5.1 Learner uptake

A somewhat contentious issue in both theoretical and methodological terms is the significance of learner uptake. With the aim of examining the illocutionary force of CF and to refer to the range of possible utterances made by students in response to CF, Lyster & Ranta (Reference Lyster and Ranta1997) borrowed the term uptake from speech act theory (Austin Reference Austin1962). Uptake was defined as a discourse move and not as an instance of acquisition, although some researchers have suggested that uptake may be ‘related to learners' perceptions about feedback at the time of feedback’ (Mackey, Gass & McDonough Reference McDonough and Mackey2000: 492) or ‘facilitative of acquisition’ (emphasis in the original; Ellis, Basturkmen & Loewen Reference Ellis, Basturkmen and Loewen2001: 287). Quantity of uptake is predicted by Robinson's (Reference Robinson and Robinson2011) cognition hypothesis to be a positive indicator of task complexity, with uptake and retention of linguistic forms associated with recasts provided during cognitively complex tasks to a greater degree than during simpler tasks. A prediction that has arisen from descriptive studies of different types of CF and uptake, and that can only be tested in experimental studies, is that different types of repair are likely to affect L2 development differentially over time, because different types of repair entail different types of processing.

Lyster & Ranta (Reference Lyster and Ranta1997) were not the first to quantify learner responses immediately following feedback (e.g. Chaudron Reference Chaudron1977; Brock et al. Reference Brock, Crookes, Day, Long and Day1986; Gass & Varonis Reference Gass, Varonis and Eisenstein1989; Doughty Reference Doughty and Alatis1994; Oliver Reference Oliver1995). Oliver (Reference Oliver1995), for example, coded learner responses to CF in one of four ways: respond, ignore, no chance (to respond), continue. In Lyster & Ranta's taxonomy, learner uptake was coded as either (a) utterances still in need of repair or (b) utterances with repair. Learner responses with repair are of greater value than responses still in need of repair, because the latter include simple acknowledgements such as ‘yes’, hesitations, off-target responses, partial repair, and occurrences of either the same or a different error. Learner repair entails the correct reformulation of an error and thus differs from modified output, which may or may not be a correct reformulation. Learner repair can be either a repetition or self-repair, and these different types of immediate repair are specific to one type of CF or another: recasts and explicit correction can lead only to repetition of correct forms by students, whereas prompts can lead, not to repetition, but either to self-repair or peer-repair.

Lyster & Ranta's (Reference Lyster and Ranta1997) data-driven model of an error treatment sequence comprising CF and uptake options served to reveal discrepant occurrences of immediate repair across instructional settings: infrequent repair has been observed after recasts in French immersion classrooms in Canada (Lyster & Ranta Reference Lyster and Ranta1997), English immersion classrooms in Korea (Lee Reference Lee2007), adult ESL classrooms in Canada (Panova & Lyster Reference Panova and Lyster2002), and EFL classrooms in Hong Kong secondary schools (Tsang Reference Tsang2004), but more frequent repair following recasts has been observed in Japanese immersion classrooms in the US (Mori Reference Mori2002), adult ESL classrooms in New Zealand (Ellis et al. Reference Ellis, Basturkmen and Loewen2001), and adult EFL conversation classes in Korea (Sheen Reference Sheen2004). The low rate of repair following recasts in some contexts was due to the fact that teachers often followed recasts with topic-continuation moves that prevented students from responding. If these instances had been excluded from the analyses, occurrences of repair following recasts would have increased, as in Oliver's (Reference Oliver1995) study of child dyads, which showed an increase from 10% to 35% in the number of repetitions after recasts. Yet because the teachers observed in classrooms with low rates of repair after recasts had used them in ways that prevented immediate repair, accounting for such instances seems critical if the research objective is to examine the overall capacity of recasts for drawing learners' attention to form. The coding of instances where learners have no opportunity to respond is nonetheless important, because it serves to demonstrate, as Oliver (Reference Oliver2000) acknowledged, ‘that the nature of whole class interactions diminishes the opportunity for students to respond to the feedback’ (p. 126).

As to whether repetitions of recasts are of any value, some studies have found that, when learners repeat a recast, they are more likely to perceive its corrective intention subsequently, during stimulated recall sessions (e.g. Mackey et al. Reference Mackey, Gass and McDonough2000; Egi Reference Egi2010). Révész, Sachs & Mackey (Reference Révész, Sachs, Mackey and Robinson2011) found in their study of high school EFL learners in Hungary that immediate repair following recasts of past progressive forms was a strong positive predictor of development when learners engaged in less complex, but not more complex, tasks. Loewen & Philp (Reference Loewen and Philp2006) found in their classroom study with adult ESL learners, however, that successful repair following recasts did not predict accuracy on posttest scores. Other studies conducted in laboratory settings also suggest that the potential effects of recasts on L2 development may be unrelated to immediate repair (Mackey & Philp Reference Mackey and Philp1998; Leeman Reference Leeman2003; McDonough & Mackey Reference McDonough and Mackey2006; McDonough Reference McDonough and Mackey2007), leading Mackey & Philp (Reference Mackey and Philp1998) to propose that repetitions of recasts ‘may be red herrings’ (p. 338).

