Introduction
Corrective feedback (CF) refers to responses to errors learners make in producing or comprehending a second language (L2). CF has been at the core of the theory, pedagogy, and research of second language acquisition (SLA), and its significance is evidenced by the large body of research that has been accumulated during the past two decades. The large amount of research has been synthesized in a number of reviews including state-of-the-art reviews (Ellis Reference Ellis and Batstone2010; Li Reference Li and Liontas2018a), meta-analyses (Norris & Ortega Reference Norris and Ortega2000; Lyster & Saito Reference Lyster and Saito2010; Brown Reference Brown2016; Li Reference Li and Gurzynski-Weiss2017a) and monographs (Mackey 2015; Nassaji Reference Nassaji2015; Nassaji & Kartchava Reference Nassaji and Kartchavain press). The research has examined CF from various perspectives, such as the incidence of CF in the classroom (Lyster & Ranta Reference Lyster and Ranta1997; Choi & Li Reference Choi and Li2012), the effectiveness of CF (Révész Reference Révész2009; van de Guchte et al. Reference van de Guchte, Rijlaarsdam, Braaksma and Bimmel2015), the impact of individual difference variables on CF effectiveness (Yilmaz Reference Yilmaz2013; Li Reference Li and Han2014), learners’ perceptions of CF (Philip Reference Philip2003; Egi Reference Egi and Mackey2007), and teacher and learner attitudes toward CF (Li Reference Li, Nassaji and Kartchava2017b). However, one topic that has received little attention in the research but is of theoretical and pedagogical significance is the timing of CF, that is, whether CF should be provided immediately after an error is made during a communicative task or whether it should be postponed until the task is completed. In this article, I call for the replication of two recently published studies on the comparative effectiveness of immediate and delayed feedback: Li et al. (Reference Li, Zhu and Ellis2016) and Arroyo & Yilmaz (Reference Arroyo and Yilmaz2018).
These two studies deserve replication because of the following reasons. First, the two studies address a new and original topic. Given that they are the first and probably the only published studies that have investigated the timing of feedback as an independent variable, replication studies are needed that will verify and shed light on the initial findings and that will demonstrate whether the impact of feedback timing varies as a function of learner- and context-related factors, such as learners’ L2 proficiency or instructional context. Second, both studies address a significant topic. Whether to provide CF during or after a communicative task is key to understanding SLA theories’ claims and explanations about the mechanism of language learning, and it has been a major theme of investigation of multiple strands of research, such as CF, task-based instruction, and form-focused instruction. Replicating the two studies will lead to firmer conclusions about the influence of CF timing and afford valuable implications for theory-building, classroom practice, and policy-making (Porte Reference Porte and Porte2012). Third, the two studies address the same topic using different methods: Li et al.’s study was carried out in the classroom while Arroyo & Yilmaz's study was conducted in the laboratory; in Li et al.’s study the feedback treatment was delivered in face-to-face interaction, whereas in Arroyo & Yilmaz's study the treatment was implemented in a computer-mediated communication mode. Replications of the two studies will enable us to make meaningful comparisons between the two different types of research and obtain a clearer picture of the comparative effectiveness of the two types of CF in different instructional settings and different modes of interaction. Fourth, both studies utilized rigorous methods. The treatments of the studies were theoretically justified and consistently implemented, and the treatment effects were assessed via valid measures. Of course, the rigor of the studies is also buttressed by the fact that they were published in prestigious journals in the field: Modern Language Journal (Li et al. Reference Li, Zhu and Ellis2016) and Language Learning (Arroyo & Yilmaz Reference Arroyo and Yilmaz2018). Fifth, both studies provided detailed information about their procedures and made their materials and instruments available on IRIS (Marsden, Plonsky & Mackey Reference Marsden, Mackey, Plonsky, Mackey and Marsden2016)—a free, open depository of research materials in applied linguistics—or as supplementary information at the journal website. In fact, Arroyo & Yilmaz called for replication of their study in various sections of the article including the title.
In the following sections, I provide an overview of the background of the theoretical and pedagogical significance of the timing of CF, followed by a synthesis of the relevant research. I proceed to summarize the two studies, reporting and critiquing their methods and findings, and following a summative analysis of each study, I make recommendations on how to replicate the studies. A conclusion is offered at the end.
