How Effective are Recasts in Providing Corrective Feedback?-An Analysis of Recent Perspectives

Inspired by the previous recast literature, this paper aims to investigate whether more explicit recasts are more effective than implicit recasts in second language grammar teaching and learning. Four empirical studies (two classroom-based and two laboratory-based) on implicit and explicit recasts were reviewed. In the two classroom-based studies, recasts were coded as explicit or implicit according to their length, intonation, number of changes, type of change, prosodic emphasis, etc. The effectiveness of recasts was measured by learners’ successful uptake rate and test scores. Conclusion drawn from these two studies is that more explicit recasts (e.g., short, declarative, reduced, one-change) tend to be more effective than implicit recasts in facilitating second language learning. In the two laboratory-based studies, recasts were divided as explicit or implicit mainly according to their intonation and the number of recast moves. Testing instruments included a spontaneous production test, an oral imitation test, and an untimed grammatically judgement test in one study; and an immediate and delayed error identification and correction tasks in the other study. Test scores and error correction rate were collected to measure the effectiveness of different recasts. Results suggest that, in laboratory settings, more explicit recasts might be more effective than implicit recasts or as effective as implicit recasts in grammar teaching and learning.


INTRODUCTION
A great amount of classroom research (e.g., Loewen & Philp, 2006;Mackey, 2012;Sheen, 2006) has revealed that corrective feedback in the way of recasts focuses most commonly on grammar errors. Over the past two decades, the effect of recasts in second language (L2) morphosyntax has attracted unprecedented research attention (Gooch et al., 2016). Defined as "a corrective feedback technique that reformulates the learner's immediately preceding erroneous utterance while maintaining his or her intended meaning" (Ammar & Spada, 2006, p. 545), recasts are considered to be effective in promoting learners' L2 development, since they can provide both positive evidence (by providing the target-like utterance) and negative evidence (by indicating that the utterance is non-target-like) at the same time (Nassaji, 2009). The juxtaposition of these two types of evidence is likely to help learners to notice the discrepancy between their incorrect utterance and the correct utterance.
In spite of these merits, recasts are not without their drawbacks. For example, due to their ambiguity, they are likely to be interpreted by students as a confirmation of the content of their utterance rather than of the form (Lyster, 1998a researchers (e.g., Loewen & Philp, 2006;Nicholas et al., 2001) suggest that this problematic issue might be solved by making recasts more explicit, thus the corrective intention of recasts can be made more salient to the learners. The question then arises: is the effectiveness of recasts positively related to their explicitness? To my knowledge, very few empirical studies have been conducted to investigate this topic. Therefore, I believe this study can further inform the field of second language acquisition (SLA) by providing implications on how to make recasts more effective in classroom teaching. Aiming at investigating the potential relationship between the explicitness of a recast and its effectiveness, the research question which motivates this paper is: whether more explicit recasts are more effective than implicit recasts in L2 grammar teaching and learning. This paper will begin by introducing the theoretical issues related to recasts, as well as the factors influencing their explicitness. It will then review four empirical studies -two classroom-based studies and two laboratory-based studieson implicit and explicit recasts in order to answer my research question. Lastly, it will conclude with the limitations and pedagogical implications on what issues need to be considered when using recasts in classroom teaching.