At least three laboratory studies that controlled for immediate repair following different CF types suggest that its effects on L2 development vary according to CF type. First, Leeman (Reference Leeman2003) compared recasts with a CF move referred to as negative evidence that contained a repetition of the error preceded by ‘but you said. . .’ but without opportunities for immediate repair in either condition. Although there were no statistically significant differences between the recast and negative evidence groups, the recast group outperformed the control group while the negative evidence group did not. Second, McDonough (Reference McDonough and Mackey2007) investigated the effects of recasts and clarification requests, both of which included opportunities for immediate repair, and found no significant differences overall between recasts and clarification requests, although both CF types were more effective than no CF. Third, Lyster & Izquierdo (Reference Lyster and Izquierdo2009) compared recasts and clarification requests (followed by a repetition if necessary), including opportunities for immediate repair after the prompts but not after recasts, and found both CF types to be equally effective. Together, these results suggest that, in dyadic interaction with a researcher, learners are likely to benefit equally from different CF types provided intensively and repeatedly on the same target feature, provided learners are given opportunities for immediate repair after prompts, but not necessarily after recasts. This is because, in contexts of dyadic interaction, learners receiving recasts benefit from repeated exposure to positive exemplars as well as from opportunities to infer negative evidence, whereas learners receiving prompts benefit from repeated exposure to negative evidence as well as from opportunities to practise using the target form as they modify their output. In dyadic interaction, therefore, positive learning outcomes are associated with self-repair following prompts but not necessarily with learner repetition following recasts, whose variable effectiveness may instead be associated with specific input features such as length and prosody.

5.2 New vs. partially acquired knowledge

A second contentious issue is whether CF is intended to activate the acquisition of completely new knowledge or to consolidate partially acquired knowledge. With respect to CF, Long (Reference Long2007: 102) claimed that ‘acquisition of new knowledge is the major goal, not “automatizing” the retrieval of existing knowledge’ (see also Goo & Mackey Reference Goo and Mackey2013). However, the ultimate goal of instruction is not to continuously present only new knowledge to students, without providing enough subsequent opportunities for assimilation and consolidation of that knowledge. In school-based learning, students need repeated opportunities to retrieve and restructure their knowledge of the target language. This is because students may have target language knowledge that continues to be accessible for comprehension but that requires further activation before becoming readily available for accurate production. CF can be used to revisit target items and grammatical subsystems in ways that encourage the gradual development of a network of associations that become increasingly accessible for learners during communicative interaction.

Proponents of the CF-for-new-knowledge position claim that recasts are effective because they target forms as yet unknown to learners (Goo & Mackey Reference Goo and Mackey2013). It remains unclear, however, how a recast can be provided on a form about which the learner has zero knowledge; if the student produced the form, albeit erroneously, then some knowledge exists. Moreover, research tends to show that learners with developing knowledge of target forms in fact benefit more from recasts than do learners at lower stages of development. For example, Mackey & Philp (Reference Goo and Mackey1998) found that adult ESL learners at stage 3 in their use of English question forms made better use of recasts than did learners at stage 2. Accordingly, Nicholas et al. (Reference Nicholas, Lightbown and Spada2001) argued 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). Similarly, recasts provided in Ammar & Spada's (Reference Ammar and Spada2006) study were more effective for learners who had pretest scores above 50% than for learners with pretest scores below 50%. This contrasts sharply with the view that the purpose of CF is to initiate the development of new knowledge and that recasts, in particular, serve to enhance the acquisition processes of new knowledge (Long Reference Long2007). Hattie & Timperley (Reference Hattie and Timperley2007) make the compelling argument that ‘feedback has its greatest effect when a learner expects a response to be correct and it turns out to be wrong. [. . .] Conversely, if response certainty is low and the response turns out to be wrong, feedback is largely ignored’ (p. 95). This suggests that, even if it were feasible to provide CF following erroneous forms about which learners have no prior knowledge, such feedback (i.e. CF requiring the integration of completely new information) would not be as effective as feedback triggering associations between existing knowledge structures.

5.3 CF with or without instruction

A third contentious issue is whether research should examine CF only in isolation or also in conjunction with instruction (see Nassaji Reference Nassaji2009; Ellis Reference Ellis2012; Goo & Mackey Reference Goo and Mackey2013). Although theoretical justifications abound for isolating the effects of CF from instruction, CF and instruction may be best seen as complementary in pedagogical terms. In support of feedback provided in tandem with instruction, Hattie & Timperley (Reference Hattie and Timperley2007) argued that ‘feedback and instruction are intertwined in ways that transform the process into new instruction rather than informing the learner only about correctness’ (p. 82).

