Looking through key SLA journals (Applied Linguistics, Language Awareness, Language Learning, Language Teaching Research, Studies in Second Language Acquisition, TESOL Quarterly), one cannot help but be struck by the sheer number of articles that address recasts, especially in the last few years. Since Nicholas, Lightbown, and Spada's (2001) review article on recasts, there have been more than 12 recast articles published in these journals, more than on any other single topic. Most of these (Braidi, 2002; Hauser, 2005; Iwashita, 2003; Leeman, 2003; Mackey, Oliver, & Leeman, 2003; Panova & Lyster, 2002; Philp, 2003; Sheen, 2004) have been cross-sectional descriptive studies that document the use of recasts in different settings and under different participatory structures (e.g., native speaker [NS]-nonnative speaker [NNS] as opposed to NNS-NNS interaction), often also examining the learners' uptake of recasts. Three (Han, 2002; Ishida, 2004; Nabei & Swain, 2002) were longitudinal. Two (Ayoun, 2004; Lyster, 2004) were experimental studies designed to investigate the relative effectiveness of recasts as opposed to other instructional strategies on acquisition. Altogether, these articles demonstrate that recasts currently constitute an area of inquiry of great interest to SLA.
Recasts occur in episodes of this type1
All examples of recasts in this article, unless otherwise indicated, are taken from the Korea EFL database used in Sheen (2004).
A learner utterance that contains (or is perceived as containing) an error (line 2) triggers the recast, which reformulates the utterance in accordance with the interlocutor's own linguistic norms (line 3). Optionally, the learner might also respond to the recast in some way (commonly referred to as uptake), sometimes repairing his or her original error and sometimes not [e.g., the final turn in (1) constitutes an example of uptake without repair].
Recasts are just one of several possible corrective strategies that teachers employ to deal with learner errors. Lyster and Ranta (1997) identified five corrective strategies other than recasts (i.e., explicit correction, clarification requests, metalinguistic information, elicitation, and repetition), whereas Panova and Lyster (2002) added one more—translation. Although these other strategies have received attention from researchers, none of them has received the same intensity of attention as recasts.
One reason is that recasts generally occur with great frequency in interactions with second language (L2) learners, especially if they occur inside a classroom.2
Recasts are not invariably frequent. Nabei and Swain (2002) reported that there were only 25 teacher error-treatment episodes (of which 23 involved recasts) in 420 min of recording time. The explanation given for this low level of corrective feedback was that the teacher was primarily oriented to meaning and that she was very familiar with Japanese students.
There are other reasons for researchers' interest in recasts. These are theoretical. Studying recasts serves as a means of investigating two issues of general theoretical importance in SLA. The first issue concerns the roles of positive and negative evidence in SLA. The position adopted by strict generativists such as Schwartz (1993) is that negative evidence might play a role in the development of learned linguistic knowledge (i.e., explicit knowledge of the L2) but not in the acquisition of competence (i.e., implicit knowledge of the L2), which relies entirely on positive evidence. Schwartz considered learned linguistic knowledge and competence to be distinct and unconnected. However, other theories (e.g., skill-building theories of the kind advocated by DeKeyser, 1998), have claimed that the two are connected and that knowledge that begins in declarative form is gradually proceduralized and restructured as a result of, among other types of learner experience, access to negative feedback. Clearly, a carefully designed study that investigates the effects of recasts along with other forms of corrective feedback on a developing L2 can serve as a means of testing the claims of these rival theories—assuming, of course, that recasts can be legitimately viewed as a source of negative evidence and that assessment devices exist for distinguishing performance based on learned linguistic knowledge and competence. As we will see, there have been a small number of studies that have attempted this.
The second theoretical issue that has the potential to be addressed by research on recasts concerns the relative impact of implicit and explicit types of negative feedback. Recasts are viewed as an implicit strategy. Explicit correction, as illustrated in (2), directly signals to the learner that an error has been committed.
There are two related questions of importance here: (a) Can explicit types of feedback contribute to the development of learners' competence (implicit knowledge) or do they simply feed into greater metalinguistic awareness (explicit knowledge)? and (b) If both explicit and implicit types of feedback can contribute to the development of learners' competence, is one type of feedback more effective than the other? Long (see Long, 1996; Long & Robinson, 1998) viewed recasts as implicit and claimed that they work for acquisition precisely because they are implicit. He considered them effective because they induce a joint focus on form and meaning, thereby encouraging form-function mapping in the context of, and without disturbing, the communicative flow of the interaction. Furthermore, they allow for cognitive comparison of erroneous and target language forms in a context in which the learner is primed to notice the difference. In contrast, explicit forms of correction involve treating language as an object and interrupting the flow of communication and, thus, will not assist form-function mapping; for these reasons, Long considered them unlikely to promote acquisition.
Other researchers (e.g., R. Ellis, 1993; Swain, 1995), however, have argued that raising learners' consciousness by means of explicit techniques can contribute either directly or indirectly to interlanguage development. It would follow from such a position that explicit correction of learner errors is effective, possibly more so than implicit techniques (such as recasts), which always run the danger of not being perceived as corrective in purpose.
For these reasons, recasts have been a focus of attention in SLA research. Conveniently, they are readily available for study because of the large number of recasts to be found in interactions that involve learners. Recasts, however, present a number of problems that need careful consideration. The main purpose of this article is to examine these problems.
WHAT IS A RECAST?
Nicholas et al. (2001) concluded their review of the recast literature by suggesting that L2 researchers, in contrast to first language (L1) researchers, have failed to solve definitional differences. This is perhaps the most obvious problem because it makes it very difficult to compare the results obtained by different L2 studies of recasts, given that often enough these studies were not looking at the same thing. For example, Doughty and Varela (1998) conducted an experimental study of what they called corrective recasting, which consisted of two moves: (a) a teacher repetition of a learner's error, with emphasis placed on the erroneous word(s), and (b) a reformulation of the complete learner utterance, as in (3).
Compare this recast with an example of a recast that occurs naturally in the context of a communicative activity, as in (4).
In one respect, these recast episodes are similar in that they are both complex; that is, they involve more than one feedback move. In (3), a repetition precedes the recast move, whereas in (4), the recast is repeated. In other respects, however, they are very different. In (3), the erroneous and corrected forms are highlighted by means of emphasis, but in (4), they are not. In (3), the illocutionary force of the recast is self-evidently that of correcting the learner, as reflected in the learner's uptake. In (4), the illocutionary force is not so self-evident. The teacher's turn could have been easily understood as a confirmation check and, judging from the learner's response, was in fact so taken. It is not at all clear that we are looking at the same discoursal phenomenon in these two examples.
