The effects of feedback timing on L2 development in written SCMC

Abstract

Numerous studies have delved into the effects of interactional corrective feedback provided in the oral or written mode in the CALL environment (e.g.video-conferencing or text-based chat). Although previous research shows that several factors influence its effectiveness, a research area that merits more attention is the role of feedback timing operationalized as immediate or delayed intervention. The current study explores the role of timing when recasts, a corrective feedback technique, are provided in an unfocused, incidental manner addressing semantic and morphosyntactic features as they arise during written synchronous computer-mediated communication (SCMC). Fifty-eight young learners of English (M = 12.75 years old, SD = 1.2) were assigned to one of two experimental conditions that differed as to whether they received immediate recasts while performing a communicative task or delayed recasts after completing a task. The results show that delayed recasts led to greater L2 gains than immediate ones in the immediate postinteraction test especially when addressing semantic features; however, no difference was found between the two conditions two weeks after the intervention in the delayed postinteraction test.

Keywords:

• Written synchronous computer-mediated communication (SCMC)
• text-based chat
• corrective feedback timing
• recasts
• linguistic focus

Introduction

Following cognitive-interactionist perspectives, the focus on form approach posits that drawing L2 learners’ attention to linguistic features while engaging in meaningful interaction assists interlanguage development (Long, 1996). An interactionist approach to corrective feedback, that is a focus on form technique, contends that feedback leads to interlanguage development when it enables learners to notice form-meaning mappings and modify their erroneous utterances (Gass & Mackey, 2015). From a socio-cultural perspective, interactional feedback is viewed as assistance from a more advanced interlocutor (i.e. scaffolding) that gradually enables learners to become independent and achieve self-regulation (Nassaji & Swain, 2000).

Although corrective feedback has been a subject of debate in the past (see Truscott, 1996 vs. Ferris, 1999), a plethora of studies have demonstrated its positive effects and meta-analyses support its use both in the face-to-face (FTF) mode and in the CALL environment (e.g. Li, 2010; Lyster & Saito, 2010; Russell & Spada, 2006; Ziegler, 2016). However, corrective feedback is not a monolithic construct as it is provided in various ways that involve a more implicit or explicit focus on form (e.g. recasts, prompts, explicit correction, metalinguistic feedback) (see Lyster & Ranta, 1997 seminal study). The focus of this paper is to explore feedback in the form of recasts supplied during written synchronous computer-mediated communication (SCMC) (i.e. text-based chat). Recasts are defined as reformulations of learners’ erroneous output without altering the meaning learners intend to convey (see example 1).

(1) Recast.

Researcher: Where does he place the baby?

Learner: He place the baby in a καρότσι (pram).

Researcher: He places the baby in a pram.

Previous studies have demonstrated that several factors influence the effectiveness of written SCMC recasts, including working memory and contingency (Lai et al., 2008), task factors (Baralt, 2013) and type of target error (Smith, 2012). SLA researchers have also set out to elucidate the role of feedback timing operationalized as immediate and delayed correction (see Quinn, 2021 for a review). Immediate corrective feedback is supplied directly after the erroneous production while engaging in interaction; whereas delayed corrective feedback is provided at a later time (e.g. after students have completed a task). When focused recasts are supplied targeting predetermined linguistic features, previous CALL studies show that there is no significant difference between immediate and delayed feedback conditions (e.g. Henderson, 2019; Shintani & Aubrey, 2016) or that immediate feedback is more beneficial (e.g. Arroyo & Yilmaz, 2018). However, the role of timing when unfocused recasts are provided remains an underresearched area. From an information processing perspective (see Skehan, 1998), it is important to distinguish focused from unfocused feedback when examining the effects of feedback timing. Considering that unfocused recasts require processing of a variety of linguistic features, providing them after learners have completed a task may enable deeper processing and more attention allocated to the target language, especially when learners’ level of proficiency is relatively low. To test this hypothesis, the current study explores the role of feedback timing operationalized as immediate and delayed SCMC recasts addressing a variety of linguistic features (i.e. unfocused recasts). The study also examines the effects of recasts for different linguistic categories and different types of errors.

Background literature

Corrective feedback in the SCMC mode

In the CALL and MALL environment, the role of corrective feedback has been explored in several platforms and modes. Previous studies have examined the effects of oral feedback supplied during interaction taking place via mobile applications or video-conferencing and they have demonstrated positive effects on several aspects of language development, including speaking ability (Xu & Peng, 2017), noticing and learning of grammatical features (Monteiro, 2014; Rassaei, 2022), pronunciation (Dai & Wu, 2021), learning of Mandarin tones (Bryfonski & Ma, 2020) and opportunities for peer interaction, increased motivation, engagement and active learning (e.g. Chun, 1994; Dao et al., 2021; Warschauer, 1996).

Numerous studies have also examined the role of corrective feedback provided during written SCMC - text-based chat. Researchers have suggested that written SCMC feedback might be beneficial for L2 learners because unlike oral feedback, it remains permanently on the screen; hence, learners have more time to process the feedback, monitor their output and engage in online planning (Lin et al., 2013; Smith, 2012), that is planning a message while performing a task (Yuan & Ellis, 2003). In line with these arguments, Sachs and Suh (2007) showed that both textually enhanced and textually unenhanced recasts on reported speech (i.e. backshifting of verbs from past to past perfect) were successful in facilitating interlanguage development. L2 gains have also been demonstrated when comparing recasts delivered in the SCMC and FTF mode in Yilmaz (2012) and in Baralt (2013); however, the latter explains that task complexity moderated the effectiveness of recasts.

Although previous studies have provided evidence that written SCMC recasts facilitate L2 development, their linguistic focus is a factor that may influence their benefits. For example, Smith (2012) found that semantic and syntactic recasts were noticed significantly more than morphological recasts. Other studies also show that written SCMC recasts are less effective when addressing low saliency features (i.e. the present third person singular in Kourtali, 2022; the past tense in Loewen & Erlam, 2006; the English zero article followed by abstract uncountable nouns in Sauro, 2009; Turkish locative case morphemes in Yilmaz, 2012). A potential explanation for this is that when learners receive immediate recasts while engaging in text-based chat, they may devote more attention to meaning at the expense of form; and hence they allocate less attention to low saliency and communicatively redundant linguistic features. For example, Kourtali (2022) found that many learners responded to interrogative SCMC recasts on the present third personal singular morpheme (e.g. ‘He works?’) by providing additional information related to the pictures of the task without correcting the target feature (see example 2).

(2) Recast from Kourtali (2022).

Researcher: Alex?