Many classroom studies investigating the effects of CF have done so in conjunction with instruction that served to create meaningful contexts of interaction, which enabled teachers to provide CF intensively on the target features. Some of these studies were designed to tease apart the effects of CF and those of other instructional activities, whose inclusion was based on the premise that ‘feedback can only build on something; it is of little use when there is no initial learning or surface information’ (Hattie & Timperley Reference Hattie and Timperley2007: 104). For example, in the classroom studies by Ammar & Spada (Reference Ammar and Spada2006), Lyster (Reference Lyster2004), Saito & Lyster (Reference Saito and Lyster2012a), and Yang & Lyster (Reference Lyster and Saito2010), the CF and no-CF groups alike were all exposed to the same instructional activities, which included focused production tasks that provided practice opportunities in all groups and contexts for CF provision in the treatment groups. In addition, some studies used typographical enhancement (Saito & Lyster Reference Saito and Lyster2012a) as well as consciousness-raising tasks embedded in the students’ regular curriculum (Lyster Reference Lyster2004). Engaging all groups in these tasks helped to level the playing field for comparisons between CF and no-CF groups and also to ensure that those receiving CF were at a similar level in their knowledge of the target feature. As Li (Reference Li2010) explained, with all groups receiving the same instruction but different CF treatments, any effects observed in the between-group comparisons must be due to the CF treatments. In contrast, the results of CF studies may be less robust when the control groups do not participate in equivalent instructional activities. For example, in the studies by Ellis et al. (Reference Ellis, Loewen and Erlam2006) and Sheen (Reference Sheen and Mackey2007), the CF groups participated in focused tasks that were designed to encourage the use of target forms during story retelling. These narrative-retelling tasks provided the context for CF. However, the control groups did not participate in the same tasks and thus missed the exposure to positive exemplars of the target forms in the original narratives and also in their classmates’ retelling of the stories, as well as in the oral interaction with their peers as they planned their retelling.

Related to the question of whether CF is best delivered in isolation or in tandem with instruction is the issue of providing intensive CF (a pre-planned focus on a single form) versus extensive CF (an incidental focus on a range of linguistic forms) (Ellis Reference Ellis2001). Whereas Ellis & Sheen (Reference Sheen2006) suggested that ‘extensive recasts directed at whatever errors learners happen to make might be of greater practical value’ (p. 597), Doughty (Reference Doughty and Robinson2001) argued that ‘recasting is most likely to be effective when it is targeted at only one or a few features’ (p. 255). Similarly, Lightbown (Reference Lightbown and Han2008a) argued that if students constantly receive diffuse CF they might develop the expectation that the teacher will always provide a corrected version and thus either cease to notice the CF or lose their motivation for self-monitoring and effortful generation of the target language. Arguing in favour of actually reducing the amount of CF in classrooms, Lightbown (Reference Lightbown and Han2008a) stated that ‘When feedback is focused on a limited number of objects or available in some classroom activities but not others, learners can take greater responsibility for creating and monitoring their own output’ (p. 41). Noteworthy in this regard, however, is that the meta-analysis by Russell & Spada (Reference Russell, Spada, Norris and Ortega2006) found no significant differences between intensive and extensive treatments, whereas Mackey & Goo (Reference Mackey, Al-Khalil, Atanassova, Hama, Logan-Terry and Nakatsukasa2007) found no differences at the time of immediate posttests but significantly larger effects for intensive treatments on delayed posttests.

6.1 Laboratory studies

For the most part, research demonstrating the effectiveness of recasts has been conducted in laboratory settings, where variables can more easily be controlled than in classroom settings and CF can be delivered intensively in consistent ways on specific linguistic targets. These laboratory studies have shown positive effects for recasts on L2 development (e.g. Long, Inagaki & Ortega Reference Long, Inagaki and Ortega1998; Mackey & Philp Reference Mackey and Philp1998; Han Reference Han2002; Iwashita Reference Iwashita2003; Leeman Reference Leeman2003; Ishida Reference Ishida2004; McDonough & Mackey Reference McDonough and Mackey2006) and are often invoked in reviews to support the effectiveness of recasts (see Long Reference Long2007; Mackey & Goo Reference Mackey, Goo and Mackey2007; Gass Reference Gass and Kaplan2010; Goo & Mackey Reference Goo and Mackey2013).

Some laboratory studies invoked in support of recast effectiveness, however, are not about oral recasts but rather about written reformulations presented on a computer screen (e.g. Ayoun Reference Ayoun2001; Sachs & Suh Reference Sachs, Suh and Mackey2007), while in others, recast groups were compared to control groups receiving no CF (Han Reference Han2002; McDonough & Mackey Reference McDonough and Mackey2006; Sagarra Reference Sagarra and Mackey2007); in Ishida's (Reference Isaacs and Trofimovich2004) study, there was no control at all, because a time-series design was used to track the progress of participants receiving recasts. In the study by Long et al. (Reference Long, Inagaki and Ortega1998), recasts were not compared with other types of CF, but rather with models: positive exemplars provided to learners before they speak. They found short-term benefits for recasts over models among learners of Spanish on adverb placement (but not object topicalization) and among learners of Japanese on required adjective ordering and a preferred locative construction. Iwashita (Reference Iwashita2003) also compared the effects of different interactional moves containing either positive evidence (models) or negative evidence (recasts) on two locative-initial constructions (word order and particle use) and te-form verbs in Japanese. She found positive effects for recasts on te-form verbs but not for the locative-initial constructions, which benefited more from positive evidence.

Mackey & Philp (Reference Mackey and Philp1998) conducted a laboratory experiment with adult learners of ESL to compare the effects of interaction with and without recasts on the production and development of question forms. Each participant performed three information-gap tasks with a native-speaker researcher who either recast intensively or withheld recasts and instead used ‘negotiation’. However, Mackey & Philp acknowledged an overlap in these CF types, because recasts, in some cases, ‘are part of negotiation sequences and function as confirmation checks’ (p. 342; see also Oliver Reference Oliver2000; Loewen & Philp Reference Loewen and Philp2006; Sheen & Ellis Reference Sheen, Ellis and Hinkel2011). Nevertheless, Mackey & Philp found that interaction with intensive recasts was more effective than interaction with fewer recasts, but only for advanced learners.