Implicit and Explicit Corrective Feedback
Corrective feedback (CF) has long been considered an effective way of promoting noticing and thus facilitating L2 learning (Lee & Lyster, 2016;Li, 2010;Mackey, 2007). The value of CF has theoretical support. For example, Long's (1996) "interaction hypothesis" suggests that the feedback learners receive on their erroneous utterance during the negotiation for meaning triggers their selective attention, thus helping them notice the gap between their non-target-like utterance with the target-like utterance. Hence, CF helps to create a favourable environment for L2 development.
Taking the form of responses to the learners' erroneous utterances, CF differs widely in explicitness. Based on Ortega's (2013, p. 75) definition of explicitness -"the perceptual salience (e.g. intonation) and linguistic marking (e.g. by metalanguage) with which the negative information is delivered and thus the corrective intent is made clear to learners", implicit feedback can be perceived as not containing a salient indicator to remind students that their original utterances are incorrect, while explicit feedback does . In a large number of L2 studies (e.g., Ammar & Spada, 2006;Lyster & Yang, 2010;Yilmaz, 2012), implicit feedback has typically taken the form of recasts, whereas the explicit type has been operationalised in a variety of forms, such as explicit error correction and prompts -"a signal of the student errors, but without provision of correct forms" (Gooch et al., 2016, p. 117). Prompts can be executed via clarification requests, metalinguistic information, repetition, and elicitation (Gooch et al., 2016).
Among the different forms of CF used in language classrooms, studies have shown that recasts are the most frequently used Li, 2010;Loewen & Philp, 2006). Explanations of why they seem to facilitate L2 learning are derived from several theoretical frameworks. One oft-cited explanation is Schmidt's (1990, cited in Nicholas et al., 2001) "noticing hypothesis", which claims that to acquire new linguistic items, learners must first notice these new features in the input. By providing an immediate contextualised reaction, which keeps the original meaning of the learners' utterance, recasts attract learners' attention to linguistic form without interrupting the flow of communication (Erlam & Loewen, 2010). This noticing the gap between the incorrect and correct forms is an essential component of SLA (Mackey, 2012). Van Pattern's (1990, cited in Ammar & Spada, 2006) "input processing model" provides another theoretical support for the beneficial role of recasts. He argues that since learners are limited capacity processors, they cannot attend to meaning and form simultaneously. Nevertheless, if the meaning of the input is easy to comprehend, L2 learners may consciously focus on form. Given that recasts provide both positive and negative evidence while keeping the meaning constant, they are considered helpful because they allow learners to attend to form by freeing up learners' attention to meaning. The fact that recasts are "contextualised, semantically contingent upon learners' own utterances, temporally juxtaposed with these utterances, and informationally redundant" (Mackey, 2012, p. 14) makes them the ideal CF technique in theory (Doughty, 2001).
Previous empirical research, however, shows that recasts may not be as effective as other feedback options in positively affecting students' accuracy. For example, Jafarigohar and Gharbavi (2014) examined the effects of recasts and prompts on Iranian learners' grammatical achievement, and they discovered that recasts were inferior to prompts. Similarly, in a study conducted by Ammar and Spada (2006) investigating the benefits of recasts and prompts for students of different English proficiency levels, recasts were found less effective than prompts for low-proficiency learners, whereas these two forms of feedback had similar impact on high-proficiency learners.  also compared the effectiveness of recasts and metalinguistic information, and they concluded that learners benefited more from metalinguistic information than from recasts.
Given the available evidence, it is difficult to generalise findings from one research to other cases. However, the overall trend of results shows that when several different types of CF are compared, the more explicit forms result in greater gains than implicit types (Ortega, 2013). Some researchers (e.g., Nicholas et al., 2001;Yilmaz, 2012) tend to attribute this disappointing finding to the implicit nature of recasts (Jafarigohar & Gharbavi, 2014). They claim that because the corrective purpose of recasts is relatively ambiguous, students tend to interpret a teacher's reformulation as a confirmation of the meaning of their utterance rather than of the linguistic structure (Lyster, 1998a). However, according to , there seems to be some confusion as to whether recasts are merely implicit. Research by Sheen (2006) demonstrates that like CF, recasts differ in explicitness, and therefore different forms of recasts will be introduced in the following section.

