that participants in this study completed an Ospan WM test designed to tap into executive control WM. According to the TA data, HL writers were engaging in explicit retrieval processes of these forms by recruiting their linguistic resources from long-term memory. With respect to morphosyntactic knowledge, these findings are in line with Zalbidea’s (Reference Zalbidea2017) results on the relationship between WM and morphosyntactic constructions in L2 Spanish writing. However, unlike L2 writers whose knowledge tends to be more explicit (Wen, Reference Wen2016), the nature of HL writers’ morphosyntactic knowledge may be stored, in many cases, as implicit memory events. This leads HL writers to test what morphosyntactic forms sounded right to them in the HL, a finding also observed in Schwartz’s (Reference Schwartz, Roca and Colombi2003) case studies. Consequently, their writing outcomes will depend in large part on the nature of their linguistic knowledge having undergone restructuring patterns associated with the HL bilingual experience such as morphological agreement (e.g., Polinsky & Scontras, Reference Polinsky and Scontras2020). In contrast, executive control WM did not contribute to pauses between words that were primarily related to lexical access during the Spanish writing task, aligning with previous studies that found no significant relationship between WM and lexical performance in adult L2 writing (Lu, Reference Lu, Wen, Mota and McNeill2015; Yi & Luo, Reference Yi and Luo2013). This may indicate that when writers, including L2 bilinguals, have no lexical knowledge to activate from long-term memory into the shared executive WM space, executive WM resources are unnecessary due to a lack of stored linguistic knowledge to address lexical encoding episodes. However, more empirical studies are needed to confirm this trend.
Last, concerning language dominance, scores indicated that HL participants in the current study ranged from Spanish to English dominant. This highlights the importance of providing independent evidence for language dominance rather than assuming that all HL participants are dominant in English, especially in linguistically diverse contexts like Southern California. Importantly, though, language dominance did not significantly contribute to HL bilinguals’ writing behaviors. However, a marginally significant trend did show that English-dominant HL writers made fewer word revisions. Future research needs to confirm this trend to understand and explain this behavior. As previously noted, bilingualism scholars have argued for theoretical distinctions between language dominance and language proficiency (Montrul, Reference Montrul, Silva-Corvalán and Treffers-Daller2016b). These results support such claims and imply that HL writers’ easier access to Spanish or English by dimensions of language use, language proficiency, language attitudes, and others did not influence their writing behaviors and that writing proficiency may have mattered more. This claim is supported by examining the data from Spanish-dominant participants, who, on average, also self-rated their writing proficiency in Spanish lower than in English (see Table 1). That is, despite being Spanish dominant, many of these participants still considered their Spanish writing skills as being weaker. These findings differ from other studies in which language dominance predicted, for example, HL bilinguals’ performance on linguistic tasks examining phonetic features in the HL (Amengual, Reference Amengual2016). However, the current study indicates that language dominance as a global multidimensional measure does not explain variability in the domain of HL writing. This can be because the Bilingual Language Profile questionnaire is biased toward spoken communication, given that language dominance is partly associated with how active one of the bilingual’s languages is for engaging in speaking (e.g., Harris et al., Reference Harris, Gleason, Aycicegi and Pavlenko2006).
Pedagogical implications
The current study points to the relevance of expanding HL writers’ lexical knowledge to tackle argumentative writing tasks in the HL that require specialized vocabulary with respect to societal issues (e.g., education). This is due to our HL writers’ large percentage of pauses between words to access lexical items to encode their ideas. Following task-based language teaching methodology, prior to HL students’ engagement in a writing task of such topics, scaffolding activities during a pretask phase are strongly recommended to allow HL writers to activate their linguistic resources in both languages to address any gaps in their lexical knowledge. In other words, instructors ought to allow HL students the flexibility to brainstorm content in any of their languages or combination thereof during a first writing draft and, then, to access dictionary and thesaurus resources to help them translate and express such content in a relevant written register in the HL. Additionally, given our finding that only HL writers with higher WM spent more time addressing orthographic and morphosyntactic encoding episodes during pauses within words, to level the playing field for all HL students, upon completion of a first writing draft, instructors can guide HL writers’ attention to morphosyntactic and orthographic issues during a posttask phase at an individual level or in collaboration with peers. In line with Torres and Baralt (Reference Torres, Baralt, Pascual y Cabo and Torres2022), it is more advisable to address language forms during a posttask phase with HL students due to their highly variable linguistic experience, which typically leads to varying degrees of knowledge of linguistic constructions in the HL.
Limitations and conclusion
Although the current study was an exploration of WM, the design could have benefited from other measures that specifically investigated phonological and visuospatial WM components to gain a more nuanced understanding of these different measures and writing behaviors. Also of relevance is that the moderating effects of WM may have been compromised given that the writing tasks were not timed. Future studies ought to also consider the effects of WM on timed writing tasks with a HL population to determine whether the findings of this study replicate. Finally, including an objective measure of writing proficiency could have provided a more robust analysis of the contribution of proficiency to the results than language dominance. Despite these limitations, the current study advances our understanding of HL bilinguals’ writing processes by integrating keystroke-logging and TA data as well as the contributions of WM and language dominance. The findings revealed that although HL writers do not show significant differences in their pauses and revisions when completing writing tasks in the heritage (Spanish) and societal (English) languages, the nature of their writing processes fluctuated between writing tasks. Writing in the HL required more attentional resources devoted to linguistic encoding issues, whereas participants focused more on generating and organizing the content of their essays in English. HL writers with higher WM spent significantly more time within words addressing orthographic and morphosyntactic encoding episodes in both Spanish and English. Finally, despite the range of language dominance scores in the current sample, language dominance did not play a critical role in participants’ writing pausing and revision behaviors, suggesting that future studies ought to examine the role of proficiency in writing processes more closely.
Supplementary materials
To view supplementary material for this article, please visit http://doi.org/10.1017/S027226312300013X.
Acknowledgments
I would like to thank the guest editors, Dr. Rosa Manchón and Dr. Cristina Sanz for their guidance as well as the anonymous reviewers for their feedback on an earlier version of the manuscript. Further, I would like to thank the following research assistants for their assistance with data collection: Kenneth Díaz, Diana Esparza, Justin Figueroa, Emily Guilfoyle, Jasmin Hernández, Alexandra Román, Nicholas Sullier, and Nicole Vargas Fuentes. All remaining errors in the article are my own.
Competing interests
The author declares none.