Learner: Alex on Saturdays 9-11 is at a home and work

Researcher: He works? (recast)

Learner: acountent (accountant) (additional information)

When analysing learners’ responses, Kourtali (2022) argued that

the interrogative mode of the SCMC recasts, in combination with the learners’ lack of previous experience in receiving CF [corrective feedback] on morphology in the SCMC environment, might have encouraged that group to perceive recasts as clarification requests requiring information about the meaning of their output. (p. 13)

In other words, the corrective intention of recasts remained unattended. Considering that recasts are more effective in facilitating noticing of linguistic elements when their intention to promote a focus on form is clear to learners (Nicholas et al., 2001), exploring whether delayed and declarative SCMC recasts are less ambiguous, especially when addressing low saliency features, merits more attention. Research on feedback timing is discussed in greater detail in the following section.

Feedback timing

In this section, we will present theoretical arguments that could be used in support of immediate and delayed feedback. We will also synthesize research findings that examined the role of feedback timing in the CALL environment.

Theoretical background

Potential benefits of immediate feedback could be explained by cognitive-interactionist perspectives in SLA. According to Long’s (1996) focus on form technique, the provision of recasts immediately after learners’ non-target-like utterances might encourage them to make a cognitive comparison between their erroneous output and the target-like feature modelled by the recast (Doughty, 2001; Long, 2007). In particular, Long (2007) explains that:

…recasts convey needed information about the target language in context, when interlocutors share a joint attentional focus, and when the learner already has prior comprehension of at least part of the message, thereby facilitating form-function mapping. (p. 77)

Long’s work focuses on oral-FTF interaction and his ideas have been mainly used in traditional classrooms; nevertheless, focus on form techniques are relevant to written SCMC. Although feedback delivered in this mode is written, there are similar interactional features in the discourse used in oral communication and instant text messaging. For example, both modes involve short turns, informality, real-time, direct and interpersonal communication (Arroyo & Yilmaz, 2018; González-Lloret, 2014; Kern, 1995; Lin et al., 2013). Considering these similarities, the provision of immediate feedback when learners perform written SCMC tasks may have similar positive effects as in oral interaction; for example, this may enable learners to engage in cognitive comparison between their erroneous output and the juxtaposed target-like model.

Another theory that may explain the potential benefits of immediate feedback is the Transfer Appropriate Processing (TAP). According to TAP, learners better transfer and remember what they have learned ‘if the cognitive processes that are active during learning are similar to those that are active during retrieval’ (Lightbown, 2007, p. 27). Following this argument, if learners attend to SCMC feedback on linguistic features as they arise during communicative tasks, they might be better able to deploy their gains when engaging in similar tasks. Drawing on the Skill Acquisition Theory (Anderson, 1993; DeKeyser, 2007), communicative tasks provide evidence for learners’ procedural knowledge; hence, receiving immediate recasts during meaningful interaction in the SCMC mode might be highly beneficial for learners’ procedural knowledge in the process of automatization, especially when communication takes place in the same mode.

Regarding delayed feedback, a theory that could be used to account for its possible benefits in L2 learning is Skehan’s (2009, 2014) Limited Capacity Model. Skehan contends that limited attentional resources restrict learners from focusing on meaning and form simultaneously. In particular, Skehan argues that learners devote less attention to linguistic encoding when engaging in tasks that impose great cognitive demands. This could arguably raise some questions about the benefits of immediate feedback, especially when several linguistic constructions are targeted (i.e. unfocused feedback) and when learners’ level of proficiency is relatively low. From an information processing perspective, it could be suggested that providing unfocused feedback after learners have completed a task may free up attentional resources; and as a result, learners might be able to allocate greater attention to the linguistic features addressed by the feedback. Nevertheless, it should be highlighted that Skehan’s ideas have been mainly explored when learners perform oral tasks. As compared to oral interaction, text-based chat poses less time pressure but it does require real-time communication. More empirical research is needed to explore how attentional limitations may influence interaction in the SCMC environment and how different SCMC conditions (e.g. different tasks or feedback timing) may affect learners’ performance.

Although different theoretical arguments could be used either in favour of immediate or delayed feedback, empirical research plays a crucial role in elucidating how feedback timing influences learning outcomes. A synthesis of the findings is presented in the following section with a focus on CALL studies.

Previous studies

In the FTF environment, previous research has provided support for the provision of immediate, focused feedback as compared to delayed feedback (e.g. Fu & Li, 2022; Li et al., 2016). In the CALL environment, researchers have also explored the role of feedback timing when learners engage in oral interaction during video conferencing (e.g. Canals et al., 2021) or written interaction that involves Google forms (e.g. Shintani, 2015; Shintani & Aubrey, 2016) and text-based chat (e.g. Arroyo & Yilmaz, 2018; Henderson, 2019). Overall, these studies demonstrate mixed findings.

Canals et al. (2021) examined the effects of feedback timing when oral explicit correction was provided during or after video conferencing. The researchers found that on a grammaticality judgment test (GJT), there was no difference between immediate and delayed feedback conditions and the control group. On the oral production task, both immediate and delayed feedback led to greater gains than the control group but no difference was found between the two experimental conditions. Shintani and Aubrey (2016) also explored immediate feedback (i.e. simultaneous editing using Google forms) and delayed feedback delivered after learners completed the task. Their study demonstrated that unlike the control group, both feedback conditions yielded L2 gains but there was no significant difference between the immediate and delayed feedback. However, when comparing the two experimental conditions to the control group, only the immediate feedback group outperformed the ‘no feedback’ condition. In a follow-up case study, Shintani (2015) used stimulated recall methodology to examine how two learners perceived the feedback supplied during and after the task. The study showed evidence of noticing and metalinguistic understanding in both conditions; however, more successful self-corrections were elicited when feedback was delivered during the task.

In the text-based chat environment, Arroyo and Yilmaz (2018) explored the effects of immediate and delayed feedback operationalized as partial reformulations. On an oral production test, immediate feedback was more beneficial than delayed feedback and the ‘no feedback’ condition (control group) on both posttests. On the GJT, both the immediate and delayed feedback groups outperformed the control group on both posttests but no differences were found between the two feedback conditions. In a similar vein, Henderson (2019) delved into the effects of immediate and delayed feedback delivered in the form of recasts preceded by error repetition on vocabulary development of L2 learners of Spanish. The study demonstrated that there were no significant differences between the two feedback conditions; however, both experimental groups achieved significantly greater development than the control group.