The literature on recasts is in fact replete with a whole host of terms that describe the different kinds of recasts found to occur. Farrar (1992, p. 92), in a study of recasts in child language acquisition, distinguished between corrective recasts (“a recast that corrects a target error”) and noncorrective recasts (“a recast that does not correct a target but models a target”). In SLA research, Lyster and Ranta (1997) used the same terms but defined noncorrective recasts as reformulations of learners' error-free utterances. Another distinction of potential importance is that between full recasts and partial recasts. All of the preceding examples are full recasts (i.e., they involve a reformulation of the whole erroneous utterance). Example (5) illustrates a partial recast, in which only the part of the erroneous utterance containing the error is repeated.
Recasts can also be single or multiple; sometimes the interlocutor repeats the recast, as in (6).
As Braidi (2002) noted, recasts can also occur in one-signal negotiated interactions or in extended negotiated interactions. The given examples all illustrate one-signal negotiated interactions, but extended negotiated interactions can also occur quite frequently (they constitute more than 50% of Braidi's data). Recasts also differ in terms of whether they are simple or complex (Nelson, Denninger, Bonvillian, Kaplan, & Bake, 1983), which depends on whether the changes to the learner's erroneous utterance are minimal or substantial (see also Philp, 2003) and on the nature of the change—that is, whether it entails a substitution of the erroneous form, an addition, a deletion, or a reordering of the trigger utterance. In short, recasts are chameleonlike.
To what extent, therefore, can all recasts be legitimately considered a single genus? There are two ways to answer this question. One is to treat them as a single genus and deal with the variations in terms of a coding system that distinguishes their major characteristics. The other approach is to acknowledge that the different types are in fact so distinct as to require a reclassification of recasts into a number of separate categories. In essence, this problem is very similar to the one facing biologists who seek to classify fauna and flora in a principled manner.
In fact, though, the research to date has often fudged the whole problem. Often, researchers operate with a general definition of recasts, a sample of which are shown in Table 1. However, with a few exceptions (e.g., Braidi, 2002; Loewen & Philp, in press; Philp, 2003; Sheen, 2006), these researchers have made no attempt to investigate subcategories of recasts in accordance with their differentiating characteristics. Lyster and Ranta (1997), in their often-cited study, simply counted recasts, although Lyster (1998b), in a further analysis of the same data, distinguished four types of recasts. Even when researchers acknowledge that recasts are variegated, they are likely to generalize about their role in the acquisition of an L2 as if they possess a homogeneous identity.
Definitions of recasts
The definitions in Table 1 are subtly but significantly different. Neither Lyster and Ranta's (1997) nor Braidi's (2002) definitions made reference to the teachers' and learners' orientation to the discourse—that is, whether the primary focus of attention is on language as an object or on message-conveyance. Given the contexts in which their studies were carried out (immersion classrooms and task-based interaction), it can be assumed that the primary focus was on the message, although it would seem likely that, even in these contexts, some repair sequences occurred that were motivated by form rather than message. Indeed, as they stand, their definitions of recasts would permit the inclusion of reformulated utterances from interactions that arise in traditional, form-focused lessons. Long's (1996, 2006) definitions seek to exclude such form-focused reformulations; to qualify as a recast, the reformulation must occur in the context of message-centered communication. There is, however, an interesting difference between the two Long definitions. Long's (1996) definition states that a recast rephrases an erroneous learner utterance “while still referring to its central meaning” (p. 434).3
Hauser (2005) raised a different objection regarding the way recasts have been defined and coded. He pointed out that definitions such as Long's (1996) make reference to recasts that maintain the meaning of the learner's initial utterance. He suggested this is problematic because meaning, whether viewed as propositional content or action, is not established by the learner's initial utterance but, rather, is “open to negotiation” and “emerges through the interaction” (p. 310); he also illustrated how this takes place. He concluded that coding practices based on the idea of maintaining meaning “obscure what is happening in the interaction” (p. 310). However, Lyster and Ranta's (1997) and Braidi's (2002) definitions make no mention of meaning; they are based on purely formal criteria, namely that a recast (a) reformulates and (b) corrects a preceding learner utterance. Thus, Hauser's critique of the coding practices of recast studies seems misplaced.
An anonymous SSLA reviewer pointed out that what is crucial is the focus of the interlocutor receiving the recast, rather than the provider. Thus, whether the recast is communicative or didactic in purpose matters less than how the learner orientates to the recast. We would agree, and we address the whole question of orientation later in the article. One of the problems with definitions of recasts such as Long's (1996, 2006) is that they do not take account of the participants' orientation—or of the very distinct possibility that the orientations of the provider and receiver of the recast might not match.
To clarify this important point, let us consider the examples in (7) and (8).
In (7), it would appear that the teacher had no difficulty understanding what the learner meant to say but, nevertheless, offered a recast that corrected the learner's grammatical error. The teacher's reformulation clearly refers to the central meaning of the learner's utterance, but rather than focus on meaning, it momentarily treats language as an object. It can be surmised that Long, in accordance with his later definition, would not wish to consider the reformulation in (7) a recast. In contrast, the teacher's reformulation in (8) appears to be motivated by an attempt to understand what the learner meant. In this episode, the interlocutors seem to be focused on meaning throughout; thus, the reformulation constitutes a recast in Long's later terms. The distinction between (7) and (8) is crucial for Long because, for him, recasts must necessarily be implicit and must not interrupt the communication. Arguably, the reformulation in (7) is not implicit, and given that it treats language as an object—albeit momentarily—it constitutes an interruption in the flow of communication. It should be noted that the claim that recasts are implicit and that they must not interrupt the flow of communication constitutes two distinct claims. It should also be noted that it is possible for recasts to be viewed as implicit without also insisting that they do not interrupt the flow of the communication. Such would appear to be Lyster's (1998b) position.
To decide whether “throughout the exchange the focus of the interlocutors is on meaning not language as an object” (Long, 2006), it is necessary to determine the interlocutors' intentions. Is the reformulation didactically motivated or is it motivated by a need to clarify what is being said? In (7) and (8), the teacher's intention can be determined with reasonable certainty by inferring it from the surrounding discourse, but in many cases, this is simply not possible. In (9), for example, it is not possible to tell whether the teacher is trying to clarify the learner's meaning or is simply correcting him.