Implicit and Explicit Recasts
Traditionally, recasts have been regarded as implicit (Long, 2007, cited in Erlam & Loewen, 2010). Yet, recent classroom-based studies reexamining the role of recasts in SLA suggest that it might be better to conceptualise recasts as "varying on a continuum of explicitness" (Erlam & Loewen, 2010, p. 879). Drawing on several studies that cast a light on the understanding of recasts (e.g., Erlam & Loewen, 2010;Loewen & Philp, 2006;Sheen, 2006), factors that are likely to affect the explicitness of a recast are listed as follows: Firstly, intonation. In general, declarative recasts are considered more explicit than interrogative recasts (Erlam & Loewen, 2010;Loewen & Philp, 2006), since the rising intonation of the latter is likely to invite the learners to interpret them as confirmation checks on the content, rather than on the form, of their original utterances (Lyster, 1998a;Sheen, 2006).
Secondly, intensity of focus. Erlam and Loewen (2010) define intensive recasts as focusing on one or two specific linguistic features; extensive recasts focus on more than two items. Research evidence suggests that intensive recasts seem to be more salient to learners than extensive ones (Lyster, 1998b).
Thirdly, length of feedback. Short recasts, as defined by Sheen (2006), consist of one word or a short phrase with only one content word, while long recasts contain three or more words. Short recasts tend to be more salient than long recasts (Loewen & Philp, 2006) since they pinpoint the non-target-like form for the learners, whereas long recasts may place a burden on learners' working memory (Sheen, 2006).
Another factor is the number of feedback moves. Unlike single-move recasts which consist of only one recast, multimove recasts involve at least two feedback moves in a single teacher turn (Erlam & Loewen, 2010). For instance, Sheen (2006, p. 371) divides multiple recasts moves into three categories: (a) "corrective recasts", including a repetition of the error and a recast, (b) "repeated recasts", comprising more than one recast, and (c) "combination recasts", containing a recast and other forms of CF, except explicit correction. Multi-move recasts appear to be more explicit than single-move ones due to their double focus on the erroneous form (Loewen & Philp, 2006;Sheen, 2006).
In addition, external factors such as social or instructional context (i.e., whether recasts are given in the laboratory setting or in the classroom setting) might have an influence on the explicitness of recasts as well (Ellis & Sheen, 2006;Lyster & Ranta, 2013;Nassaji, 2009;Ortega, 2013). Nicholas et al. (2001) give two possible reasons why a laboratory setting seems to make recasts more salient to learners. On one hand, the dyadic nature of the interactions in the laboratory context might help learners recognise the corrective intent of recasts. On the other hand, the consistency that only one or two linguistic structures have received treatment in the laboratory setting may draw learners' attention to the language forms more easily.
Last but not least, the overall instructional focus of a classroom is likely to determine how explicit recasts can be (Loewen & Philp, 2006;Nicholas et al., 2001). In content-based classrooms, where the primary focus is on the negotiation of meaning rather than form, learners may tend to treat recasts as reactions to the content; while in primarily structure-based classrooms, older learners, adult students in particular, might not have difficulty recognising recasts as corrective (Egi, 2007;Erlam & Loewen, 2010;Nicholas et al., 2001).
Based on the aforementioned factors, it is safe to say that both features inherent to recasts and external factors could influence the explicitness of a recast. However, although different forms of recasts have been distinguished according to their explicitness in recent L2 literature, thus far, very few empirical studies have been conducted to investigate and compare the relative effectiveness of these different forms (Ellis & Sheen, 2006;Nassaji, 2009;Sheen, 2006). Drawing on four of these research papers, the following section will attempt to cast a more definitive light on the issues above as well as aim at answering the research question: are more explicit recasts more effective than implicit recasts in grammar teaching and learning?

Empirical study of Sheen (2006)
By first presenting a classification of the recasts found in L2 communicative classrooms, this study explores how different characteristics of recasts are related to learner uptake and repair.
The database for this research consisted of two data sets: the first one came from two English as a second language (ESL) communicative lessons in New Zealand; and the second one came from two English as a foreign language (EFL) classrooms in Korea. Participants were 24 adult learners in the ESL setting and 10 adult learners in the EFL setting, with their English proficiency levels ranging from intermediate to high-intermediate. Both data sets comprised a total of 12 hours of meaning-based English lessons.
All the recast moves were first identified following Lyster and Ranta's (1997) error treatment sequence: learners' erroneous utterance -teacher's recast -student's reaction to the recast. After that they were coded according to the coding system (see Appendix A). Following the above two steps, uptake and repair moves were then coded. Uptake refers to students' immediate response to a teacher recast (Ammar & Spada, 2006). In this study, if the uptake was successful, which means the initial error had been corrected, it was coded as "repair". Otherwise it was coded as "needs repair", indicating that this uptake still needed to be repaired.
After coding, the database altogether yielded 295 recast moves, including 233 single-move and 62 multi-move recasts. The significance of the relationship between characteristics of recasts and uptake/repair was measured by the uptake/repair rate among all single-move recast moves respectively.
The analysis showed that, of all the characteristics of single-move recasts, length (word-or short-phrase length) and type of change (substitution) were found significantly linked to learner uptake. These two characteristics, along with three additional ones -mode (declarative intonation), reduction (reduced form) and the number of changes (single change) -were significantly related to learner repair. One possible explanation for these findings is that recasts with these characteristics are more explicit, and therefore they tend to be more effective in getting learners' attention than those without such added emphases (Lyster, 1998a).
Despite the relatively small number of recasts, this study provides empirical support for the previous claims in the CF literature (e.g., Nicholas et al., 2001) that declarative, shorter, and reduced recasts produce greater noticing, since they make the corrective intention of recasts more salient to the learners. It further suggests that the more explicit recasts are, the more salient they are likely to become, and the more effective they might be.