Overall, CALL studies that have investigated feedback timing show that either immediate feedback is more beneficial or there is no difference between feedback conditions. Nevertheless, it should be highlighted that these studies have explored the effects of feedback timing when focused feedback is supplied on predetermined language features (e.g. Spanish noun-adjective gender agreement in Arroyo & Yilmaz, 2018; -ing/-ed participial adjectives in Canals et al., 2021; vocabulary development in Henderson, 2019; hypothetical conditional in Shintani & Aubrey, 2016). To the best of our knowledge, the role of timing when feedback is provided in an incidental, unfocused manner addressing a variety of language features is an area that has not been explored in the SCMC environment. In the following section, we explain the difference between focused and unfocused feedback and why we believe that this distinction is crucial when we examine the effects of feedback timing.

Focused and unfocused feedback

When learners receive focused feedback, teachers or researchers target specific predetermined linguistic features. Conversely, unfocused feedback is supplied to address a variety of errors and it lacks specific linguistic focus. The effects of focused feedback have been widely explored in experimental research using a pretest-posttest(s) design (see Beuningen, 2021; Nassaji, 2016 for a review). Considering learners’ limited attentional capacities, it has been argued that focused feedback is highly effective in drawing learners’ attention to specific linguistic constructions (Ellis et al., 2008). Nevertheless, more research on the effects of unfocused feedback is needed. First, research on unfocused feedback may have high ecological validity as findings might be more relevant to the way teachers address learners’ errors in classrooms (Nassaji, 2016). Unfocused feedback might also be more effective in facilitating learning of a wide range of linguistic constructions that will enable students to become successful communicators during more authentic interaction. As Ellis and Sheen (2006) explain, unfocused feedback that is ‘directed at whatever errors learners happen to make might be of greater practical value’ (p. 597). Although a few studies have examined the role of unfocused feedback during oral interaction (e.g. Nassaji, 2017), research on the role of unfocused feedback in the CALL environment is still limited, especially when the effects of feedback timing are explored. Examining feedback timing when unfocused feedback is provided is crucial because targeting a variety of linguistic features may impose great cognitive demands on learners’ working memory and attentional resources. In such conditions, delivering delayed feedback may enable deeper processing of linguistic features, especially when the recipients of feedback are low intermediate EFL learners. When looking at unfocused feedback, another issue that merits more attention is whether there are differences between immediate and delayed feedback for broad linguistic categories (e.g. semantic and morphosyntactic recasts) and different types of errors within each category.

The present study

In order to fill the gaps described above, in the current study we used unfocused recasts delivered in the written SCMC mode and we examined the effects of feedback timing operationalized as immediate versus delayed correction. We also looked at the effects of the two conditions on development of different linguistic categories and types of errors. Finally, our participants were low intermediate EFL learners of English; an underresearched population. It should be noted that previous studies that examined feedback timing recruited adult learners. Considering the role of age effects in SLA (see García Mayo & García Lecumberri, 2003), feedback timing may have different effects on younger, low intermediate EFL learners. To explore this area further, we formulated the following research questions:

1. Do low intermediate EFL young learners of English demonstrate development after receiving unfocused, written SCMC recasts? If yes, does feedback timing influence their development?

2. Do low intermediate EFL young learners of English benefit from unfocused, written SCMC recasts with different linguistic focus? If yes, does feedback timing influence their development?

Method

Participants

The first author recruited participants by sharing an information sheet with the owners of several language schools in Greece and the parents of L2 learners. A total of 58 low intermediate learners of English were recruited for the study (33 females and 25 males). They had never lived in an English-speaking country prior to the study and they were L1 users of Greek (n = 51) or they were Greek-Albanian (n = 4), Greek-Romanian (n = 2) or Greek-Russian (n = 1) bilinguals born in Greece. Their tutor had enrolled them in low intermediate classes (for more information see the proficiency guidelines suggested by ACTFL); however, we also analysed their accuracy levels at the preinteraction stage in order to ensure the two groups were similar in their linguistic performance (see the ‘preliminary analysis’ section). Participants’ ages ranged from 11 to 15 years (M = 12.7, SD = 1.19 in the immediate feedback group and M = 12.81, SD = 1.22 in the delayed feedback group). Participants’ length of English study prior to the experiment ranged from 3 to 10 years (M = 5.9, SD = 1.44 in the immediate feedback group and M = 6, SD = 1.46 in the delayed feedback group). A series of independent-samples t tests targeting the variables of age and length of previous English study confirmed that the two groups were comparable: age, t = .326, p = .746, d = .05; English study, t = .271, p = .788, d = .04. The majority of participants also reported that they were learning French, German, Italian or Russian (n = 54) as a foreign language.

All the participants had experience with face-to-face English classes in state schools and private language schools, and they had some experience with language learning in the online mode. The researcher asked the participants and their parents how long they had been attending online English classes and they reported that their English classes were delivered online for 7 months due to the Covid-19 pandemic. Hence, both groups (i.e. immediate and delayed feedback) had a similar experience with online learning.

Study design, materials and data collection procedure

As we explored the role of incidental and unfocused recasts, it was not possible to predict the linguistic features recasts would address in advance. Hence, it was not possible to employ a pretest-posttest design similar to studies that examine focused feedback on predetermined features. We also decided not to use individualized tailor-made tests because they do not provide sufficient evidence of learners’ prior knowledge of target features (Loewen, 2005; Nassaji, 2016). Alternatively, drawing on a method suggested by Nassaji (2009), we followed the procedure below (also Flowchart 1 in Appendix A illustrates the experimental schedule described here).

After the participants and their parents read the information sheet and consented to take part in the study, the first author arranged two online sessions via Zoom with the participants and the parents. The researcher met each participant separately and engaged in dyadic online interaction. The first session consisted of three stages: preinteraction, dyadic interaction and immediate postinteraction error identification/correction (i.e. immediate posttest). The second session involved delayed postinteraction error identification/correction (i.e. delayed posttest).

At the preinteraction stage of the first session, the researcher explained to the participants that they would watch five scenes from a Charlie Chaplin film, ‘The Kid’, and they would use the chat to answer several questions about the plot. The researcher shared her screen so as to show the film and a word document with questions written in English (Appendix B). Prior to each scene, the participants had the opportunity to read the questions in advance and ask for clarification if needed. After they read the questions, the researcher played a specific scene. When the researcher stopped the scene, she shared the questions again and the participants were asked to answer them in the chat. When the participants answered all the questions, the same procedure was followed for the other scenes of the film and the researcher saved the chat log when this process was complete. At the preinteraction stage, the participants did not receive any recasts.

At the dyadic interaction stage, in the same session, the participants were assigned to one of two groups: immediate feedback (n = 29) and delayed feedback (n = 29). Both groups were asked to watch the same scenes again and use the chat to answer the same questions presented in the word document. Similarly to Nassaji’s (2009) work, the researcher asked the participants to provide answers that were as close as possible to their initial answers at the preinteraction stage. However, at this stage, the participants received incidental and unfocused recasts addressing semantic and morphosyntactic errors.