It might be possible to ascertain the interlocutors' intentions by means of stimulated recall (Gass & Mackey, 2000), but this would pose an enormous burden on researchers, nor could intentionality be established retrospectively with any reliability. In short, definitions that rely on interpreting the interactants' intentions are highly problematic.
Perhaps the best solution to these problems would be to work with a very general definition of recasts, along the lines of Sheen's (2006) definition. This type of approach recognizes that the recast must occur in the context of an effort to communicate (as opposed to some activity of the focus on forms type) but makes no attempt to determine whether the recast is didactically or communicatively motivated. It would then be possible to subcategorize recasts into distinct types depending on clearly distinguishable formal characteristics (e.g., length, number of changes, and type of change), as in Lyster (1998b). Lyster recoded the recasts from Lyster and Ranta (1997) in terms of four types: isolated declarative recasts (a reformulation in an utterance with falling intonation and no additional meaning), isolated interrogative recasts (a reformulation in an utterance with rising intonation and no additional meaning), incorporated declarative recasts (a reformulation in an utterance with falling intonation and additional information), and incorporated interrogative recasts (a reformulation in an utterance with rising information and additional information). What becomes crucial in developing such a taxonomy is determining on the basis of which features recasts should be classified as different types. Given that there is currently no available theory to guide the development of a taxonomy of recasts, the identification of relevant types will have to be undertaken empirically. One way to accomplish this is by exploring the relationship between different types of recasts and successful uptake (as in Lyster, 1998b, and Sheen, 2006). A better way is by examining what effect the different types of recasts have on acquisition (as in Loewen & Philp, in press).
HOW IMPLICIT ARE RECASTS?
The prevailing view in the recast literature is that recasts constitute an implicit form of negative feedback. Lyster (1998a) referred to their “function of implicitly [italics added] providing a reformulation” (p. 59), whereas Long (2006) asserted unequivocally that “a recast is a discourse move that is by definition implicit [italics added].” However, what exactly does implicit mean here? Does it mean that there are no overt linguistic signals that a recast is corrective in nature? Or does it mean that the learner who receives the recast remains unaware of the corrective nature of the recast? In either case, there are problems with the definition.
If we adopt a purely linguistic perspective, we need to decide which linguistic signals denote explicitness and which ones denote implicitness. This is no easy task. For example, it might be argued that corrective recasts of the kind used in Doughty and Varela's (1998) study [see (3)] contained clear signals that made the corrective force of the recast explicit. First, the teacher repeated the student's erroneous utterance, stressing the deviant parts. Second, the teacher reformulated the utterance, again stressing the corrected items. This procedure was specifically designed to make the corrected items salient to the learners. In so doing, it arguably made the corrective nature of the recasts quite explicit. Doughty and Varela's recasts were in fact didactic rather than communicative in nature. Didactic recasts, in general, are likely to be quite explicit. Consider again (5)–(7), all of which would seem to contain didactic recasts and all of which made the corrective nature of the recast quite explicit. In (5), the explicitness derives from the fact that the recast was a single word, spoken with a falling intonation, which signals that a statement rather than a question was intended. In Lyster's (1998b) taxonomy, this is an example of an isolated declarative recast. In (6), the teacher made it clear that he (the teacher was, in fact, male) was correcting the learner by repeating the recast twice. In (7), the teacher used emphatic stress to signal that he was making a correction. Arguably, recasts should not be viewed as necessarily implicit but, rather, depending on the linguistic signals that encode them and the discoursal context, they should be taken as more or less implicit or explicit. In other words, recasts can lie at various points on a continuum of linguistic implicitness-explicitness. However, precisely which linguistic and discoursal features distinguish implicit and explicit recasts remains to be discovered.
An alternative approach is to adopt the receiver's perspective and attempt to establish how learners view the recasts to which they are exposed. There are two ways in which this might be undertaken. One way is by examining to what extent the recasts result in greater metalinguistic awareness. The reasoning that underlies this approach is that if learners treat recasts explicitly as corrections, they are more likely to reflect on them and develop some conscious representation of the underlying pattern or rule. Conversely, if they respond to them implicitly, they might internalize but not demonstrate awareness of the pattern or rule. It is also possible that some learners who respond to recasts implicitly will neither become aware nor internalize. Long et al.'s (1998) study provided evidence to suggest that learners do in fact develop metalinguistic awareness as a result of exposure to recasts. All eight students who learned Spanish adverb word order (i.e., verb-adverb-object) through recasts were able to explicitly and correctly formulate an explanation of the rule. If we adopt the learners' perspective, it would appear that the recasts were responded to explicitly, although it is possible that the explicit representation of the rule was derived after the actual treatment. Further evidence of metalinguistic awareness through exposure to recasts comes from Han's (2002) study of eight ESL learners, as evidenced in their self-correction behavior. Learners who received the recasts were more likely to self-correct past tense errors, the target of the recast treatment, than a control group. This might be seen as evidence of the fact that they had formed an explicit rule. It can be argued that learners only achieve this kind of metalinguistic awareness from recasts if they are treated as explicit feedback. Indeed, the fact that the learners received frequent, intensive recasts directed at a single structure (as in Doughty & Varela's, 1998, and Han's studies) encouraged the treatment of recasts as explicit feedback. It might also be argued, however, that learners with well-developed metalinguistic knowledge (such as adults who have experienced traditional form-focused instruction) are more likely to perceive recasts as explicit correction than learners with less well-developed metalinguistic knowledge (such as children learning in an immersion context). Sheen (2004) showed that adult learners in EFL and ESL classes produced higher levels of successful uptake following recasts than children in immersion classrooms, a finding that might reflect differences in these learners' metalinguistic knowledge as well as the extent to which they perceived the recasts as explicit.
A second way of investigating how learners respond to recasts is to ask them directly. Using stimulated recall, Mackey, Gass, and McDonough (2000) investigated L2 learners' perceptions of feedback moves, including recasts. This study found that feedback on morphosyntax was seldom perceived as being about morphosyntax, whereas feedback on phonology and lexis was perceived more accurately. Feedback on morphosyntax typically took the form of recasts, whereas feedback on phonology and lexis involved other strategies (e.g., requests for clarification). This study thus indicated that learners were not aware of the corrective force of the recasts and did not attend to the reformulated morphological features, which suggests evidence of the implicit nature of the recasts. It should be noted, however, that the recasts were not focused on any specific grammatical target in this study and, judging from the examples provided, the recasts were predominantly communicative rather than didactic in nature.