Empirical study of Loewen & Philp (2006)
This study examines the relationship between the explicitness of recasts and their effectiveness by investigating whether particular characteristics of recasts are related with successful uptake and accurate posttest scores.
The database for this research consisted of four L2 classroom observation sessions of a language school in New Zealand. For the purpose of this study, classroom activities which were not meaning-based were excluded from the database. The final data involved a total of 17 hours of communicative interaction. Participants were 118 adult learners, whose English proficiency ranged from intermediate to upper intermediate.
The way of identifying recast moves in this research was similar to that of aforementioned Sheen's (2006) study. After the identification process, recasts were coded according to a battery of characteristics, as described in Appendix B. Then uptake moves were coded into four categories: "successful uptake"; "unsuccessful uptake"; "no uptake"; or "no chance", if the learner was not given an opportunity to respond to the recast.
The effectiveness of recasts was assessed by two methods: first, the successful uptake rate following the different forms of recasts; second, the immediate and delayed tailor-made posttest scores. Only 73 of the 118 learners were chosen to take these two post-tests.
The results revealed that certain characteristics of recasts, such as prosodic emphasis (stressed), number of changes (one change), intonation (declarative), number of feedback moves (multi-move), were the best predictors of successful uptake. In other words, the more explicit recasts tended to generate more repair.
The results also showed that recasts with short length, recasts with one change, and recasts with interrogative intonation were likely to result in more accurate test scores. The first two types accorded with the previous claim that short and single-change recasts produced greater noticing (Nicholas et al., 2001). Nevertheless, one surprising finding is that interrogative intonation, other than declarative intonation, was related to correct test scores. One possible explanation is that though declarative recasts are more explicit than interrogative recasts, they might simply encourage the learners to repeat the teacher's target-like utterance.
One limitation of this study is that it does not use pretests to measure learners' previous knowledge, thus it seems impossible to know the learners' original mastery of the linguistic features tested in the two post-tests. Despite this limitation, this study is of importance for L2 recast literature since it puts a grain of salt on findings about recasts that they vary in explicitness, and their differences in explicitness seem to affect their effectiveness.

Empirical study of Erlam & Loewen (2010)
This study examines the effectiveness of implicit and explicit recasts on French non-adjective agreement errors.
Fifty second-and third-year American university students of French were selected as participants, whose proficiency level was assessed by an academic year test. They were allocated to three groups: 19 students for the implicit recast group; 21 for the explicit recast group; and 10 for the comparison group, who received no feedback. In this study, an implicit recast was operationalised as a single recast with rising intonation; an explicit recast was operationalised as comprising two uninterrupted recast moves: a single repetition of the student's error with interrogative intonation, followed by a single recast provided in declarative form. Except the intonation and the number of recast moves, the other variables which may affect the explicitness of recasts, such as intensity of focus and the length of the recasts, were operationalised consistently across the two treatment groups.
During the two half-hour noun-adjective agreement treatment sessions, participants in each group worked on a series of four meaning-focused tasks, which were designed to elicit students' use of the target structure. It should be mentioned that the implicit/explicit recast moves were directed at individual students so as to maximise the likelihood that all participants in the two treatment groups attended to the correction.
A pretest-posttest-delayed posttest design was employed to examine participants' mastery of the target structure at different stages. Each testing occasion consisted of three tests: a spontaneous production test and an oral imitation test to measure learners' implicit language knowledge, and an untimed grammaticality judgement test to measure their explicit knowledge. In addition, two short oral interviews, one after the first posttest and the other after the delayed posttest, were conducted respectively to collect data on learners' perceptions of the CF and their awareness of the target grammatical structure used in the instructional sessions.
Results showed that the type of recasts that learners received in this study had no differential effects on L2 knowledge of the target structure.
Though there are several limitations to this study, such as the difficulty to keep the same number of participants in each group, and the impossibility of controlling their interaction with the target structure outside the treatment sessions, it enhances our understanding of CF. It also calls for further research to investigate whether different types of feedback have different impacts on different grammatical structures, and whether one certain form of feedback has a differential effect on both implicit and explicit L2 knowledge.