In the immediate feedback condition, written recasts were supplied by the first author in the chat immediately after learners’ erroneous output (example 3) so as to enable learners to engage in cognitive comparison between their production and target-like features. In the delayed feedback condition, the first author provided recasts in a word document approximately one minute after the participants answered all the questions of each scene (example 4). The word document with the delayed recasts had different tables for each scene and each table consisted of two columns. The researcher copy pasted learners’ errors made during interaction in the first column of the table and added recasts in the second column. The researcher supplied recasts next to the learners’ errors so as to encourage them to compare their erroneous output with the target-like features as the immediate feedback group also had this opportunity (i.e. immediate recasts were provided after learners’ production). Appendix C shows examples 3 and 4 in the Zoom platform.

Because the participants’ level of proficiency was low intermediate and recasts addressed a variety of features, the researcher provided delayed feedback after they completed one scene (mini task) rather than all five scenes. The participants did not take notes and they had one minute to read the delayed feedback as the pilot study indicated this was sufficient for the amount of feedback each scene elicited. Both the delayed and the immediate feedback groups received the same types of written recasts. Recasts were operationalized as full reformulations of learners’ erroneous output and they were declarative without additional input enhancement. For example, when answering the question ‘Why is the baby there?’, S49 from the immediate feedback group replied ‘Because they left parents’ while P50 from the delayed feedback group wrote ‘Because he left the parents’. In both cases, learners received the syntactic recast ‘Because his parents left him’. In both conditions, errors were not presented in isolation but as part of complete sentences in order to help learners create form-meaning mappings by associating their output with the scenes of the film. When the dyadic interaction stage was complete, the researcher saved the chat log of both groups and the word document with the recasts provided to the delayed feedback group. In the immediate feedback condition, recasts were part of the chat.

The postinteraction error identification and correction stage of the first session (i.e. immediate posttest) was the same for both groups and took place via Zoom. In particular, the researcher shared a document online with the initial answers the participants provided at the preinteraction stage and she asked them to read these answers and identify and correct errors if they could. Correction of errors that received a recast would show evidence that feedback was successful. The participants were not asked to share their thoughts while revising their answers and there was no time limit for this task. Two weeks later, the researcher arranged a second online session via Zoom for the delayed postinteraction error identification and correction task (i.e. delayed posttest). She played the scenes again and shared online the same questions about the plot of the film and the same document with the participants’ first answers (i.e. answers they had provided at the preinteraction stage in the chat via Zoom). The researcher asked them to review their answers and identify and correct errors they considered necessary without time pressure. The delayed task was included so as to examine whether the benefits of recasts are sustained two weeks later. Following Nassaji’s (2009) suggestion, we did not inform the participants about the purpose of the delayed task in advance as this could encourage them to revise linguistic features they noticed at previous stages.

It is important to emphasize that the method we employed explored L2 gains by focusing on learners’ correction of errors at the immediate and delayed postinteraction stages; however, we did not use think aloud protocols, stimulated recall or eye-tracking methodology to examine the extent to which learners attended to immediate or delayed recasts at the dyadic interaction stage. It is possible that learners noticed the feedback without being able to correct errors after interaction.

(3) Immediate feedback during dyadic interaction in the chat (participant 39).

Researcher: Where is the baby? How would you describe the place where the baby is laying?

Learner: He is in a swing.

Researcher: The baby is laying in a sling. (recast)

Researcher: What is the baby doing?

Learner: It is drinking milk.

Researcher: What is the baby using?

Learner: It is using a μπιμπερό. (baby bottle)

Researcher: It is using a tea pot. (recast)

Researcher: What is Charlie doing and why?

Learner: He making a γιογιο (pottie) for the baby

Researcher: He is making a home-made pottie for the baby. (recast)

Researcher: Do you think he is a good father? Why or why not? (a question about a specific scene)

Learner: I think it’s a good father because he makes the baby laught and he takes care of him.

Researcher: He is a good father because he is making the baby laugh and he is taking care of him. (recast)

(4) Delayed feedback after each scene.

Table

SCENE 3
Your answers (participant 13)	My comments
It is in a κούνια. (crib) It drinks some milk. It uses a κανάτα. (jug) He cut a piece of a chair to make a yoyo for the baby. Because he cares the baby.	It is laying in a sling. It is drinking some milk. It is using a tea pot. He is cutting a chair to make a home-made pottie for the baby. He is taking care of the baby.

Note. Learners referred to the baby as ‘it’ or ‘he’. Although ‘he’ is more appropriate, the researcher accepted both pronouns because the film referred to the baby as ‘it’.

Preliminary analysis

First, we needed to ensure that the two groups that received immediate and delayed feedback were comparable. Although the tutor of the participants had enrolled them in low intermediate classes, we needed to make sure that the two groups achieved similar accuracy levels in their performance. To this end, participants were assigned to one of two conditions (i.e. immediate versus delayed feedback) through stratified random sampling taking into account their accuracy score in the description they produced at the preinteraction stage before receiving feedback. The accuracy analysis of the preinteraction data was important to ensure the two groups had a similar starting point.

Accuracy was assessed using a weighted clause ratio (Foster & Wigglesworth, 2016). Learners’ output was divided into clauses and these clauses were assigned to four levels taking into account error gravity. Each level was given a different score (see Table 1). A final accuracy score was calculated by adding up the clause ratings and dividing the sum by the total number of clauses. Unlike other measures of accuracy that involve a binary distinction of errors (i.e. error-free versus not error-free), weighted clause ratio is more nuanced and provides stronger evidence of learners’ accuracy level. Because we needed to ensure that the immediate and the delayed feedback groups had a similar mean accuracy score, this analysis was done by the researcher during the preinteraction session (i.e. before learners engaged in dyadic interaction that involved recasts) when participants watched different scenes of the Charlie Chaplin film and shared their answers with the researcher before watching another scene. The accuracy score was calculated while the participants were watching a different scene of the film and while they were answering the questions for that scene. This was possible because the participants were young, their level of proficiency was relatively low, they produced simple language that was easy to analyse and their pace was slow enough for the researcher to complete the accuracy analysis and assign the participants to one of two conditions. 25% of the data were coded twice by both authors at different stages of data collection in order to identify potential discrepancies. Kappa was calculated to determine whether the two raters were consistent and it was found sufficiently high (.81).

Table 1. Categorizing clauses for accuracy.