Determining the implicitness or explicitness of recasts in terms of learners' perceptions of them forces us to address another problem: whether the reformulated feature in the recast is noticed by the learner and whether noticing implicates explicitness. Noticing implies at least some degree of conscious attending (Schmidt, 2001); however, it need not imply that the learner has treated the recast as a correction. Learners can carry out the kind of cognitive comparison (R. Ellis, 1993) that has been hypothesized to foster acquisition while continuing to view the recast communicatively (e.g., as a request for confirmation or simply as a conversational lubricant); that is, noticing can occur irrespective of how the learner interprets the illocutionary force of the recast. Of course, it remains a matter of interest whether the noticing that occurs when the recast is responded to as implicit negative feedback is more or less effective in promoting acquisition than the noticing (and concomitant metalinguistic understanding) that arises when the recast is responded to as explicit negative feedback. To the best of our knowledge, there is no study that has attempted to address this crucial question. This is another reason why we should not reject recasts that didactically involve a time-out from communicating. It might be that such recasts are precisely the ones that are most useful for acquisition.
To sum up, recasts cannot be viewed as a purely implicit form of negative feedback. In many cases, their illocutionary force as corrections is quite transparent and, therefore, they should be seen as a relatively explicit form of negative feedback. Again, what is important is to resist theorizing about recasts as if they are homogeneous and monolithic and, instead, to acknowledge the diversity of their form and function and of learners' responses to them.
DO RECASTS SUPPLY POSITIVE OR NEGATIVE EVIDENCE?
Positive evidence consists of samples of what is grammatical or acceptable in the L2; it arises naturally in communication. Negative evidence is information about what is ungrammatical or unacceptable; it is available not only through feedback but also through explicit grammar teaching. We have already seen that the whole question of the role of negative evidence in SLA is controversial (see Carroll, 2001, and Long, 1996), but this is not our current concern. Here we consider whether recasts constitute a source of negative evidence.
Nicholas et al. (2001) considered that it is not clear whether recasts are to be viewed as providing negative evidence or just positive evidence. Whether recasts afford positive or negative evidence is again tied up with how learners interpret their illocutionary force. If learners have no conscious awareness that the recast is intended to be corrective, then recasts are to be seen as providing only positive evidence. Conversely, if they are interpreted as corrective, then they constitute a source of negative evidence. Thus, recasts can constitute (a) positive evidence only or (b) both negative and positive evidence. As we have seen, how learners interpret recasts is, in part, a function of the formal signals contained in the recast and the discoursal context. It is also, as Sheen (2004) pointed out, a matter of their orientation to the interaction—whether they are functioning as learners, who treat language as an object to be studied, or as language users, who treat language as a tool for communication. The general question of the learner's orientation is considered in a later section.
Leeman (2003) designed a study to investigate whether recasts work for acquisition as positive or as negative evidence. She compared the learning that resulted from three treatments: (a) recasts, (b) negative evidence (i.e., indicating the source of a problem but without correction), and (c) enhanced salience (i.e., using stress and intonation to make the target form salient). The recasts in this study were designed to provide only implicit negative evidence about noun-adjective gender agreement (e.g., una taza roja “a cup red-fem”). Each recast consisted of a comprehension signal, a targetlike reformulation of the erroneous element of the learner's utterance without any emphasis, and an immediate topic continuation move, as in (10). Such recasts can claim to be implicit on the grounds that they do not include any of the explicit linguistic signals discussed previously.
Leeman found that only the recast and enhanced salience groups outperformed the control group on the posttests; that is, no benefit was found for simply indicating that an error had been committed. Furthermore, the recasts and input with enhanced salience had very similar effects. Leeman concluded that recasts facilitate acquisition because they provide positive evidence, not because they constitute negative evidence.
However, we cannot conclude from Leeman's (2003) study that all recasts work for acquisition because they afford positive evidence. Leeman designed her recasts so that the corrective force of the recasts was hidden. It is possible that if the recasts had been more explicitly corrective, they would have constituted negative evidence and resulted in greater acquisition than the enhanced input. As we will see, other studies (e.g., Carroll & Swain, 1993) found that explicit forms of negative feedback led to greater acquisition than implicit forms. Thus, whereas Leeman's study demonstrated that recasts can assist acquisition simply by providing positive evidence, it certainly did not demonstrate that this is the only way or even the most efficient way to do so.
To sum up, the answer to the question of whether recasts supply positive or negative evidence is that, depending on their characteristics, they can constitute positive evidence alone (i.e., their corrective force is not recognized by the learner) or both positive and negative evidence. Although there is evidence that recasts affording just positive evidence can facilitate acquisition, it remains possible (but not yet demonstrated) that recasts that supply both positive and negative evidence are even more effective.
DO RECASTS FACILITATE ACQUISITION?
A number of researchers have queried whether recasts in general are an effective means of promoting acquisition. The doubts stem from the problems that learners are believed to face in identifying the corrective force of recasts (i.e., in viewing recasts as providing negative evidence). Given the multifunctional nature of recasts, learners are likely to have difficulty determining how to respond to them. Lyster (1998b) argued that the acquisitional value of recasts is diminished by the fact that recasts are inherently ambiguous, such that their reformulating, corrective function is not transparent. The problem is made worse by the fact that interlocutors also repeat learners' correct utterances, so learners can never be sure whether their interlocutors are echoing what they have said as a way of supporting their conversational efforts or in order to provide them with negative evidence. Han (2002) noted that “recasts are among the least clear and direct forms of negative feedback” (p. 550). Mackey et al.'s (2000) study, referred to previously, provided evidence that learners do indeed often fail to perceive recasts as corrections, especially when morphosyntactical reformulations are involved.