Background
To prepare for the later discussion of different types of CF, it is necessary to provide a taxonomy of CF. In their seminal observation study, Lyster & Ranta (Reference Lyster and Ranta1997) identified six types of CF in French immersion classes: recasts, explicit correction, elicitation, clarification requests, repetition, and metalinguistic clues. Recasts and explicit correction, which provide the correct form, were called reformulations. The remaining four CF types withhold the correct form and encourage self-correction, and they were collectively referred to as prompts. Lyster & Ranta pointed out that teachers should provide prompts rather than reformulations because prompts push learners to conduct deeper cognitive processing. However, it can be argued that the two CF types facilitate L2 development in different ways because reformulations are more effective for learning new linguistic forms while prompts are better at consolidating previously learned forms (Goo & Mackey Reference Goo and Mackey2013; Lyster & Ranta Reference Lyster and Ranta2013). CF can also be divided into explicit and implicit depending on whether learners’ attention is drawn to linguistic forms. Based on this taxonomy, recasts, elicitation, repetition, and clarification are relatively implicit while explicit correction and metalinguistic feedback are more explicit. In the next section, I discuss the theoretical claims on CF timing.
Most SLA theories do not make overt claims about the ideal time to provide CF. However, their perspectives on CF timing can be inferred by examining their claims about the role of CF in language learning or the mechanism through which CF affects L2 development. To start with, according to the currently popular Interaction Hypothesis (Long Reference Long2015), CF embedded in negotiated interaction facilitates learning by helping the learner notice the gap between the erroneous production and the correct form. The theory further argues that the recast, which refers to the reformulation of an erroneous utterance, is an ideal form-focusing device on the grounds that it provides both positive (information about what is correct) and negative (information about what is wrong) evidence; it juxtaposes the correct and wrong forms so the learner can make a direct comparison; it keeps the meaning intact and thus frees up learners’ cognitive resources for linguistic forms; it is provided at a propitious moment when the learner is grappling with the appropriate form and is therefore motivating. Given that the theory stipulates that CF must be embedded in interaction and provided immediately after errors are made during a communicative task, delayed CF is not an ideal option.
Skill acquisition theory claims that learners need to have a solid base of declarative knowledge—which refers to knowledge about how language works—before engaging in practice activities where the knowledge is applied or proceduralized in actual language use. Skill acquisition theory has been drawn on as the theoretical model by researchers who advocate the use of prompts (Lyster Reference Lyster2004). Prompts, according to Lyster, push the learner to retrieve and use linguistic items from the existing repertoire, thus facilitating the proceduralization of previous knowledge. Skill acquisition theory also states that practice must be skill-specific (DeKeyser Reference DeKeyser, VanPatten and Williams2015); in other words, in order for declarative knowledge to be usable in a particular skill such as speaking, writing, listening, or reading, the knowledge must be practiced in activities involving that skill. Based on this argument, if CF aims to facilitate the proceduralization of the knowledge used in communication, it must be provided during, but not after, communicative tasks. Therefore, scholars who champion skill acquisition theory would also support immediate feedback and disfavor delayed feedback.
A third perspective that can be drawn on is sociocultural theory (Aljaafreh & Lantolf Reference Aljaafreh and Lantolf1994). According to this theory, the CF provider cannot make a priori decisions about the type and amount of CF to be provided; rather, CF must be tailored, graduated, and contingent. By “tailored” it is meant that CF must be provided in a manner that fits the need and level of individual learners. Graduated CF means that for the same individual, CF should start from the most indirect and become progressively more direct depending on how much assistance is needed to fix the error. “Contingent” refers to the idea that CF is provided only when it is necessary and that CF is withdrawn when it is no longer necessary. This type of CF is called scaffolding, which can only be provided in the learner's zone of proximal development (ZPD)—a metaphorical space with a lower and upper bound. The lower bound refers to independent performance and the upper bound to improved performance resulting from scaffolding. In what way does the theory relate to the timing of CF? Since scaffolding can only happen in the ZPD, and a learner's ZPD can only be fathomed and determined during the interaction between an expert and a novice, it can be inferred that in this theory CF must be immediate and take place during, rather than after, a communicative task.