Clause Category	Score	Definition	Examples from the current study
Entirely accurate	1.00	The clause is accurately constructed.	He is a good father.
Level 1	0.80	The clause has only minor errors that do not compromise meaning.	He want to keep the baby.
Level 2	0.50	The clause contains serious errors but the meaning is recoverable, though not always obvious.	and to took some money
Level 3	0.10	The clause has very serious errors that make the intended meaning far from obvious and only partly recoverable.	fly stones

Note. The classification and description of different levels is from Foster and Wigglesworth (2016).

This process enabled us to ensure that the mean accuracy score of the two groups was similar at the preinteraction stage prior to the provision of recasts (M = .717, SD = .15 in the immediate feedback condition and M = .716, SD = .18 in the delayed feedback condition).

Before exploring potential L2 gains, the second author also examined whether the two groups were similar in (1) the total number of recasts they received, (2) the number of errors they made, and (3) the number of changes recasts involved in each condition. Regarding the number of recasts, 559 recasts were supplied in the immediate feedback condition (M = 19.28, SD = 4.68) and 566 recasts in the delayed feedback condition (M = 19.52, SD = 5.36). Independent-samples t tests showed there was no significant difference between the two groups (t = .182, p = .856, d = .06) in the number of recasts they received. As for the number of errors that were the same as the preinteraction description, the second author identified 973 errors in the immediate feedback group (M = 33.55, SD = 13.08) and 963 errors in the delayed feedback group (M = 33.21, SD = 13.25). To investigate the reliability of the coding, the first author coded 25% of the errors. Kappa was computed to examine consistency and was found high (.96). Independent-samples t tests demonstrated there was no significant difference between the two groups in the number of errors they made (t = .100, p = .921, d = .03).

The second author also coded recasts for the number of changes they involved. Following Loewen and Philp (2006), they were classified as ‘one change’ recasts (example 5) and ‘two or more changes’ (example 6). The first author coded 25% of the recasts. Kappa was found high (.99) showing that the two coders were consistent. When looking at the two groups, in the immediate feedback condition, 38.64% of the recasts involved one change (M = 40.58, SD = 17.23) and 61.35% two or more changes (M = 58.97, SD = 17.97). In the delayed feedback condition, 39.92% of the recasts made one change (M = 41.74, SD = 16.37) and 60.07% had two or more changes (M = 58.25, SD = 16.37). Independent-samples t tests indicated there was no significant difference between the two groups (t = .263, p = .793, d = .09 for one change recasts and t = .158, p = .875, d = .05 for recasts that involved two or more changes). It is also important to mention that the two groups did not modify their output after receiving immediate or delayed recasts.

(5) One change recast.

Participant: Because he don’t want to keep it.

Researcher: Because he doesn’t want to keep it.

(6) Recast - two or more changes.

Participant: He think why is the baby at the curb.

Researcher: He is thinking why the boy is sitting on the curb.

Analysis of the preinteraction-interaction-postinteraction data and L2 gains

After ensuring the two groups were comparable, we proceeded with the analysis of learners’ performance at preinteraction-interaction-immediate and delayed postinteraction stages in order to measure L2 gains. When analysing the preinteraction data of both groups, the second author identified learners’ errors and coded them by linguistic category. We used a classification first suggested by Liu (2008) and later adopted by Smith (2012) (see Table 2). We used their coding scheme because we agree with Smith (2012) that it is clear and precise; however, unlike Liu and Smith, we decided to ignore spelling errors because it was not possible to distinguish typos from spelling errors. Another difference is that although we used both authors’ coding procedure to identify morphological and syntactic errors separately, we combined these two categories in our analysis (i.e. morphosyntactic errors) because both groups made only a few syntactic errors.

Table 2. Procedures for marking errors. From Liu (2008) and Smith (2012).

Type of error	Description	Examples from the current study
Morphological errors	All errors in verb tense or form; plural or possessive ending incorrect, omitted, or unnecessary; subject-verb agreement errors; article or other determiner incorrect, omitted, or unnecessary.	He want to keep the baby.
Syntactic errors	Errors in sentence/clause boundaries (run-ons, fragments, comma splices), word order, other ungrammatical sentence constructions.	Because they left parents (instead of ‘His parents left him’).
Semantic errors	Errors in word choice, including preposition and pronoun errors; omitted words or phrases, unnecessary words or phrases.	It is using a μπιμπερό (baby bottle).

After the analysis of the preinteraction data, the second author continued with the analysis of the data at the interaction stage following the same procedure for both conditions (i.e. immediate and delayed feedback). First, recasts were identified and coded for linguistic category following the classification presented in Table 2. To verify the reliability of the coding scheme, the first author independently coded approximately 25 per cent of the data, randomly selected through stratified random sampling from both groups. Kappa was computed to determine consistency between the two raters and was found to be sufficiently high (.83).

Then, we compared the preinteraction and interaction data to identify whether the errors made at the preinteraction stage and those that triggered recasts at the interaction stage were the same or similar (e.g. the same type of morphological error using a different verb). Drawing on Nassaji’s (2009) methodology, we only focused on those recasts that addressed errors made at both the preinteraction and interaction stage.

Finally, the second author analysed the immediate postinteraction data by looking at the changes the participants made when they were asked to identify and correct their errors from the preinteraction stage. Similarly to Nassaji’s (2009) method, we only focused on those errors that received recasts so as to measure learners’ development after receiving feedback. The errors were coded by linguistic category using the same classification as at previous stages. When students made an accurate linguistic change, they received one point even if other linguistic elements were erroneous. For example, if a student changed a word accurately without correcting the tense, they received one point. If they changed both the word and the tense, they received two points. The score was zero when (1) they did not make any change, (2) they revised the content of their output without addressing their error, or (3) they made a non-target-like linguistic change (e.g. they wrote an irrelevant word). In order to examine whether L2 development of semantic and morphosyntactic features was sustained, the second author did the same analysis for the delayed postinteraction data which involved error identification and correction two weeks after receiving recasts. The first author coded 25% of the data twice and Kappa was .95 for the immediate postinteraction test and .96 for the delayed postinteraction test. Appendix D shows an example of how we calculated a participant’s identification and correction of errors at the immediate and delayed postinteraction stages.

Statistical analyses

For the first research question, we compared L2 gains (i.e. error corrections) in the immediate and delayed feedback conditions without taking into account linguistic focus. In other words, we analysed L2 development from both semantic and morphosyntactic recasts. Before analysing the data, we looked at the kurtoses and skewness values, the histogram and we ran a Shapiro-Wilk test to examine whether the data were normally distributed. These measures indicated that either transformations or non-parametric tests were required (Shapiro-Wilk test p < .05). After successfully transforming the data, we used independent-samples t tests to explore differences between the two groups and we also calculated Cohen’s d to examine effect sizes. Drawing on Plonsky and Oswald (2014), the effect sizes were interpreted as small when Cohen’s d was close to .40, medium when Cohen’s d was close to .70, and large when it reached 1.00.