It should be noted, however, that the problem of identifying the corrective function of recasts does not negate their acquisitional potential. First, as Leeman's (2003) study showed, recasts that supply only positive evidence can have a positive effect on acquisition. Second, learners have been shown to identify the corrective force of a fair number of recasts, including those that address morphosyntax, and to notice the features that have been reformulated. As a number of studies (e.g., Doughty & Varela, 1998; Han, 2002) have shown, the ambiguity of recasts can be reduced by ensuring that they focus on a single linguistic feature and that their corrective force is linguistically signaled by, for example, the use of emphatic stress on the target item. It might also be argued that the ambiguity of recasts is actually an advantage, as learners clearly need to be able to not pay attention to linguistic form at least some of the time in order to focus on message content. The multifunctional nature of recasts gives learners the choice of whether to focus on their communicative or linguistic content. Good language learners focus on form and meaning, switching from one to the other within the course of a single interaction (Lennon, 1989). Recasts cater to this.
Another factor that needs to be taken into account in evaluating the acquisitional worth of recasts is the literacy level of learners. There is clear evidence that alphabetic literacy has an impact on the way in which adults process oral language. Reviewing research in cognitive psychology on this issue, Tarone and Bigelow (2005) concluded that adults lacking in alphabetic literacy have difficulty in oral processing tasks that require an awareness of linguistic segments. The studies reviewed by Tarone and Bigelow controlled for the impact of social and educational background; the key factor appears to be literacy itself. The question then arises whether this interferes with learners' ability to notice and thereby learn from recasts, given that noticing presumes some degree of phonological analysis. Bigelow, delMas, Hansen, and Tarone (in press) set out to examine this possibility by investigating how a group of Somali ESL learners, who differed in terms of whether their literacy level was moderate or low, responded to recasts of their erroneous questions while performing communicative tasks. They reported that the more literate learners were better at recalling the recasts than the less literate learners. The researchers concluded that “literacy may augment learners' ability to visualize formal lexical representations in short term memory in processing negative corrective feedback.”
In general, recasts have been examined from an exclusively cognitive perspective (i.e., whether they constitute the kind of data on which the acquisitional mechanisms can operate); the social and sociocognitive dimensions of recasting have been neglected (see Firth & Wagner, 1997, for a more general critique of interactionist studies in SLA). In particular, there is a need to consider the orientation of the learner receiving the recasts. Batstone (2002), for example, argued that learners often find themselves between contexts; that is, they are simultaneously in a learning context and a communicative/social context. The former encourages an orientation to language as an object and, thus, is likely to lead to awareness that the recast is corrective and to noticing of the corrected form; the latter encourages an orientation to language as message by treating language as a tool for achieving communication. In this case, the learner might not become aware of the corrective force of a recast and might be less likely to carry out the cognitive comparison hypothesized to be important for learning. Some studies (e.g., Nabei & Swain, 2002; Mackey et al., 2000) have attempted to establish what learners attend to in recasts, but there has been an almost complete absence of research that has investigated attention to recasts in relation to learners' general and shifting orientations to the discourse.
What all this is saying, however, is not that recasts cannot work for acquisition but, rather, that they do not always do so. In this respect, they are almost certainly no different from other types of speech act that characterize input to learners. The fact that recasts have been found to occur with such high frequency in both naturalistic and classroom settings in comparison to other corrective feedback strategies (which is a direct outcome of their multifunctionality) ensures that they have acquisitional potential at least some of the time. Recasts are not necessary for acquisition; they are only facilitative when they occur in supportive social contexts, and this is true only some of the time, depending on the learner's orientation.
There is now substantial evidence that exposing learners to a rich diet of recasts can lead to acquisition. Both short-term experimental studies (e.g., Mackey & Philp, 1998; Ortega & Long, 1997) and longitudinal studies (e.g., Han, 2002; Ishida, 2004) have demonstrated that recasts, especially when they are directed in a concentrated fashion at a specific grammatical structure, promote acquisition. However, as Nicholas et al.'s (2001) review showed, recasts do not always work. Thus, the key question is not whether recasts facilitate acquisition but, rather, when and, even more crucially, how recasts work for acquisition.
WHAT FACTORS INFLUENCE WHETHER RECASTS ARE FACILITATIVE?
Recasts and Uptake
As a way of beginning to answer this question, another controversial issue—the significance of uptake and its role in acquisition—must be examined. For the purposes of their study of corrective feedback in an immersion context, Lyster and Ranta (1997) defined uptake as “a student utterance that immediately follows the teacher's feedback and that constitutes a reaction in some way to the teacher's intention to draw attention to some aspect of the student's initial utterance” (p. 48). Uptake can constitute repair (i.e., the uptake move corrects the initial error) or be characterized as needs repair (i.e., the uptake move does not correct the initial error). Controversy centers on whether the learner's uptake does or does not contribute to acquisition.
A number of studies that investigated recasts (e.g., Braidi, 2002; Ellis et al., 2001; Loewen, 2005; Loewen & Philp, in press; Lyster, 1998a; Lyster & Ranta, 1997; Oliver, 2000; Panova & Lyster, 2002; Sheen, 2004, 2006) also examined the extent to which learners respond to recasts and, more interestingly, the extent to which they repair the errors in their initial utterances by incorporating the correct forms from the recasts. Corrective episodes (3), (7), and (8) are examples of uptake with repair. Episodes (1), (2), (4), (5), and (6) either contain no uptake move or uptake in need of repair. Sheen (2004) showed that both uptake and repair have been demonstrated to vary according to setting. In her study, both uptake and repair were found more likely to occur in contexts in which learners are more oriented to language as an object (such as adult EFL classes) than in contexts in which the interlocutors are predominantly concerned with content (such as immersion classrooms). One reason for this difference appears to be that in classrooms in which there is a strong focus on message content, teachers often do not allow time for students to uptake their recasts, preferring, instead, to continue with topic development. Learner uptake of feedback is also more likely in a teacher-fronted lesson than in a pairwork situation (Oliver). Oliver suggested that this was because many of the recasts in the pairwork situation consisted of confirmation requests for which the appropriate response was simply yes or no. Thus, the extent to which uptake with repair occurs depends to a large extent on the instructional or social context of the interaction.
Researchers agree that although uptake with repair provides evidence that learners have noticed the reformulated forms and are able to deploy them, it cannot be used as evidence of acquisition. However, they disagree on two points: (a) whether noticing can be said to have occurred only if uptake with repair is present and (b) whether deployment of the correct forms contributes to their acquisition.