One theoretical model that appears to have a direct link with the timing of CF is the Transfer Appropriate Processing (TAP) theory. Based on this theory, the context where learned knowledge is retrieved and applied must match the context in which the knowledge is acquired; that is, knowledge is not transferable between domains. This claim is reminiscent of skill acquisition theory. However, it must be pointed out that the two theories are distinct in that the TAP emphasizes the compatibility between the contexts of learning and use, whereas the central idea of skill acquisition theory is that declarative knowledge must be learned first and then practiced or applied in skill-specific activities. The TAP theory has guided the research on integrated and isolated form-focused instruction, a distinction that has been made by Spada and associates (e.g. Spada & Lightbown Reference Spada and Lightbown2008; Spada et al. Reference Spada, Jessop, Tomita, Suzuki and Valeo2014). In integrated form-focused instruction, linguistic forms are addressed during communicative activities such as through feedback or brief time-outs from the ongoing task where the teacher explains a grammar point. In isolated form-focused instruction, linguistic forms are dealt with separately and form-focused instruction is detached from communicative activities. The researchers referred to this distinction as relating to the timing of form-focused instruction. Casting their research in the TAP theory, the researchers argue that the two types of form-focused instruction contribute to different types of knowledge, that is, integrated form-focused instruction facilitates communicative competence, and isolated form-focused instruction leads to explicit knowledge or knowledge accessible in controlled tasks. Based on these arguments, immediate feedback constitutes integrated form-focused instruction, whereas delayed feedback is a type of isolated form-focused instruction. Therefore, immediate feedback will enhance communicative competence (or implicit knowledge) and delayed feedback will favor knowledge used in drills and grammar exercises (or explicit knowledge).
It would seem that all SLA theories support the utility of immediate feedback, primarily because of the emphasis they put on the importance of correcting errors during communication in acquiring communicative competence and because of their dismissal of decontextualized form-focused instruction. However, it can be argued that delayed feedback has an advantage over immediate feedback in certain respects. For example, delayed feedback allows learners to concentrate on form and alleviates them of the burden of having to attend to form and meaning simultaneously imposed by immediate feedback. Also, from a pedagogical perspective, those who embrace communicative approaches to language teaching worry that immediate feedback may disrupt the flow of communication and affect the development of fluency (Hedge Reference Hedge2000; Harmer Reference Harmer2007). Furthermore, from the perspective of task-based instruction, Willis & Willis (Reference Willis and Willis2007) consistently recommend providing form-focused instruction in the posttask stage on the grounds that addressing linguistic forms before or during a task may predispose or obsess learners with language and make them turn the task into a grammar exercise, thus undermining the meaning-primary principle of task-based instruction. In fact, this concern has been confirmed by Ellis, Li & Zhu's (Reference Ellis, Li and Zhu2019) study, which showed that providing explicit instruction prior to task performance had detrimental global effects on the complexity, accuracy, and fluency of their speech production. The posttask stage is ideal for form-focused instruction because (1) there is no risk of ruining the meaning-primary principle, (2) the foregoing engagement with the target form provides a context for the form-focused instruction, and (3) learners will be more motivated to benefit from the instruction after struggling with the linguistic form during task performance.
What has research demonstrated about the comparative effects of immediate and delayed feedback? Unfortunately, except for the two studies of which I am calling for replication, CF research has predominantly and exclusively investigated immediate feedback or feedback embedded in communicative tasks. This is possibly because of the dominance of the Interaction Hypothesis in the field of SLA, which attaches importance to incidental focus on form in meaning-oriented tasks. The meta-analytic reviews of the empirical CF research (Russell & Spada Reference Russell, Spada, Norris and Ortega2006; Mackey & Goo Reference Mackey, Goo and Mackey2007; Li Reference Li2010; Lyster & Saito Reference Lyster and Saito2010) all revealed significant effects for CF on L2 development. The reviews also showed that CF effectiveness varies as a function of feedback type and other constraining factors. Specifically, prompts are more effective than recasts in classroom settings; explicit feedback is more effective than implicit feedback in the short term but the effects of implicit feedback are more sustainable; the effects of feedback are more evident on free production tests; younger learners benefit more from feedback than older learners.