For the second research question, we explored differences in L2 gains for the semantic and morphosyntactic features separately. As we transformed the data successfully for the semantic features, we followed the same procedure described above (i.e. the transformed data conformed to normality; hence, independent-samples t tests and Cohen’s d measures were used). However, when we analysed morphosyntactic features, transformations did not lead to normal distribution; hence, non-parametric tests were employed. In particular, we used the Mann-Whitney U test and the effect sizes were interpreted as small when r was close to .25, medium when r was close to .40, and large when r reached .60.

Results

Research question 1

The first research question explores whether participants demonstrate L2 gains after receiving unfocused, written SCMC recasts and the extent to which feedback timing influences their learning as demonstrated by the accurate modifications they made . According to Table 3, in the immediate postinteraction error identification/correction task (i.e. immediate posttest), both groups made targetlike modifications after receiving recasts. The immediate feedback group corrected 206/973 errors (21.17%) and the delayed feedback group corrected 299/963 errors (31.04%). The mean of accurate changes in the immediate feedback condition was 21.25 (SD = 16.1) and in the delayed feedback condition, it was 33.68 (SD = 18.86). Learners made accurate changes significantly more when they received delayed feedback as compared to immediate feedback (t = −2.79, p = .007, 99% CI = −2.59, −0.06). The effect size was large (d = 1.03).

Table 3. Descriptive statistics for gain scores per group.

	Immediate posttest	Delayed posttest
Group	M (SD)	M (SD)
Immediate feedback	21.25 (16.1)	18.47 (11.92)
Delayed feedback	33.68 (18.86)	21.52 (16.76)

Regarding the delayed postinteraction error identification/correction task (i.e. delayed posttest), the immediate feedback group corrected 167/973 errors (17.16%) and the delayed feedback group corrected 185/963 errors (19.21%). The mean of correct changes in the immediate feedback group was 18.47 (SD = 11.92). In the delayed feedback condition, the mean was 21.52 (SD = 16.76). The difference between the two groups was not significant (t = −.891, p = .377, 99% CI = −0.27, 0.13). The effect size was small (d = .34). In other words, the two groups showed similar development in the delayed postinteraction test. This indicates that although delayed feedback generated significantly greater immediate gains, they were not sustained two weeks later.

Research question 2

The second research question examined whether L2 learners of English benefit from recasts with different linguistic focus and the extent to which feedback timing influences their development. According to Table 4, in the immediate feedback condition 567 (58.27%) errors were classified as semantic and learners accurately changed 136 (23.98%) of these errors in the immediate postinteraction test and 122 (21.51%) in the delayed test. The mean of accurate changes was 24.32 (SD = 16.89) in the immediate test and 21.92 (SD = 15.09) two weeks later. In the delayed feedback condition, 600 (62.3%) errors were coded as semantic. Learners correctly changed 215 (35.83%) semantic errors in the immediate test and 146 (24.33%) in the delayed test. The mean of accurate changes was 38.42 (SD = 21.19) in the immediate test and 26.07 (SD = 17.91) two weeks later.

Table 4. Descriptive statistics for gain scores of semantic features per group.

	Immediate posttest	Delayed posttest
Group	M (SD)	M (SD)
Immediate feedback	24.32 (16.89)	21.92 (15.09)
Delayed feedback	38.42 (21.19)	26.07 (17.91)

When we compare the two groups in the immediate test, the delayed feedback group corrected significantly more semantic errors than the immediate feedback group (t = −2.801, p = .007, 99% CI = −27.51, −0.67). The effect size was large (d = 1.04). In the delayed test, the difference between the two groups was not significant (t = −1.109, p = .272, 99% CI = −1.63, 0.67) and the effect size was small (d = .41). In other words, delayed feedback was more beneficial than immediate feedback when addressing semantic features but L2 gains were not sustained two weeks later.

Regarding morphosyntactic features, according to Table 5, learners made 406 (41.72%) errors in the immediate feedback condition. They accurately changed 70 (17.24%) of these errors in the immediate postinteraction test and 45 (11.08%) in the delayed test. Because non-parametric tests were used, the median and the IQR are reported. The median of accurate changes in the immediate test was 8.33 (IQR = 27.45) and in the delayed test 3.22 (IQR = 13.66). In the delayed feedback condition, 363 (37.69%) errors were classified as morphosyntactic. Learners corrected 84 (23.14%) errors in the immediate postinteraction test and 39 (10.74%) in the delayed test. The median of accurate changes in the immediate test was 13.04 (IQR = 32.33) and in the delayed test 6.66 (IQR = 15.99).

Table 5. Descriptive statistics for gain scores of morphosyntactic features per group.

	Immediate posttest	Delayed posttest
Group	median (IQR)	median (IQR)
Immediate feedback	8.33 (27.45)	3.22 (13.66)
Delayed feedback	13.04 (32.33)	6.66 (15.99)

When comparing the two groups, there were no significant differences (in the immediate posttest Mann-Whitney U = 347.50, p = .250, r = .15 and in the delayed posttest Mann-Whitney U = 408.5, p = .844, r = .02).

Because semantic and morphosyntactic categories are broad, we also conducted a more detailed analysis of different types of errors within each category. We identified three sub-categories within semantic features; use of a non-target-like word or phrase (e.g. ‘he’s flying stones’ instead of ‘he’s throwing stones’), omission of a word or phrase, and use of L1 (e.g. ‘he places the baby in a “καροτσάκι”’ instead of “pram”). Regarding morphological and syntactic subcategories, we identified errors in tenses or verb forms, determiners, a few cases of non-target-like word forms (e.g. ‘lices’ instead of ‘lice’) and a few errors in word order, fragments and embedded questions (e.g. ‘He is wondering where are his parents’). Tables 6 and 7 demonstrate the distribution of these subcategories in the two feedback conditions and the number and percentage of changes learners made in the immediate and delayed postinteraction tests. In both conditions, most changes were in the categories ‘word/phrase choice’ and ‘L1 use’, especially when students received delayed feedback. With respect to the morphosyntactic categories, both groups demonstrated greater gains in tenses and verb forms as compared to other morphosyntactic features. Interestingly, both groups exhibited limited gains in the use of determiners and syntactic constructions (i.e. word order, embedded questions, and fragments). The findings for both research questions are discussed in the following section.