In a Japanese EFL context, Loewen & Nabei (Reference Loewen, Nabei and Mackey2007) conducted a laboratory study involving a researcher who interacted with small groups of four learners in order to simulate meaning-focused activities in classroom settings. They compared the effects of recasts, clarification requests, and metalinguistic feedback provided during meaning-focused tasks on English question formation. All CF groups significantly outperformed a control group, but no significant differences were found across the different CF treatments. Also with small groups of learners, Erlam & Loewen (Reference Erlam and Loewen2010) conducted a comparative study of implicit and explicit recasts provided in the context of interactive tasks targeting gender agreement in French. Implicit recasts entailed a single recast with rising intonation, while explicit recasts included a repetition of the error with rising intonation followed by a recast with declarative intonation. No differences were found across CF types.

Other laboratory studies focusing on the noticeability of recasts have used stimulated-recall methods to probe learners' perceptions of CF and hence the extent to which CF engages learners in a cognitive comparison (Ellis Reference Ellis and Ellis1994) or in ‘focused input analysis’ (Ellis Reference Ellis2005: 328; for theoretical discussion of noticing and L2 development, see Schmidt Reference Schmidt1990, Reference Schmidt and Schmidt1995; Tomlin & Villa Reference Tomlin and Villa1994; Robinson Reference Robinson1995, Reference Robinson, Doughty and Long2003; VanPatten Reference VanPatten1996; Skehan Reference Skehan1998; Doughty Reference Doughty and Robinson2001; Gass Reference Gass, VanPatten, Williams, Rott and Overstreet2004). Mackey et al. (Reference Mackey, Gass and McDonough2000), for instance, after videotaping interactions between L2 learners and native speakers, asked learners as they watched the video clips to comment on their perceptions about the CF they received. In addition to finding that the accuracy of their perceptions depended on the linguistic nature of the targets (morphosyntactic errors being the least accurately perceived), Mackey et al. (Reference Mackey, Gass and McDonough2000) found that recasts were not perceived as corrections to the extent that the CF provider had intended. Similarly, in Carpenter et al.'s (Reference Carpenter, Jeon, MacGregor and Mackey2006) experiment, learners viewing videotaped segments of a researcher responding to a learner with a mixture of recasts and non-corrective repetition were more likely to identify recasts as non-corrective repetition than as recasts – whether or not they actually heard the learner's preceding utterance (see also Egi Reference Egi2007, Reference Egi2010; Nassaji Reference Nassaji2009).

Yet another important strand of laboratory-based research has revealed the role played by individual differences in determining recast effectiveness. For example, the positive effects for recasts are reserved for learners with high literacy levels (Bigelow et al. Reference Bigelow, delMas, Hansen and Tarone2006), developmental readiness (Mackey & Philp Reference Mackey and Philp1998), high working memory capacity (Mackey et al. Reference Mackey, Philp, Egi, Fujii, Tatsumi and Robinson2002), as well as high phonological memory, attention control, and analytic ability (Trofimovich, Ammar & Gatbonton Reference Trofimovich, Ammar, Gatbonton and Mackey2007).

Several reviews of recast studies are available that put recast effectiveness into perspective through reference to a range of linguistic, pragmatic, cognitive, and contextual constraints (e.g. Nicholas et al. Reference Nicholas, Lightbown and Spada2001; Ellis & Sheen Reference Ellis and Sheen2006; Sato Reference Sato2011). One review that stands out for its optimism is by Long (Reference Long2007), in which research failing to find unequivocal support for recast effectiveness is ascribed to ‘sceptics’ (p. 94) whose ‘doubt’ is a ‘challenge to the optimism about recasts’ (p. 104). However, as Spada & Lightbown (Reference Spada, Lightbown, Mackey and Polio2009: 165) pointed out:

We turn next to classroom-based research that ranges from studies of learners' perceptions to intervention studies comparing different types of CF.

6.2 Classroom studies

Before addressing CF intervention studies, we mention two studies that examined learners' perceptions of CF in classroom settings. The first is Mackey et al.'s (Reference Mackey, Goo and Mackey2007) study in which L2 learners of Arabic (n = 11) and their two teachers watched videotaped CF episodes from their own classroom interactions. Findings revealed that, overall, only 36% of the CF was perceived in the way that teachers had intended (e.g. instances where CF targeted a morphosyntactic error were accurately understood by the learners as morphosyntactic CF). Explicit CF (either containing metalinguistic information or clearly eliciting self-repair) was more accurately perceived than implicit CF (recasts and negotiations for meaning). With regard to recasts, interrogative recasts were more accurately perceived than declarative recasts. The other is Mackey's (Reference Lyster, Sato, Mayo, Gutierrez-Mangado and Adrián2006) study, which stands out as one of the few classroom investigations of how learners' reports of noticing CF may be related to L2 development (i.e. question forms, plurals, and past tense). Noticing data were collected by means of stimulated recalls, learning journals, a questionnaire, and focused questions in the learners' L1. Statistical analyses partially supported the relationship between noticing and L2 development: learners' self-reports of noticing were significantly related to L2 development only for question forms. This result indicates that learners reported noticing which in the end was not associated with L2 development and also that CF contributed to the learning of forms on which learners had not reported receiving CF. Thus, while many researchers argue that noticing accounts for CF efficacy, it is difficult to claim that ‘learning follows noticing, or is dependent on noticing’ (Mackey Reference Mackey2006: 423; see also Mackey et al. Reference Mackey, Philp, Egi, Fujii, Tatsumi and Robinson2002; Kim & Han Reference Kim, Han and Mackey2007; Yoshida Reference Yoshida2008b, Reference Yoshida2010). To reveal the mechanisms of how and why CF impacts L2 development, further research on learners' perceptions is warranted to examine the nature of noticing (e.g. levels of noticing: see Svalberg Reference Svalberg2007) and its measurement (Ammar & Sato Reference Ammar and Sato2010), as well as the causal relationship between noticing CF and L2 development.