Aside from the variables analyzed in the meta-analytic reviews, the primary research has examined the impact of other constraining variables for CF effectiveness. Li (Reference Li and Han2014) examined a three-way interaction between feedback type, proficiency, and target structure. Li reported that for both the simple and complex structures, metalinguistic feedback were more effective than recasts at the low-proficiency level and that the two types of feedback were equally effective at the high-proficiency level. The results are in line with Ammar & Spada's (Reference Ammar and Spada2006) study on the interface between feedback type and proficiency. They reported that for learners whose pretest scores were below 50%, prompts worked better than recasts, but for learners whose pretest scores were above 50%, the two types of CF were equally effective.
One stream of research has examined the interface between feedback type and individual difference factors. In a meta-analysis, Li (Reference Li and Gurzynski-Weiss2017a) aggregated the results of the research on the associations between CF effectiveness and learners’ individual differences in language aptitude and working memory. Li found that the two cognitive variables were more strongly correlated with the effects of explicit feedback than those of implicit feedback. Li concluded that traditional aptitude and working memory are more important for conscious or explicit learning than unconscious or implicit learning. Li called for more research into the interface between feedback type and implicit aptitude—abilities to learn an L2 unconsciously (Yilmaz & Granena Reference Yilmaz and Granena2016).
To sum up, CF research has centered on the effectiveness of immediate CF and on how its effects are affected by and correlated with learner-internal and learner-external variables. There is an urgent need for more research comparing the effects of immediate and delayed CF and how their effects are constrained by other factors such as those examined in the research on immediate CF. In the following, I closely examine the methods and results of the two studies on CF timing and make recommendations on ways of replicating the two studies.
Study 1: Li, Zhu & Ellis (Reference Li, Zhu and Ellis2016)
Li et al.’s study investigated whether immediate and delayed CF was effective in facilitating L2 development and whether the two CF types had differential effects on learning outcomes. The participants were 120 eighth-grade English as a foreign language learners at a Chinese middle school. These learners were homogeneous in terms of their linguistic and cultural backgrounds and language learning experience. The homogeneity of the sample minimized the potential influence of these factors on the results and made it possible to see the effects of the independent variable (treatment type) more clearly. The learners were divided into four groups, each with 30 learners: immediate feedback, delayed feedback, task only, and control. The researchers used the learners’ mid-term exam scores to ensure equivalency of proficiency between the four participant groups. The study adopted a pretest-posttest design where all groups attended three sessions. In the first session, the learners took two pretests, a grammaticality judgement test (GJT) and an elicited imitation test (EIT). In the second session, the treatment groups attended a two-hour treatment followed by the immediate posttests, and the control group only took the posttests. In the third session, which happened two weeks after the second session, the learners took the delayed posttests.
Each of the three treatment groups performed two dictogloss tasks following similar procedures. The target structure was the English passive voice, a new structure that had not been taught to the students prior to the treatment. In each task, the learners listened to a story told by the teacher three times, then worked in pairs to practice retelling the story, and finally took turns to report the story to the rest of the class. The immediate CF group received immediate correction of the errors they committed in using the passive voice during the reporting stage. The delayed CF group received feedback on their passive errors after both tasks were completed. The task only group only performed the tasks without receiving any feedback. The immediate feedback was a hybrid corrective package consisting of a prompt encouraging the learner to self-correct, followed by a recast if the learner failed to self-correct. The prompt was operationalized as a repetition of the wrong utterance with a prosodic emphasis on the passive error, as in “Three dogs were kill?”, and the recast was a reformulation of the erroneous utterance with the rest of the utterance intact, as in “Three people were killed”. For the delayed feedback group, the teacher drew individual students’ attention to the errors they made by quoting the error and asking the student to reproduce the sentence, for example, “Tom, you said ‘three people were kill’. Could you say that again?” The teacher provided the correct version of the sentence if the student was unable to self-correct. An EIT and a GJT, both including 30 items, were used to measure treatment effects. The tests were designed to measure learners’ implicit and explicit knowledge respectively.