Table 6. Distribution of types of errors and accurate changes in the immediate and delayed postinteraction tests in the immediate feedback group.

	Total	Immediate postinteraction changes	Delayed postinteraction changes
Types of errors	n(%)	n(%)	n(%)
Word/phrase choice	329/567 (58.02%)	80/329 (24.31%)	70/329 (21.27%)
Word/phrase omission	64/567 (11.28%)	6/64 (9.37%)	5/64 (7.81%)
L1 use	174/567 (30.68%)	50/174 (28.73%)	47/174 (27.01%)
Verb form/ tense	290/406 (71.42%)	63/290 (21.72%)	40/290 (13.79%)
Determiners	66/406 (16.25%)	2/66 (3.03%)	3/66 (4.54%)
Word form	15/406 (3.69%)	3/15 (20%)	0/15 (0%)
Word order/embedded questions/fragments	35/406 (8.62%)	2/35 (5.71%)	2/35 (5.71%)

Note. 567 errors were classified as semantic and 406 errors were classified as morphosyntactic.

Table 7. Distribution of types of errors and accurate changes in the immediate and delayed postinteraction tests in the delayed feedback group.

	Total	Immediate postinteraction changes	Delayed postinteraction changes
Types of errors	n(%)	n(%)	n(%)
Word/phrase choice	323/600 (53.83%)	112/323 (34.67%)	65/323 (20.12%)
Word/phrase omission	58/600 (9.66%)	14/58 (24.13%)	10/58 (17.24%)
L1 use	219/600 (36.5%)	89/219 (40.63%)	71/219 (32.42%)
Verb form/ tense	250/363 (68.87%)	71/250 (28.4%)	35/250 (14%)
Determiners	75/363 (20.66%)	7/75 (9.33%)	1/75 (1.33%)
Word form	13/363 (3.58%)	2/13 (15.38%)	1/13 (7.69%)
Word order/ embedded questions/ fragments	25/363 (6.88%)	4/25 (16%)	2/25 (8%)

Note. 600 errors were classified as semantic and 363 errors were classified as morphosyntactic.

Discussion

The study demonstrates that when learners engage in SCMC interaction, they benefit more from delayed recasts as compared to immediate ones. In particular, in the immediate postinteraction test, the delayed feedback group corrected their errors significantly more than the immediate feedback group. When looking at the linguistic focus of recasts, delayed recasts were more effective than immediate ones when addressing semantic features. Regarding morphosyntactic features, no difference was found between the two groups. This section discusses how feedback timing affected semantic and morphosyntactic features separately.

Semantic features

The results of the current study raise questions of why delayed SCMC recasts were more effective than immediate ones in facilitating corrections of semantic features. A possible explanation is that semantic features involve lexical items (e.g. ‘pram’, ‘tea pot’, ‘pottie’) that are more concrete and less complex than morphosyntactic constructions (Gass et al., 2003). According to previous research, vocabulary learning may involve different processes; for example, rote memorization or the keyword method (i.e. learners make connections of L1-L2 words that share acoustic or orthographic similarities) (see Sagarra & Alba, 2006 for a review). As we did not ask learners to share their thoughts while reading the feedback, the current study does not show what techniques or strategies learners used so as to remember and correct semantic features in the immediate postinteraction test.

As explained earlier, a theory that could be used to account for the benefits of delayed feedback is Skehan’s (2009, 2014) Limited Attentional Capacity Model. In the current study, we did not measure the cognitive demands of the task; nevertheless, the significant difference between the two groups when looking at corrections of semantic features at the immediate postinteraction test shows that the provision of delayed feedback enabled learners to process recasts more effectively. Although more empirical research on feedback timing in the SCMC environment is needed, it could be tentatively suggested that from an information processing perspective, providing recasts after a communicative task may enable learners to allocate more attention to semantic features and engage in deeper processing without the pressure that the flow of real-time communication naturally poses. Furthermore, focusing on form after interaction, as in the delayed feedback condition, might have increased the explicitness of recasts. It is possible that learners perceived the process of receiving delayed recasts as time dedicated to focus on language. Drawing on previous research, explicit recasts are more effective in facilitating noticing than implicit ones (e.g. Kim, 2018; Kim & Han, 2007; Nassaji, 2009). Although relatively implicit reformulations were provided in this study, as there were no explicit comments or metalinguistic information, providing recasts after interaction might have rendered their corrective intention clear to students.

When looking at different categories of semantic errors, an interesting observation is that learners benefitted the most from delayed semantic recasts that addressed L1 features. According to Table 7, the delayed feedback group corrected 89/219 features (40.63%) in the immediate postinteraction test. It is possible that when learners used their L1, it was because of a gap in their knowledge; and consequently, they noticed information supplied in the feedback (see Doughty, 2001; Swain, 1995).

Interestingly, unlike the current study, Henderson (2019) demonstrated no difference between immediate and delayed feedback on vocabulary development. An important difference between the two studies is that we provided incidental and unfocused recasts on semantic and morphosyntactic features; whereas in Henderson’s study, the focus of recasts was 10 predetermined vocabulary items. Also, our participants were young EFL low intermediate learners while Henderson recruited university students. The incongruent findings indicate that before we draw conclusions about whether immediate or delayed feedback is more effective in the written SCMC mode, we need more research to delve into how the age of participants or differences between processing of focused and unfocused recasts may influence the role of feedback timing. It is worth highlighting that one methodological decision we made when we supplied delayed feedback and has potentially contributed to its effectiveness was that recasts were delivered after a mini task corresponding to one scene from the film. In other words, participants were not required to process a large number of unfocused recasts after the film as this would not be an appropriate intervention for young low intermediate learners with shorter attentional span than adult learners.

Finally, it is important to emphasize that the delayed feedback condition was more advantageous than immediate feedback only in the immediate postinteraction test. No difference between the two feedback conditions was found two weeks later in the delayed postinteraction test. In other words, the L2 gains of the delayed feedback group were not sustained. This indicates that more interventions and opportunities for practice are needed to consolidate learners’ knowledge of semantic features (e.g. explicit instruction, task repetition accompanied by additional feedback).

Morphosyntactic features

Regarding morphosyntactic features, both the immediate and the delayed feedback groups demonstrated limited gains. In addition, no difference was found between the two conditions. The low rate of corrections in both feedback conditions also demonstrates that recasts were less effective when addressing non-salient and redundant features (e.g. determiners) or complex constructions (e.g. embedded questions, word order and fragments). Our finding is in line with other studies which also show that SCMC recasts are less successful when targeting non-salient linguistic elements (e.g. the present third person singular in Kourtali, 2022; past tense in Loewen & Erlam, 2006; English zero article followed by abstract uncountable nouns in Sauro, 2009; Turkish locative case morphemes in Yilmaz, 2012). In a similar vein, Smith (2012) also demonstrated that semantic and syntactic recasts were more effective in facilitating noticing than morphological recasts. In the current study, learners exhibited limited gains when they received syntactic recasts; however, the task elicited relatively simple language and learners made only a few syntactic errors. Hence, only tentative conclusions can be drawn about the effects of unfocused, syntactic, SCMC recasts from this empirical work.