At least two classroom intervention studies (Doughty & Varela Reference Doughty, Varela, Doughty and Williams1998; Saito & Lyster Reference Saito and Lyster2012a) compared the effects of instructional activities with and without CF, and both found the effects to be greater with CF than without. In one of the first classroom studies to control for the provision of CF and to demonstrate its effectiveness, Doughty & Varela (Reference Doughty, Varela, Doughty and Williams1998) examined the effects of what they called ‘corrective recasts’ (a repetition of the error followed by a recast if necessary) in two content-based ESL classrooms. A group of 11–14-year-old students conducted a set of experiments in accordance with their regular science curriculum. The class receiving CF during the reporting phase showed significant short- and long-term improvement in comparison to a class engaged in the same production tasks but without CF. Doughty & Varela's study did not directly examine the effects of recasts, because recasts were used solely as secondary moves in the event that the primary move – a prompt that repeated verbatim the learner's error – failed to elicit self-repair. Students appeared especially to benefit from the teacher's repetition of their non-target utterances, as evidenced by the observation that, by the beginning of the second of three treatment sessions, ‘students were beginning to self-correct before the teacher had the opportunity to recast’ (p. 135). The science tasks creating obligatory contexts for the use of the target forms in this study provide excellent models of exemplary production tasks congruent with content area curricular objectives. However, the teacher's use of a double-feedback move was arguably much more explicit than the researchers’ characterization of corrective recasts as an ‘implicit focus on form’ (p. 118).

Saito & Lyster (Reference Saito and Lyster2012a) investigated the pedagogical value of recasts on the acquisition of /ʴ/ by adult Japanese learners of English. Students in the experimental groups engaged in tasks designed to develop argumentative skills in English while drawing attention to the target form through typographically enhanced input and providing opportunities for production practice. At the same time, teachers provided CF only to those in the CF group by recasting their mispronunciation and unclear pronunciation of /ʴ/. According to the results of the listener judgment and acoustic analyses, only those who received recasts during the tasks demonstrated gains, not only at a controlled-speech level (measured via word and sentence reading) but also at a spontaneous-speech level (measured via a timed picture-description task) (see also Saito & Lyster Reference Saito and Lyster2012b for the effects of recasts on the acquisition of L2 vowels). In a subsequent study, Saito (in press) found that providing explicit phonetic information (i.e. exposing students to teachers’ exaggerated model pronunciation of /ʴ/) before the form-focused tasks significantly enhanced the generalizability and magnitude of recast effectiveness. That is, students who received both recasts and explicit information demonstrated considerable improvement with large effects (changing from use of hybrid exemplars to good exemplars of /ʴ/), not only for familiar lexical items that appeared during instruction but also for unfamiliar lexical items.

Quasi-experimental classroom studies comparing different types of CF have shown overall positive effects for CF as well as some advantages for prompts and explicit correction over recasts. For example, with young learners, Ammar & Spada (Reference Ammar and Spada2006) investigated the effects of recasts and prompts on the acquisition of possessive determiners by French-speaking ESL learners. While both groups receiving CF showed superior performance compared to the control group, the group receiving prompts significantly outperformed the recast group on written and oral posttests. Noteworthy in their study, as previously mentioned, was the finding that learners initially demonstrating more accurate use of possessive determiners benefited from both types of CF, while those initially demonstrating less accurate use benefited from prompts but not from recasts.

Also with young learners, Lyster (Reference Lyster2004) investigated the effects of form-focused instruction and CF on French immersion students’ acquisition of grammatical gender. He found that instruction with prompts led to significantly higher results than instruction with recasts in written production but not in oral production. He attributed the lack of significant differences in oral production measures to a large task effect: the subsample of students participating in the oral production measures benefited greatly from the opportunities to interact one-on-one with a near-native speaker of French who provided them with valuable oral practice to an extent that was impossible to match in class. Overall, of the eight posttest measures, the comparison group was significantly outperformed by the prompt group on all eight measures, by the recast group on five, and the no-CF group on four.

In a partial replication of Lyster's (Reference Lyster2004) study, Algarawi (Reference Algarawi2010) found that all students receiving CF showed significant improvement in their use of passive forms in written production, regardless of CF types (there were no measures of oral production). Conducted in a very different instructional setting from the French immersion context of Lyster's study, Algarawi's study involved adult EFL students in form-oriented courses of English for Academic Purposes (EAP) at a university in Saudi Arabia. In EFL classrooms in China, Yang & Lyster (Reference Lyster and Saito2010) compared the differential effects of recasts, prompts, and no CF on the use of regular and irregular past tense forms by undergraduate English majors. The effects of prompts were larger than those of recasts for increasing accuracy in the use of regular past-tense forms, while prompts and recasts had similar effects on improving accuracy in the use of irregular past-tense forms.