It was found that in terms of the GJT scores or explicit knowledge, both feedback groups outperformed the control group at the time of the immediate posttest. However, at the time of the delayed posttest, only the immediate feedback group performed significantly better than the control group, and the delayed feedback condition failed to maintain its effects. In terms of the EIT scores or implicit knowledge, there were no significant differences among the four groups on the immediate posttest or the delayed posttest. Based on these results, the researchers concluded that immediate feedback had an advantage over delayed feedback but the difference was only evident in terms of explicit knowledge, not implicit knowledge.
Approach to replication
One way to replicate the study is to ask the delayed feedback group to perform another task after receiving feedback. The purpose of this change is to verify Li et al.’s explanation for the superiority of immediate feedback over delayed feedback. The authors speculated that the advantage of immediate feedback is primarily because the learners under this condition had an opportunity to apply the knowledge they learned from the feedback immediately in subsequent production. The authors further argued that the requirement for the immediate use of the learned knowledge also means the learners may have been more motivated to attend to the feedback provided by the teacher. In contrast, the learners who received delayed feedback were not required to apply the learned knowledge in production practice, and due to the lack of participatory demands, they probably did not feel the need to engage with the feedback. Adding a third task after providing feedback on the errors learners make in the two preceding tasks will make it possible to verify these speculations. This is because the learners in the delayed feedback condition will have an opportunity to apply the learned knowledge in the following task, and performing another task will provide motivation for attending to the feedback. To make the two feedback conditions comparable otherwise, the immediate feedback group will also perform the third task. If the authors’ hypotheses were correct, then the addition of a third task should even out the differences between the effects of the two feedback types. If the two feedback types are equally facilitative of L2 development, it would be advisable to provide delayed feedback to minimize the disruption immediate feedback may cause to the flow of conversation. An alternative for the revamped design is to provide feedback at the completion of each of the first two tasks under the delayed feedback condition so that the feedback is more spaced and the learners in this condition have an opportunity to apply the knowledge in both the second and third tasks.
A second recommendation for replication is to use a different or additional measure of implicit knowledge. Recall that the TAP theory predicts that immediate feedback will be more effective for implicit knowledge and delayed feedback will be more effective for explicit knowledge. However, the hypothesis was not borne out in Li et al.’s study, which showed that immediate feedback showed an advantage for explicit knowledge but there was no difference between the two feedback types in terms of implicit knowledge. One methodological concern the authors pointed out is that the EIT that was used to measure implicit knowledge was too challenging for this group of young learners and may have failed to capture any change that happened in their implicit knowledge. The test required the learners to listen to some statements read by a native speaker, judge the plausibility of each statement (for the purpose of focusing learners’ attention on meaning), and then repeat the statement in correct English within the allotted time. The test does appear to be demanding, especially for these young learners who were not familiar with the native speaker's accent and who had no experience with this test format. To address this issue, future replication studies may include an oral production task similar to the treatment tasks as a measure of implicit knowledge. Ellis (Reference Ellis2005) reported that L2 learners’ performance on oral production and elicited imitation tasks loaded on the same factor, suggesting that they measured the same construct—implicit knowledge.
Two other suggestions for replication are in order. One is to increase the length of the treatment. In Li et al.’s study, the learners only attended a two-hour treatment. Given that a major objective of the study was to ascertain whether the two feedback types have differential effects on the development of explicit and implicit knowledge, it is necessary to expose learners to longer instructional treatments because the development of implicit knowledge requires extensive and intensive exposure to linguistic input. Another suggestion is to replicate the study in a laboratory setting where the teacher or interlocutor meets with the learners on a one-on-one basis (e.g. Li Reference Li, Wen, Mota and MacNeil2015). It is well-known that the classroom and the laboratory are two different research settings that may generate different results (Li Reference Li2010, Reference Li, Gudmestad and Edmonds2018b). Replicating the same study in a different setting may help us reach firmer conclusions and detect the cause for any discrepancies between the findings of the two settings. To date, there has been no study, either in feedback research or SLA research in general, that replicates the same treatment in a classroom setting and a lab setting.