Repair serves as an indicator of learner noticing. If learners correct their original errors by incorporating the target forms from recasts, then they must have noticed these forms at some level. However, failure to repair the original errors cannot be taken as evidence of a failure to notice the target forms. Indeed, as Ohta's (2000) study of Japanese foreign language classrooms showed, learners' responses to recasts often take the form of private rather than social speech, nor does repair signal that a learner has identified the corrective function of a recast, as echoing the interlocutor's recast might simply constitute an acknowledgment of the propositional veracity of the interlocutor's statement—a reply designed to keep the channel of communication open. It follows, then, that the fact that recasts result in fairly low levels of repair in at least some learning contexts (Lyster & Ranta, 1997; Panova & Lyster, 2002) cannot be used as an argument against the acquisitional potential of recasts.
The key question, however, is whether repair constitutes an acquisitional mechanism. More specifically, does repair itself facilitate acquisition? There are theoretical arguments both for and against this proposition. Swain's (1995) output hypothesis claims that output serves as a means by which learners can develop greater metalinguistic awareness and fill in gaps in their interlanguage. N. Ellis (1996) suggested that practice serves as one of the ways in which patterns are entrenched in learners' L2 systems; uptake as repair can be seen as a kind of contextualized practice. However, it can also be argued that what is important for acquisition is the input available from the recasts rather than the output that may or may not occur in the learner's response to the recasts. An input-driven view of the acquisition process views output only as a means of increasing control over knowledge derived initially from input. According to this view, the acquisition of new linguistic forms is entirely dependent on input.
Ultimately, whether uptake facilitates acquisition must be determined empirically. It is unfortunate that, to date, there have been very few studies that have examined this issue. Mackey and Philp (1998) reported that successful uptake (i.e., repair) was not a significant factor in acquisition. Loewen and Philp (in press) found that successful uptake correlated with success in posttests, but only when the feedback consisted of elicitation or metalinguistic information. Uptake following recasts was not significantly related to success in their posttests. However, Nabei and Swain (2002) showed that the learner they investigated was more likely to be successful in posttests with items that she had uptaken from recasts. Given the limited amount of research that has investigated the relationship between uptake and acquisition, it is not yet possible to conclude whether it is the recasts alone or recasts + repair that is more effective.
There is a greater certainty regarding the impact of a number of other factors on the capacity of recasts to facilitate acquisition. These factors relate to the learner, the linguistic target of the recasts, and the type of recast.
Learner Factors
As with all types of form-focused instruction, the key learner factor is developmental readiness—that is, the extent to which individual learners have reached a stage of development that will enable them to incorporate the target forms addressed in the recasts into their interlanguage. The notion of developmental readiness derives from early work in SLA, which showed that learners follow a relatively fixed, universal order of acquisition and manifest clear developmental sequences in the acquisition of specific structures (see R. Ellis, 1994, chap. 3). Thus, if the recasts target features that a learner is developmentally primed to acquire, potentially those recasts will be effective; if the recasts target features that lie too far beyond the learner's current stage of development, they are likely to fail. Mackey and Philp (1998) provided convincing evidence that this is the case, at least for question forms. Nicholas et al. (2001) suggested that “recasts can be effective if the learner has already begun to use a particular linguistic feature and is in a position to choose between linguistic alternatives” (p. 752). This, however, is to stake a claim for recasts increasing control over existing knowledge rather than triggering new knowledge and, as such, stands in contrast to the claim that the real value of recasts lies in their promotion of new knowledge. Thus, what is needed is research that investigates the acquisition of features that learners have not yet begun to acquire. Such research is currently lacking, as studies to date have examined features that learners have already partially acquired.
It is important to distinguish the theoretical question of recasts in relation to developmental readiness from the practical question of how interlocutors (in particular, teachers) can ensure that the recasts that they provide are developmentally appropriate. Interlocutors (especially teachers in large classes) are not in a position to determine learners' stages of development, nor will they know with any precision whether learners are “in a position to choose between linguistic alternatives” (Nicholas et al., 2001, p. 752). This only matters, however, if the recasts are of the focused kind (i.e., directed intensively at some preselected linguistic feature). If the recasts are incidental and extensive (i.e., directed at whatever errors learners happen to make), this is not a problem. True, some recasts might miss their mark, but others will score a bull's-eye. Loewen and Philp (in press) reported that 50% of the recasts provided were usable by the classroom learners they investigated. Also, there are many linguistic items (lexical and grammatical) that are not subject to the constraints of developmental readiness. In short, developmental readiness should not be seen as a serious threat to the efficacy of recasts in contexts in which recasts are plentiful and extensive.5
However, developmental readiness is a potential threat to experimental studies of recasts that target specific grammatical structures. Clearly, researchers planning such studies need to establish that the target structure lies within the learners' acquisitional capacity.
It is also possible that other individual learner difference factors such as language aptitude, personality, general motivation, and attitudes toward correction will influence learners' receptivity to corrective feedback and, hence, its effectiveness. To date, no study has examined the moderating effects of these individual difference variables on the acquisitional worth of recasts, but two studies have examined them for corrective feedback in general. DeKeyser (1993) found that learners who had low anxiety and low extrinsic motivation benefited from teaching that included regular error correction, whereas learners with high extrinsic motivation improved more when the teaching was devoid of it. Havranek and Cesnik (2001) reported that what they called ideal learners (i.e., advanced learners who were only slightly irritated and not really embarrassed by correction, who reported finding correction important, who found speaking English easy and enjoyable, and who had high verbal intelligence and strong motivation) benefited most from corrective feedback. However, such learners might be expected to benefit from any kind of instruction, with or without corrective feedback. There is an obvious need to examine more closely how individual learners react to and benefit from corrective feedback (but see Sheen, in press).
Choice of Linguistic Target
One explanation for some of the mixed findings of the recast research is that recasts are differentially effective depending on the linguistic target. Ortega and Long (1997) found that recasts aided learners' adverb placement but not their use of clitic pronouns. Havranek and Cesnik (2001) reported that corrective feedback (which included large numbers of recasts) had a greater effect on verb inflections and rule-governed auxiliary use than on prepositions and tense choice. Leeman (2003) reported that learners benefited to a greater extent in Spanish number agreement than in gender agreement. Iwashita (2003) found that recasts predicted posttest scores for one Japanese structure (the te- verb form) but not for another (initial locative constructions). Ishida (2004) reported that the correct resultative use of the -te i (ru) structure in Japanese increased to a greater extent than the progressive use of the same structure after a diet of focused recasts.