Empirical study of Nassaji (2009)
This study identifies several implicit and explicit forms of recasts and examines whether one specific type is more beneficial to L2 learning than the other.
Participants -42 adult intermediate-level learners -were selected from an intensive ESL program at a Canadian university. In this study, implicit recast moves referred to recasts that "reformulated the error within its larger context with no additional intonational or verbal signals to highlight the error, or those that expanded on it with a confirmatory tone", and more explicit recasts referred to those that "isolated the error and reformulated it outside the context with a rising intonation and/or added stress and/or those that combined the feedback with additional more explicit verbal prompts (e.g., "Is that what you mean?") to push the learner further to respond to feedback". Data was collected through four procedures. Firstly, students were asked to sequence four randomly ordered pictures in logical order. After that they were required to submit a written description of the plots the four pictures represented. Then they participated in the treatment session -an oral dyadic interaction, in which they were told to describe the scenario of the pictures orally to the teacher, and their oral descriptions were expected to be as close as possible to the previous written versions. Lastly, learners were given back their written descriptions and were asked to correct the errors. All the four steps were implemented in the same session.
It should be noted that the target structures were not preselected. The same errors arising in an individual's written and oral description, which were given recasts by the teacher were defined as target structures for this particular learner. Hence, each learner had his/her tailor-made target structures.
To determine the long-term effect of implicit and explicit recasts, a delayed tailor-made error identification/correction task -correcting the errors in their original written descriptions -was given to students in two weeks.
Results from both the immediate and the delayed posttests revealed that the more explicit recasts led to a higher rate of error correction than the implicit recasts.
Despite the relatively short treatment session and the limited sample sizes, this research is of great value since it highlights the role of salience as an important feature of recasts. Therefore, it lends empirical support to the argument that the effectiveness of recasts is likely to be determined by their degree of explicitness (Ellis & Sheen, 2006;Loewen & Philp, 2006;Nicholas et al., 2001;Ortega, 2013).

FINDINGS AND DISCUSSION
This section seeks to answer my research question, which is, whether more explicit recasts are more effective than implicit recasts in L2 grammar teaching and learning. However, due to the distinct nature of classroom-based settings and laboratory-based settings (i.e., the latter seems to make linguistic structures more salient to learners), the two classroom-based studies and the two laboratory-based studies will be discussed separately.