Study 2: Arroyo & Yilmaz (Reference Arroyo and Yilmaz2018)
Different from Li et al.’s study which was conducted in the classroom and in a face-to-face mode, Arroyo & Yilmaz's study was conducted in a laboratory setting and in text-based synchronous computer-mediated communication (SCMC). Text-based SCMC was operationally defined as communication through online instant text messaging. In the introductory section of the article, the authors identified some benefits of SCMC compared with face-to-face communication. For example, in SCMC learners participate more, they have more processing time, and learners have the opportunity to review foregoing messages. Citing a meta-analysis by Ziegler (Reference Ziegler2016) and other studies (e.g. Yilmaz Reference Yilmaz2012), the authors pointed out that SCMC has been found to lead to more learning gains than face-to-face interaction. Against this background, the authors examined whether immediate and delayed CF affected L2 development differently in SCMC.
The participants were 45 L2 Spanish learners at a US university whose first language (L1) was English. They were divided into three groups of 15: immediate feedback, delayed feedback, and control, and participated in a treatment involving dyadic interaction with a native speaker through text messaging on Skype. The target structure was Spanish adjective-noun gender agreement—a non-salient, challenging structure for L1 English speakers. During the treatment, which lasted 35 minutes, each learner completed an information gap task with the native speaker interlocutor. The immediate feedback group received recasts on their errors in using the target structure, and to maintain consistency with the delayed feedback, they were not given opportunities to respond to the feedback. The learners in the delayed feedback condition received a Word document at the end of the task listing the errors they made during the task followed by the correct versions of the incorrect sentences. They were allowed five minutes to study the list. The control group only completed the task without receiving any feedback. Treatment effects were measured via two tests: a GJT that required the learners to identify and correct errors built in discrete sentences, and an oral production test where the learners were asked to describe paired photos to identify their differences.
The results showed that on the oral production test, immediate feedback was significantly more effective than delayed feedback and no feedback on both the immediate and delayed posttests; delayed feedback did not lead to higher gains than no feedback. On the GJT, both feedback groups outperformed the control group on both posttests, but there was no significant difference between the effects of the two feedback types. The authors concluded that immediate feedback was more effective than delayed feedback. They attributed the larger effects for immediate feedback to the short interval between errors and feedback in this treatment condition; the short interval made it possible to make a direct comparison between the correct and wrong forms. By contrast, the learners under the delayed feedback condition received feedback after completing the task, and it is possible that the error traces had decayed by the time the learners received feedback. Furthermore, like Li et al. (Reference Li, Zhu and Ellis2016), the authors speculated that the finding was probably also due to the lack of practice opportunities for the delayed feedback condition.
Approach to replication
First, the authors speculated that the advantage of immediate feedback might be due to the proximity between errors and feedback. However, the authors expressed an uncertainty over how the learners processed the feedback during online chatting, that is, whether the learners only attended to the most recent feedback or also reviewed previous feedback episodes. Future replications may consider two ways to prevent learners from reviewing earlier feedback. One is to replicate the study in a face-to-face interaction in which it is impossible to review previous feedback given the fleeting and spontaneous nature of this mode of communication. Replicating the study in face-to-face communication also makes it possible to compare the results of immediate feedback with those of other studies using similar methods (Révész Reference Révész2009; Kim, Payant & Pearson Reference Kim, Payant and Pearson2015; Li Reference Li, Wen, Mota and MacNeil2015). Another way is to use a software program that prevents the participant from reviewing earlier messages, as the authors suggested.
A second recommendation is to try another type of feedback, such as the hybrid feedback package utilized in Li et al.’s study, which consists of a prompt that encourages self-correction, followed by a recast in the absence of self-correction. The prompt in the hybrid package makes the feedback more salient than a recast alone and seems particularly useful for a non-salient structure; the second move is a recast which provides positive evidence if necessary. This type of feedback fits all levels of learners and is suitable for both salient and non-salient structures. The hybrid feedback may enhance the effects of delayed feedback by increasing learner engagement, because the interlocutor will work together with the learner instead of asking the learner to read a list of corrections silently, as happened in Arroyo & Yilmaz's study. Alternatively, to overcome the possible lack of learner engagement and to increase the saliency of delayed feedback, the learner may be required to study a list of corrections and discover the regularities based on the feedback. This practice would make the delayed feedback slightly inconsistent with the immediate feedback, thus undermining the internal validity of the study. However, it will increase external validity if we consider the authors’ concern about the practicality of providing immediate feedback in SCMC, hence their suggestion for exploring delayed feedback as an alternative. It is worth pointing out that even though immediate feedback is favored by theorists, it is disfavored by practitioners who worry about its harmful effects on communication flow and fluency. Therefore, researchers have the responsibility to find ways to use delayed feedback without sacrificing learning gains.