It is possible that these mixed findings are related to the learners' developmental readiness (i.e., they were more ready to acquire some features than others), but this does not appear to be entirely the case [e.g., acquisition of the progressive use of the -te i (ru) structure typically precedes acquisition of the resultative use]. It might also have to do with the salience of the different structures in the recasts. Structures that are salient might be more easily noticed and, hence, acquired. Salience, however, remains a difficult construct to operationalize (but see Goldschneider & DeKeyser, 2005). Thus, the mixed results await a convincing explanation. Again, though, this is more of a theoretical than a practical problem if the linguistic environment is one in which recasts are incidental and extensive rather than planned and intensive.
Different Types of Recasts
What are the characteristics of those recasts that work best for acquisition? One general finding is that recasts are more effective when they are intensive (i.e., focused repeatedly on the same linguistic feature) and when the target features are enhanced in some way (e.g., by means of emphatic stress). Doughty and Varela's (1998) and Han's (2002) studies testified to this. This is tantamount to saying that recasts work well when they become a technique for delivering planned form-focused instruction. However, the same can be said of many other techniques—in particular, the provision of explicit instruction (Norris & Ortega, 2000). Of greater interest are those characteristics of incidental and extensive recasts that promote acquisition. Such recasts constitute what is natural and normal for most language learners. Understandably, given that these cannot be investigated experimentally but, rather, require careful analysis of data collected naturalistically, there have been few studies that have investigated recasts in this way to date.
A few studies have examined the relationship between different types of recast and uptake with repair. Sheen (2006) found that mode (i.e., whether the recast was declarative or interrogative in form), linguistic focus (i.e., whether the recast targeted phonological, lexical, or grammatical features), and type of change (i.e., whether the change involved substituting an item in the learner utterance or some other kind of change) influenced repair. Repair was more likely to occur if the recast was declarative, was directed at pronunciation and lexis rather than grammar, and involved substitution. Philp (2003) examined learners' ability to recall recasts immediately after hearing them. She found that short recasts were recalled better than long recasts, especially by less proficient learners, and recasts with just one or two changes were recalled more accurately by all learners, irrespective of proficiency. These studies, then, indicate which characteristics of recasts are more likely to promote noticing of the targeted features.
Only one study to date investigated the relationship between different types of recasts and acquisition. Loewen and Philp (in press) investigated the relationship between recast characteristics and scores on tailor-made posttests completed by individual learners. The characteristics that they examined were linguistic focus, length of recast, segmentation (i.e., whether the recast repeated all or just part of the learner's utterance), number of changes, and complexity (i.e., whether the corrective episodes were simple or complex, involving several turns). Of these, only the feature concerning the number of changes predicted test scores: If the recast made only a single change, the learners were more likely to get the right answer when tested.
To sum up, whereas there is clear evidence that recasts can facilitate acquisition, we still do not really have a clear picture of when they will do so. Learner factors, the nature of the targeted features, and the characteristics of the recasts help to determine, in complex ways, when recasts work for acquisition and when they do not. If recasts are intensive, focused, and individualized (as has been the case in laboratory studies), they are likely to be effective; however, so are other form-focused techniques. Of interest, then, is not just whether recasts facilitate acquisition but whether they do so more efficiently than other techniques, a question addressed in the next section. Researchers have begun to investigate the characteristics of extensive recasts (of the kind found in immersion and language classrooms) that predict uptake with repair, but the relationship between uptake and acquisition remains uncertain and controversial. Currently, little is known about which characteristics of extensive recasts are important for acquisition.
ARE RECASTS MORE EFFECTIVE THAN OTHER FORM-FOCUSED TECHNIQUES?
Comparisons between recasts and other form-focused techniques have examined the relative effectiveness of recasts in relation to (a) models and (b) other corrective feedback strategies. These studies have been motivated by both theoretical considerations (in particular, the contributions of positive and negative evidence in the acquisition of a L2) and practical issues (i.e., what kind of form-focused instruction to recommend to teachers).
Models have typically been operationalized as statements or instructions that learners are asked to listen to and then repeat. To provide a communicative context for such models, Long et al. (1998) asked learners to listen to instructions relating to the positioning of dolls in a drawing of a room and then to relay these instructions to the researcher, who placed the dolls as directed. Long et al. pointed out that such models differed from recasts in two crucial respects: (a) whether the utterance containing the grammatical information preceded the learners' own utterances (models) or followed them (recasts) and (b) whether the learners owned the content of their utterances (as was the case with the utterances that triggered the recasts) or not (as was the case with the models that they just copied). For experimental purposes, it is clearly necessary to contrive a context in which learners repeat the models that they hear in order to control for output (as learners must necessarily engage in output to elicit recasts). However, models operationalized in this way constitute a somewhat unusual discourse move, as there are few communicative contexts that require learners to relay instructions that they hear from one person to another. It is also possible that requiring the learners to repeat the propositional content of the models might have interfered with their ability to attend closely to the grammatical forms that encoded them. No such need exists with recasts.
The results of research that compared the effects of models and recasts on acquisition are very mixed. For the four grammatical structures that Long et al. (1998) investigated, models resulted in higher gain scores on one structure, recasts on one structure, and there was no difference for the other two. However, the only statistically significant difference (for adverb placement in Spanish) was in favor of recasts. In contrast, Ayoun (2004) found no difference between groups receiving models or recasts, but reported that a group receiving traditional grammar instruction outperformed both. Leeman's (2003) study, mentioned previously, found no difference in the posttest scores of learners exposed to models or recasts. In this study, models were operationalized somewhat differently because the participants only heard the models (they were not asked to repeat them) and the grammatical target was enhanced suprasegmentally.
A conservative reading of the research completed to date would be that models and recasts are equally effective (or equally ineffective, in comparison to explicit grammar instruction). Some degree of skepticism regarding studies that seek to compare generic recasts and models is also warranted. As emphasized previously, recasts are enormously varied, with evidence increasingly indicating that some types might be more facilitative than others. The same is almost certainly true of models. Ayoun (2004) noted that it was difficult to distinguish qualitatively between recasts and models in her study. Ultimately, what probably matters is not whether the data that learners have to work with occur before or after they speak, but the extent to which (a) the targeted forms are salient and (b) the learners are motivated to attend to them.