Discussion and Findings of Classroom-based Studies: Sheen (2006) and Loewen & Philp (2006)
These two studies both investigated the effectiveness of different forms of recasts in L2 adult classrooms. As Table 1 illustrates, they had a lot in common.
Firstly, the way they implemented the research was similar: data came from L2 classroom observations; all classroom activities were meaning-focused; all the participants were adult learners, whose average English proficiency was intermediate to high-intermediate level; target structures were selected from "learner error -teacher recast -student reaction" interaction immediate after the instructional sessions, other than pre-selected, since incidental focus on form is not predictable in classroom settings (Loewen & Philp, 2006). In addition, the way the three researchers identified and coded different forms of recasts was similar (see Appendix A and B). As for the test instruments, they both used the learners' repair/successful uptake rate as an indicator of the effectiveness of different forms of recasts, while in Loewen and Philp's study, another test instrumentimmediate and delayed posttest -was adopted.
The relatively high similarity of these two studies made it possible to compare their findings together. They appeared to generate similar results as well.
They both suggested that recasts varied in explicitness. As such, they challenged the commonly held view in the L2 recast literature (e.g., Doughty, 2001) that recasts are implicit in nature.
Furthermore, they attested that the explicitness of recasts may affect their effectiveness. To be more specific, the more explicit recasts (e.g., short, declarative, reduced, onechange) tend to generate higher rate of successful uptake, and in Loewen and Philip's study, higher test scores as well.
Based on the viewpoint that successful uptake may facilitate acquisition , it seems safe to conclude that more explicit recasts tend to be more effective in facilitating L2 learning.
Based on the analysis of these two classroom-based studies, the answer to my research question is: in classroom settings, more explicit recasts tend to be more effective than implicit recasts in grammar teaching and learning.
However, it should be noted that, though both studies concluded that explicit recasts resulted in more uptake, they also suggested that the explicitness of a recast did not necessarily guarantee its effectiveness. Other factors, such as learner factors (Sheen, 2006) and learners' developmental readiness (Loewen & Philp, 2006), determined the extent to which students might benefit from recasts as well.
Evidence has shown that unlike classrooms, laboratory settings make linguistic structures more salient to learners (Ellis & Sheen, 2006;Lyster & Ranta, 2013;Nassaji, 2009;Ortega, 2013). The following two studies will examine if laboratory-based studies generate the same answer to my research question.

Discussion and Findings of Laboratory-based Studies: Erlam & Loewen (2010) and Nassaji (2009)
By comparing the effects of implicit recasts and explicit recasts in conversational interaction, these two studies aim to find out which one is more effective in L2 grammar learning and teaching. Unlike the above two classroom-based studies, these two laboratory-based studies were conducted in a relatively different way.
As can be seen from Table 2, in Nassaji study, the adult learners' L2 language proficiency level was intermediate, while Erlam and Loewen did not provide information about their adult learners' target language proficiency. The target structures in Nassaji study were tailor-made, whereas in Erlam and Loewen's research, the target structure was preselected. Another difference is the way implicit/explicit recasts were conceptualized. Compared to Nassaji, Erlam and Loewen defined recasts into a narrower scope. In addition, the effectiveness of different types of recasts was measured using different test instruments.
These two studies concluded with different results. Nassaji found that more explicit recasts resulted in a higher proportion of error correction, which indicated that more explicit recasts appeared to be more facilitative in grammar learning. Erlam and Loewen did not find difference between the effectiveness of implicit/explicit recasts.
The result of Erlam and Loewen's study seems to go against with that of Nassaji study. Nevertheless, they are not surprising if we take into account the learners' perceptions regarding the feedback and the target structure used in the treatment sessions. In Erlam and Loewen's research, a similarly high proportion of students in either the implicit or explicit recasts group realised that they had been given feedback during the treatment sessions. This means that the explicitness of the two types of feedback, though operationalised differently by researchers and were supposed to be a variable in this research, appeared to be similar for the two groups. Erlam and Loewen also gave us two possible explanations for the similar awareness of the feedback: Firstly, the laboratory setting, where the intensive focus on errors related to the target structure was likely to attract learners' attention to the target linguistic form, and thus helped participants to recognise the corrective intent of recasts. The fact that students in both treatment groups had similar level of awareness of the target structure provided evidence for this explanation.
Secondly, the fact that the control group also made gains without receiving any CF indicated that not only the CF, but the communicative tasks used in the interaction sessions might facilitate learning. These tasks might have met the criterion of "task-essentialness" (Loschky & Bley-Vroman, 1993, p. 138), in the sense that they were designed to elicit the target structure, and they cannot be completed without the target structure.
According to Erlam and Loewen's explanation, it seems not unreasonable to hypothesise that, theoretically, explicit recasts were expected to be more effective than implicit recasts in this study. However, this hypothesis can only be attested by more research.
In conclusion, based on these two laboratory-based studies, the answer to my research question is: in laboratory settings, more explicit recasts might be more effective than implicit recasts or as effective as implicit recasts in grammar teaching and learning.