With adult ESL learners, Ellis et al. (Reference Ellis, Loewen and Erlam2006) investigated the differential effects of recasts and metalinguistic feedback (i.e. a repetition of the error followed by metalinguistic information: ‘Kiss – you need past tense’ 2006: 353) on the acquisition of regular past tense in English (see also Ellis Reference Ellis and Mackey2007). The main purpose of their study was to examine whether learners with a great deal of explicit knowledge could develop more implicit knowledge (measured via timed oral imitation task) as a result of two different types of CF treatment. The metalinguistic feedback was more effective overall than recasts and the effect was found more in the delayed posttest than the immediate posttest. Also with adult ESL learners, Sheen (Reference Sheen and Mackey2007) compared the effects of recasts and metalinguistic corrections on the use of English articles. Metalinguistic corrections included provision of the correct form followed by metalinguistic explanation (e.g. ‘You should use the definite article the because you've already mentioned fox’). The metalinguistic group significantly outperformed both the recast and control groups, and its positive gain scores were correlated with both language analytic ability and attitudes towards CF. In contrast, the recast group did not significantly outperform the control group and its gain scores were related neither to language analytic ability nor to attitudes towards CF. In a related study with a subsample of these participants, Sheen (Reference Sheen2008) investigated the effects of language anxiety on recast effectiveness, confirming that low-anxiety learners receiving recasts significantly outperformed not only high-anxiety learners receiving recasts but also low-anxiety learners receiving no recasts in a control group. No significant differences emerged between the high-anxiety recast group and the control groups.

To summarize, experimental classroom studies of CF consistently confirm that oral CF is significantly more effective than no CF and also reveal a tendency for learners receiving prompts or explicit correction to demonstrate more gains on some measures than learners receiving recasts (for similar summaries, see Sheen Reference Sheen2011; Ellis Reference Ellis2012). Specifically in terms of significant differences, repetitions of learner errors followed by recasts if necessary were more effective than no CF (Doughty & Varela Reference Doughty, Varela, Doughty and Williams1998), and recasts were more effective than no CF for improving pronunciation of familiar, but not unfamiliar, items (Saito & Lyster Reference Saito and Lyster2012a). Recasts were as effective as prompts for young ESL learners with high pretest scores but less effective than prompts for learners with low pretest scores (Ammar & Spada Reference Ammar and Spada2006). In the case of young immersion students, recasts were less effective than prompts in written production measures but equally effective in oral production measures (Lyster Reference Lyster2004). With adult learners in EAP courses, no significant differences between CF types were apparent in the accuracy of passive forms used in written production (Algarawi Reference Algarawi2010). Adult EFL students in China benefitted more from prompts than recasts in improving their accurate use of regular past-tense forms but benefitted equally from both CF types in improving accuracy of irregular forms (Yang & Lyster Reference Yang and Lyster2010). With adult ESL learners, metalinguistic feedback proved more effective than recasts (Ellis et al. Reference Ellis, Loewen and Erlam2006; Ellis Reference Ellis and Mackey2007) and explicit corrections with metalinguistic explanations were also more effective than recasts (Sheen Reference Sheen and Mackey2007).

With respect to the relative benefits of recasts and prompts, Lyster & Saito (Reference Lyster and Saito2010) interpreted the similar findings yielded by their meta-analysis of 15 classroom studies of oral CF to mean that classroom learners are able to benefit from the positive evidence available in recasts as well as from the opportunities recasts provide to infer negative evidence, but may benefit even more from the negative evidence available in prompts and from the greater demand they impose for producing modified output. With respect to the benefits of explicit CF, Ortega (Reference Ortega2009: 75) stated that ‘when two or more implementations of negative feedback are compared, the more explicit one leads to larger gains’ and went on to say that this finding ‘is hardly illuminating’ because it mirrors the findings of the meta-analysis by Norris & Ortega (Reference Norris and Ortega2000) concerning the superiority of explicit instructional treatments over more implicit ones (see also Spada & Tomita Reference Spada and Tomita2010). Ellis (Reference Ellis2012) cautioned, however, that the reasons that one type of CF leads to more gains than another are difficult to interpret: ‘These strategies are not as “pure” as they are sometime presented in the literature. Recasts, in particular, occur in many different forms. Prompts are a mixture of implicit and explicit strategies’ (p. 263). Therefore, if we take the Ellis et al. (Reference Ellis, Loewen and Erlam2006) study as an example, it is difficult to tease apart whether participants benefitted from (a) the teacher's repetition of the error ‘Kiss’, (b) the metalinguistic information conveyed by ‘you need past tense’, (c) the opportunity to self-repair, or (d) all of the above.