Other studies examined the relative effects of recasts and other types of corrective feedback on acquisition. The results of these studies are much clearer. In general, recasts do not work as well as feedback that elicits output from learners (see Havranek & Cesnik, 2001; Lyster, 2004) or feedback that provides explicit metalinguistic information (see Carroll & Swain, 1993). Feedback that leads to learner output—variably referred to as negotiation of form and prompts in the literature (see Lyster; Lyster & Ranta, 1997)—includes such strategies as clarification requests and elicitation. Whereas recasts provide the learner with data on target language forms, prompts elicit use of them by the learner. Comparing recasts and prompts serves as a way of addressing the relative contributions of focused input and pushed output (Swain, 1995) to acquisition. Studies that compared recasts with explicit correction, metalinguistic feedback, or both (e.g., Carroll & Swain, 1993; R. Ellis, Loewen, & Erlam, 2006) were motivated by the desire to examine the relative effects of implicit and explicit feedback on acquisition.
It is, of course, possible to find theoretical arguments to refute the demonstrated benefits of prompts and explicit feedback. Prompts do not supply learners with new linguistic knowledge; they merely activate (proceduralize) target forms that have already been partially acquired. In contrast, recasts supply learners with data on new as well as old forms while, if repair takes place, affording opportunities for pushed output. Explicit feedback might only appear to be more effective than implicit feedback in that it serves to develop learned linguistic knowledge but not true competence. Doughty (2001) argued that recasts are ideal because they facilitate cognitive comparison of interlanguage and target forms in a communicative context and, thus, influence learners' competence. Both of these arguments are convincing in making the case for recasts. However, they do not necessarily constitute arguments for rejecting prompts and explicit feedback. If prompts help learners to proceduralize and thereby restructure their existing knowledge, they clearly serve a valuable purpose. Explicit feedback might be considered of no value for acquisition only if (a) it can be demonstrated that it contributes to explicit knowledge but not implicit knowledge and (b) it can be argued convincingly that explicit knowledge plays no role (immediate or delayed) in the development of learners' competence. In fact, strong arguments have been advanced for claiming that learners' explicit knowledge assists the development of implicit knowledge (N. Ellis, 2005; R. Ellis, 1994), and a recent study by R. Ellis et al. (2006) provided evidence that explicit feedback in the form of metalinguistic explanation had a delayed effect on the acquisition of implicit knowledge of English past tense -ed. No such effect was found for recasts in this study.
To sum up, there is, as yet, no conclusive evidence that recasts are more effective than models, and there is some evidence to show that they are less effective than explicit types of corrective feedback and prompts, even when the measure of acquisition taps implicit knowledge. However, it remains possible that recasts will prove more effective in promoting acquisition of new linguistic features, although this remains to be demonstrated empirically, as the majority of existing studies examined features of which the learners already had partial knowledge.
CONCLUSION
Our knowledge about recasts and their role in the acquisition process is developing and will continue to grow. There are, however, a number of difficulties that researchers face in investigating recasts. The main purpose of this article has been to problematize research on recasts by identifying the difficulties. These include the following:
- Recasts do not constitute a homogeneous construct; they exist in many different forms. As Nicholas et al. (2001) pointed out, researchers have differed in how they operationalized recasts, making it difficult to compare results across studies. There is sufficient evidence now available to show that it is not appropriate to treat recasts as an undifferentiated genus.
- Recasts are also functionally different; they can serve both communicative and didactic functions. However, distinguishing recasts that negotiate meaning from those that negotiate form is difficult because there are no obvious formal signals that indicate what kind of negotiation is taking place; nor is there any empirical evidence that this distinction is significant for acquisition.
- Recasts have been considered a type of implicit corrective feedback. However, many recasts make the corrective force quite transparent. It is not difficult to find examples of recasts that are just as explicit as so-called explicit correction.
- It is not possible to say with any certainty whether recasts constitute a source of negative evidence (as is often assumed) or afford only positive evidence, as this will depend on the learner's orientation to the interaction. If learners treat language as an object to be studied, then they might detect the corrective force of recasts and thus derive negative evidence from them. However, if they act as language users and treat language as a tool, then they are less likely to see recasts as corrective and so will derive only positive evidence from them.
- Learners might experience difficulty in perceiving the corrective force of recasts because recasts are multifunctional. However, whether this is important for acquisition remains to be shown.
- Learners can adopt either a learner orientation or a communicative orientation to recasts, which might influence how they interpret them and to what they attend. Differences in learners' orientation might account for the different frequency of recasts and uptake found in different instructional contexts.
- Recasts have been viewed almost exclusively in terms of their linguistic and cognitive impact on acquisition. However, recasts do not occur in a social vacuum, and their efficacy might be influenced by sociopsychological factors that determine learners' receptivity to them.
- The role of uptake (including repair) remains uncertain. Repair constitutes an indicator of noticing, but noticing can also take place without uptake. Lyster (1998a) suggested that recasts have limited acquisitional value because they result in low levels of uptake with repair. Such a claim is justified if it can be shown that learner repair is related to acquisition. There is, in fact, some evidence that this is the case (Loewen, 2005; Williams, 2001).
- There is some evidence to show that learner factors (e.g., learner readiness and literacy level), the linguistic target of the recasts (e.g., morphosyntax as opposed to vocabulary or pronunciation), and the characteristics of the recasts themselves (e.g., the type and number of changes made in the recast) influence whether recasts result in uptake with repair or facilitate acquisition. Nevertheless, we are a long way from understanding how these various factors interact. Thus, although we know that recasts can facilitate acquisition, we do not know when they will.
- The claim that recasts are most effective when they are focused and intensive (i.e., directed repeatedly at a single linguistic feature) is of little practical significance for teachers because they are not in a position to know whether their learners are developmentally ready or whether the targeted feature is one amenable to treatment through recasts. Incidental, extensive recasts directed at whatever errors learners happen to make might be of greater practical value.
- Related to the previous point, care must be taken not to extrapolate from laboratory studies of recasts, which typically examine focused, intensive recasts, to classroom situations, in which recasts are typically of the extensive type.
- Despite the strong theoretical claims regarding the acquisitional potential of recasts, there is no clear evidence that recasts work better for acquisition than other aspects of interaction such as models, prompts, or explicit corrective strategies. Indeed, there is some evidence that the last two of these are more effective than recasts.
Finally, it would be wise to bear in mind that recasts are just one flower among many and, to extend this metaphor, that they are a very varied species of flora. Much might be gained by engaging in a careful, qualitative analysis of recasts (e.g., using conversational analysis as in Hauser, 2005) in order to tease out the variations and the contextual constraints that give rise to them.
APPENDIX
TRANSCRIPTION KEY
Symbol and Meaning