LIMITATIONS AND FURTHER RESEARCH
Apart from the fairly small number of empirical studies reviewed, analysis of said studies did not provide enough evidence to fully answer my research question due to several limitations: Firstly, findings about students' L2 development in these four studies might be interfered with by a variety of variables, such as target structures and teachers' instructions. Ellis (2007) claims that the choice of target structure has an impact on the effectiveness of the CF treatment, hence it is not easy to distinguish whether the different findings of these four studies were caused by different forms of recasts only or by the target structures as well. Moreover, in all the four studies, recasts were used in combination with form-focused instruction (FFI). Since FFI also promotes L2 development (Lindseth, 2016;Lyster, 2004;Nassaji, 2013), the benefits found for recasts cannot be attributed to recasts alone but to the conjunction of recast and FFI (Nassaji, 2009). Further research examining the effects of recasts without FFI is needed.
Secondly, it is not easy to compare across these four studies due to their differences in the design and implementation of the research. For instance, they used different ways to define and operationalise implicit and explicit recasts. In other words, implicit/explicit recasts used in these studies were not equivalent in their degree of explicitness. In addition, they used different test instruments to assess the effectiveness of recasts. As such, it seems unlikely to compare their findings together.
Lastly, participants in all the four selected studies were adult learners, with the average English proficiency level being intermediate. In addition, the sample size was relatively small in each study. Due to the fact that students' language proficiency levels and their ages might have influence on the research results (Ammar & Spada, 2006), more large-scale and long-term research targeted at learners of different ages and degrees of language proficiency are needed in order to get a more accurate answer to my research question.

IMPLICATIONS
Considering that teachers' understanding about recasts might affect the extent to which learners can benefit from this most frequent CF type, the topic of my paper is of great importance to L2 classroom teaching and learning. In spite of the aforementioned limitations, the analysis of these four empirical studies generates several pedagogical implications for classroom teachers on how to use recasts more effectively.
Teachers should first realise that like CF, recasts also vary on an explicitness continuum. The more explicit recasts tend to be more effective in facilitating L2 grammar learning than implicit recasts. As such, teachers can reduce the ambiguity of recasts by increasing their explicitness, such as adding stress, shortening the length, and targeting a single structure.
Nevertheless, it should be remembered that the effectiveness of recasts is not only related to their explicitness, but also determined by learner factors. Kim and Han (2007) state that learner's interpretation of recasts does not always overlap with teachers' intent. To reduce the possibility that students interpret recasts as feedback on content of their original utterance, it is advisable for teachers to combine recasts with other types of feedback which can push them to self-repair. On one hand, multi-move recasts appear to be more effective due to their double focus on the erroneous form (Loewen & Philp, 2006;Sheen, 2006). On the other hand, since multimove recasts cover more feedback techniques, they seem to have another advantage -some linguistic structures might not be amenable by recasts (Ellis, 2007) but by other forms of feedback.

CONCLUSION
This research paper aimed at examining the relationship between the explicitness and the effectiveness of recasts in order to answer the question: to what extent does the explicitness of a recast influence its effectiveness in grammar teaching and learning? The analysis conducted in this paper is of importance since very few empirical studies have investigated this topic. The overall trend of the findings from the four empirical studies on implicit and explicit recasts suggests that the latter might be a better choice than the former in L2 classrooms. This result not only challenges the commonly held view that recasts are implicit in nature, but also provides evidence that recasts vary in explicitness, and their differences in explicitness seem to affect their effectiveness.
However, it should be noted that the effectiveness of recasts is not only determined by its inherent characteristicsexplicitness, but also by other external factors, such as learner factors, communicative interaction, and tasks. It also comes up with several pedagogical implications on how to use recasts more effectively in classrooms, such as increasing their effectiveness by combining recasts with other types of feedback.
In order to further solidify these notions and reach a more definite conclusion regarding the effectiveness of recasts, however, further research will need to be conducted. As has been noted, the number of papers analyzed here is limited to four, and as more research is made available, further and deeper analysis should be carried out to continue testing the effectiveness not only of recasts themselves in each individual form but of CF in general.