Students’ corrections indicate that they benefited more from semantic recasts possibly because they had features that assisted successful communication to a greater extent than morphosyntactic constructions. For example, they needed to know the word ‘pram’ in order to provide information about where the baby was in the film and this information was more crucial than communicatively redundant elements (e.g. articles). Hence, our hypothesis that delayed feedback would increase the salience of morphosyntactic features and would provide learners with time to engage in deep cognitive processing of these features was not confirmed. Nevertheless, it is not clear how feedback timing would have influenced learning if focused feedback was provided on a single grammatical feature consistently without addressing vocabulary. For example, Arroyo and Yilmaz (2018) showed that when learners received focused feedback in the written SCMC environment, immediate feedback was more effective than delayed feedback.

Pedagogical Implications

Although more empirical research on feedback timing in the written SCMC mode is needed, a pedagogical implication of the current study is that when CALL tutors provide interactional and unfocused feedback in the form of recasts using text-based chat, learners may benefit more from delayed recasts delivered after they have completed a communicative task, especially when semantic errors are addressed. Tutors could look at the features of different platforms so as to explore how delayed feedback could be provided. For example, when learners perform tasks using the chat in Zoom, tutors could save chat logs and draw learners’ attention to erroneous linguistic features after interaction. Regarding more complex, communicatively redundant and non-salient morphosyntactic constructions, the current empirical work suggests that more explicit or longer interventions might be needed.

Conclusion and future directions

The study examined whether feedback timing influences L2 development of young low intermediate learners of English when unfocused recasts are provided in the written SCMC mode. Learners’ accurate modifications in the immediate postinteraction test show that they benefit more from delayed recasts delivered after they have completed a communicative task as compared to immediate recasts supplied during interaction when the linguistic focus of feedback is semantic features. However, no difference was found between the two feedback conditions in the delayed postinteraction test, indicating that additional input and opportunities for practice are needed to assist consolidation of knowledge in the long term. Regarding corrections of morphosyntactic features, there was no difference between the two groups. Considering that there was a low rate of morphosyntactic corrections in both conditions, the current study suggests that these features are less amenable to SCMC recasts and more explicit or longitudinal interventions might be needed.

Although we are hoping that these findings are informative for teachers, especially those who work in the CALL environment, there are some limitations to this study that should be acknowledged. First, we only looked at how immediate and delayed feedback influence learning outcomes by measuring error corrections after interaction. However, it is possible that learners might have noticed linguistic features when receiving feedback without being able to correct these features after interaction. Although learners exhibited gains when they received delayed semantic recasts, their behaviour during interaction is still unclear. Considering that not only learning outcomes, but also learning processes are crucial in L2 learning, future studies could employ stimulated recall (SR) methodology to measure noticing and/or eye-tracking to explore learners’ cognitive processes (i.e. what learners attend to when receiving immediate or delayed feedback). Think aloud protocols could also be used to delve into learners’ thoughts at the postinteraction stage when they are asked to correct errors. Sharing their thoughts might reveal that they identify errors that they are not able to correct or they partially remember the target-like feature.

Furthermore, we used only one outcome measure that involved error corrections. Future studies could include comprehension-based measures (e.g. grammaticality judgment tests) or learners’ ability to recognize lexical items without necessarily being able to retrieve them for production. This would be an important addition as previous vocabulary research has demonstrated that receptive knowledge precedes productive knowledge (Schmitt, 2019). Also, measuring learning outcomes in conditions similar to those that involve provision of recasts might show different results from this study. Drawing on the Transfer Appropriate Processing model, it is possible that learners might perform better when the test conditions require a written SCMC task similar to the treatment task that involves recasts.

We also acknowledge that we did not test the levels of mental effort that the immediate and delayed feedback conditions required. We assume that when unfocused feedback is provided, delayed feedback might be a less cognitively demanding condition than immediate feedback because the latter requires learners to process semantic and morphosyntactic features simultaneously during real-time communication; however, future studies could delve into the cognitive demands of each condition by employing questionnaires to compare participants’ perceptions about the mental effort they exert while receiving feedback (see Kourtali & Révész, 2020).

Another shortcoming of this paper is that there was no control group as we focused on comparing the effects of immediate and delayed feedback. Future studies could elucidate whether a control group may also exhibit L2 gains as a result of practice; however, in the current study, we only focused on errors that learners made twice at the preinteraction and interaction stages and this shows that they needed feedback. We would also like to highlight that the results cannot be generalized for adult or more advanced L2 learners, focused and intensive feedback on predetermined language features, different types of tasks, modes of interaction (e.g. oral feedback) or types of feedback (e.g. explicit correction). When examining morphosyntactic development, more explicit or longitudinal interventions might be needed to draw learners’ attention to non-salient, communicatively redundant features or complex and abstract syntactic constructions. Nevertheless, it should be noted that there are different views in SLA about the extent to which explicit presentation facilitates L2 development of complex morphosyntactic rules (see Gass et al., 2003 for a review).

Finally, in the current study, we explored the role of immediate and delayed recasts by using the chat function in Zoom. Future studies could delve into the effects of feedback timing when interaction occurs in the CALL or MALL environment via platforms with different features from Zoom. For example, some platforms (e.g. Teams) provide opportunities for simultaneous editing by sharing a word document.

Acknowledgements

We would like to express our gratitude to the participants of our study, as well as to the anonymous reviewers.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The study was supported by the Early Career Researcher and Returners Fund and RDIF, University of Liverpool.

Notes on contributors

Nektaria-Efstathia Kourtali

Dr Nektaria-Efstathia Kourtali is a TESOL and Applied Linguistics lecturer at the University of Liverpool. Her research interests lie in the field of Second Language Acquisition and Language Pedagogy. In particular, she is interested in computer-mediated communication, interactional feedback, task-based language teaching and the role of individual differences. She has also worked on projects that explore cognitive processes underlying second language learners’ behaviours while writing.

Lais de Oliveira Borges is an English lecturer at Universidade Catolica de Brasilia, Brazil. Her research interests within Applied Linguistics include Second Language Acquisition, individual differences, speech production, pronunciation attainment, and language and identity.