In the case of prompts, because they include more than one way of providing negative evidence while withholding positive evidence, it may be their variety that adds to their effectiveness. The variety in prompts has led to the criticism that comparing recasts with prompts is like comparing one variable with multiple variables (Goo & Mackey Reference Goo and Mackey2013). It is important to note, however, that the use of prompts by teachers in comparative studies of recasts and prompts has not entailed more instances of CF in the prompt condition, just more variety. Although prompts have multiple manifestations, which may indeed contribute to their effectiveness, they might be considered a single strategy insofar as they withhold correct reformulations. Similarly, recasts might be considered a single strategy (i.e., CF that provides correct reformulations) but with multiple manifestations. As Mackey & Goo (Reference Goo and Mackey2007) suggested, the term ‘recast’ ‘is increasingly recognized as being elastic in nature’, although they considered it, as many researchers do, ‘as a monolithic construct’ (p. 413). To point out just a few examples of such elasticity, other researchers have considered recasts as: conversational or didactic (Sheen & Ellis Reference Sheen, Ellis and Hinkel2011), implicit or explicit (Erlam & Loewen Reference Erlam and Loewen2010), and interrogative or declarative (Loewen & Philp Reference Loewen and Philp2006).

That instructional interventions with variety may be more effective than interventions with less variety is in line with studies demonstrating superior effects for explicit over implicit treatments. That is, whereas implicit instruction is typically operationalized in a relatively restricted way that may simply involve exposure to the target structure, explicit instruction often involves a combination of several instructional strategies that may include rule presentation, focused practice, corrective feedback, and repetition of the rule (Norris & Ortega Reference Norris and Ortega2000). As Lightbown (Reference Lightbown and Han2008a) pointed out, variety in instructional practices not only stimulates interest but may also ‘increase both the depth and the transferability of learning’ (p. 40).

Because a variety of CF types is probably more effective than consistent use of only one type, it may not be necessary or even possible for researchers to identify the single most effective CF strategy. Moreover, CF types are not the only variable affecting CF effectiveness, as shown by the different levels of noticeability yielded by different studies; that is, learner factors such as instructional contexts, proficiency levels, metalinguistic knowledge, and age are all variables that influence ‘the ultimate effectiveness of CF’ (Sato Reference Sato2011: 7). The next sections specifically address the influence of linguistic targets and age. Given the exploratory nature of these topics, we aim at the same time to suggest directions for future CF research.

7.1 Grammatical targets

Whether conducted in laboratories or classrooms, CF research has focused to a great extent on grammatical targets, reflecting the preoccupation with grammatical development in the study of SLA. Some examples of grammatical features from the aforementioned studies include questions, passive forms, past tense, articles, and possessive determiners in English; gender attribution and noun-adjective agreement in French; adverb placement, object topicalization, and gender agreement in Spanish; aspectual forms, adjective ordering, and locative constructions in Japanese. However, the effects of CF differed widely depending on the linguistic target; as Sheen (Reference Sheen2011) observed, ‘we cannot assume that because CF has been shown to assist the acquisition of one grammatical feature it will necessarily do so for all features’ (p. 165).

Two of the aforementioned classroom studies were designed to examine the differential effects of CF types on different kinds of grammatical targets. The first was conducted with ESL learners in New Zealand by Ellis (Reference Ellis and Mackey2007), comparing the variable effects of recasts and metalinguistic explanations on regular past tense –ed and comparative –er. Ellis hypothesized that these features differed in terms of grammatical difficulty. Comparative –er was considered more difficult than past –ed, because it involves both morphology and syntax, occurs less frequently, emerges after –ed in terms of learnability, and the rule to be learned explicitly is more complex. The group receiving metalinguistic explanations showed greater improvement on –er than –ed, while the recast group did not show any significant gains over the control group on any of the measures. Ellis attributed the ineffectiveness of recasts to their lack of saliency due to the shortness of the one-hour treatment and also to the shortness of the recasts themselves. In contrast, the metalinguistic feedback was deemed sufficiently salient in spite of the short treatment and especially effective for comparative –er simply because, according to the pretest scores, learners initially demonstrated well-developed explicit knowledge of past –ed but not of comparative –er, so had more room for improvement in their use of the comparative.

The second was Yang & Lyster's (Reference Yang and Lyster2010) study of the effects of recasts, prompts, and no CF on past-tense forms. Due to the predictable and rule-based nature of regular past-tense verbs on the one hand, and the complex and unpredictable yet highly salient nature of irregular past-tense verbs on the other (DeKeyser Reference DeKeyser, Doughty and Williams1998; Ellis Reference Ellis2005), the authors predicted that representational and acquisitional processes would vary between regular and irregular forms. All learners participated in a set of focused production tasks that elicited the target forms and provided contexts for the provision of CF in the recast and prompt groups. Regular past-tense forms proved more amenable to prompts than to recasts, arguably because of the negative evidence afforded by prompts but not by recasts incorporating a nonsalient morpheme (recast of ‘shop’ is ‘shopped’). Learners who may not notice the additional morpheme in recasts of regular forms yet have plenty of explicit knowledge are able to benefit from the negative evidence in prompts and also from the opportunity to self-repair. In contrast, recasts of irregular past-tense forms entail a revised intact syllable associated with high saliency (recast of ‘buy’ is ‘bought’). Given their greater noticeability, irregular forms were found to be amenable to both recasts and prompts, again because of the negative evidence afforded by prompts and also by recasts incorporating salient positive exemplars.