A third suggestion is to replicate the study with a different learner group, preferably learners at a lower or higher level, to see whether the impact of feedback timing is contingent on learners’ proficiency level. Another option is to conduct the study with learners of another language to see whether similar findings can be obtained. Note that all future replications should ensure the homogeneity of learner background. The participants of Arroyo & Yilmaz's study were either taking or had taken the Spanish courses offered at the institution, which suggests that the learners were not from the same cohort and that they were varied in their learning experiences and their Spanish proficiency. The researchers did their best to minimize the influence caused by the learners’ varied learning experiences, such as by making sure that their self-rated proficiency scores and their pretest scores for the target structure were not significantly different between the three participant groups. However, the possibility remained that the findings were confounded by extraneous variables.
Finally, some of the recommendations for replicating Li et al.’s study also apply to Arroyo & Yilmaz's study. For example, in order to verify the hypothesis that the advantage of immediate feedback is due to the practice opportunities available in this condition, learners in the delayed feedback condition may be required to perform another task after receiving feedback so as to provide them with an opportunity to apply the knowledge learned from feedback. Also, like Li et al., Arroyo & Yilmaz suspect that the results may change with longer treatments. Therefore, future replications may increase the treatment length.
Conclusion
Given the theoretical and pedagogical significance of the timing of CF, it is important to replicate the only two studies that have examined the differential effects of immediate and delayed feedback. Given that feedback research almost invariably examines immediate feedback, the introduction of delayed feedback extends the scope of research and opens up a new territory in feedback research and in SLA research in general. The methodological aspects of the research such as the way feedback is embedded in a communicative task and the way errors are addressed in the posttask stage add to the repertoire of form-focused strategies that teachers can draw on in their pedagogical practice. Also, teachers need further and more convincing information about the influence of feedback timing in order to make informed decisions about when to address linguistic forms in a task cycle.
Replications may generate results that are similar to or different from those of the original studies, and they are equally valuable. Similar results will increase the generalizability of the findings, and different results will help us identify factors that may moderate the effects of feedback timing. Furthermore, it is necessary to point out that one replication is not enough and that there needs to be a large amount of replication research before conclusions can be drawn about the comparative effects of immediate and delayed feedback and any constraining factors. Given the applied and scientific nature of the field of language learning, replication should become the norm. The need for replication research is felt acutely by research synthesists and meta-analysts when they try to reach conclusions about the effects of an instructional treatment or a relationship that has been subjected to empirical investigation. More often than not, they feel it difficult to meaningfully aggregate the data contributed by the primary research because of the methodological heterogeneity of the synthesized studies and because of the lack of replication studies conducted using the same methods. In a meta-analysis on language aptitude, Li (Reference Li2016) expressed a deep concern over the lack of replication research in the field:
The scattering of studies among subareas of a research domain may partly be driven by the desire to be different from previous research or to be original, but from the perspective of research synthesis, the value of replication research cannot be exaggerated in order for robust conclusions to be reached about a certain construct. (p. 834)
Shafoeng Li is an Associate Professor in Foreign and Second Language Education at Florida State University where he teaches courses in language education and supervises postgraduate students. Before joining Florida State University, he was a Senior Lecturer in Applied Language Studies at the University of Auckland, New Zealand. He received his Ph.D. from Michigan State University. His main research interests include CF, task-based language teaching and learning, individual differences (language aptitude, working memory, motivation, and anxiety), and research methods. His recent and forthcoming publications appear in Annual Review of Applied Linguistics, Applied Linguistics, Applied Psycholinguistics, Language Learning, Language Teaching Research, Studies in Second Language Acquisition, and System.