7.2 Lexical targets

With respect to interaction in general, Mackey & Goo's (Reference Mackey, Al-Khalil, Atanassova, Hama, Logan-Terry and Nakatsukasa2007) meta-analysis demonstrated that its effects were significantly larger for lexical than for grammatical development. Specifically with respect to CF provided on either lexical or grammatical errors, Egi (Reference Egi2007) provided some evidence that learners process recasts of morphosyntactic and lexical errors differently. In her study, learners of Japanese as a foreign language received recasts during dyadic communicative tasks with native speakers. Using stimulated-recall measures and tailor-made posttests to investigate which components of recasts (i.e., positive or negative evidence) benefit learners' morphosyntactic and lexical development, she found that the noticing of positive evidence in recasts was more likely to result in immediate interlanguage changes in vocabulary than in morphosyntax.

In Ellis & He's (Reference Ellis and He1999) study, after receiving explicit instruction on ten unknown words, adult ESL students were paired and asked to complete a collaborative problem-solving activity requiring them to use the new lexical items in interaction with their partner. To successfully complete the task, the learners provided and received clarification requests if anything was unclear. The results showed that they demonstrated significant improvement in receptive and productive use of new lexical items, especially compared to those who simply received pre-modified input (i.e., listening to task directions read by their teachers at a slow speech rate without any interaction opportunities) (for similar results, see de la Fuente Reference de la Fuente2002).

Dilans (Reference Dilans2010) teased apart CF as an independent variable and tested the relative effects of different CF types on the acquisition of ten unknown words by adult ESL students. After receiving explicit instruction on the items, the students completed a written production task (i.e., sentence creation) and an oral production task (i.e., picture labelling). During these two tasks, teachers provided either recasts or prompts in response to ungrammatical usage and mispronunciation of these words. The attention given to both grammar and pronunciation in the context of lexical development is in line with Webb's (Reference Webb2005) call for focusing on ‘multiple aspects of L2 lexical knowledge’ (p. 35) and with other L2 vocabulary researchers, according to whom knowing a word requires the acquisition not only of form and meaning but also of orthography, phonology, syntax, and the grammatical function of the item (see Schmitt Reference Schmitt2008). Both the recast and prompt groups improved equally in partial-precise and receptive-productive vocabulary knowledge, but only the prompt group demonstrated gains in depth of vocabulary knowledge (i.e., increasing awareness of syntactic and pragmatic associations with the target words). In line with previous findings in the L2 morphosyntax literature, Dilans attributed the relative effectiveness of prompts to the role of pushed output, a point that has also been raised from a cognitive psychology perspective with respect to lexical acquisition. Given that acquiring new words and then retaining them within one's long-term memory requires repetition of the words at increasing intervals over a long period of time, these repetition sequences can be even more effective when learners are pushed through prompts to recall the words instead of being presented with them in recasts (i.e., retrieval practice effects: Hulstijn Reference Hulstijn and Robinson2001; Nakata Reference Nakata2008).

To further examine how providing various types of CF through interaction can promote L2 vocabulary development, we call for future research that interfaces the existing CF research paradigm with L2 vocabulary research in an interdisciplinary manner. One direction for future research of this kind is the integration of psycholinguistic methodologies to measure the impact of instruction on L2 vocabulary development. For example, Elgort (Reference Elgort2011) employed priming techniques to examine the impact of list-learning on both representational and functional aspects of L2 lexical knowledge, providing a useful framework for future CF studies to measure the effects of more meaningful interventions (e.g. communicative input and output activities followed by CF) on enhancing automaticity in L2 vocabulary development (for discussion, see Hulstijn Reference Hulstijn and Robinson2001).

7.3 Phonological targets

With respect to CF in L2 phonology some studies, as reported earlier, have examined how instruction with and without recasts can facilitate L2 speech learning processes (Saito & Lyster Reference Saito and Lyster2012a, Reference Saito and Lyster2012b; Saito in press). Though few in number, these studies suggest that short pronunciation-focused recasts can play an important role in L2 pronunciation development, arguably because students benefit from the opportunities afforded by such recasts, first, to notice the negative evidence directed at the intelligibility of their output and, second, to practise the correct form in response to their teachers’ model pronunciation (positive evidence). Together, however, these studies also showed that provision of explicit information might be necessary for pronunciation-focused recasts to lead to larger and more generalizable changes in relatively difficult cases of L2 speech learning (e.g. the acquisition of /ʴ/ by Japanese learners of English). Explicit information enables students to attend to the phonetic unit of L2 input, and thus make the best of subsequent recasts that embed the target sound at a lexical level.

More research of this kind is needed to examine further how teachers can use CF techniques to promote various areas of L2 phonological development, such as suprasegmentals (speech rate, intonation, rhythm) and syllable structure. Another under-studied but pedagogically crucial aspect in measuring phonological CF effectiveness includes the development of global listening skills (Vandergrift Reference Vandergrift2007). A majority of L2 listening research has focused on the importance of increasing learners' awareness of metacognitive strategies via a range of pre-listening activities in a proactive manner (Vandergrift et al. Reference Vandergrift, Goh, Mareschal and Tafaghodtari2006). Recently, this line of research has begun to reveal that L2 vocabulary size is a relatively strong predictor of global L2 skills (Stæhr Reference Stæhr2009). This in turn suggests the pedagogical potential of L2 listening instruction that reinforces bottom-up processing in order to help learners build robust phonological representations of L2 words and sounds (Field Reference Field2008). Such pedagogical techniques include slow speech rate (Zhao Reference Zhao1997), repetition of oral texts (Jensen & Vinther Reference Jensen and Vinther2003), and attending to prosodic features (Harley Reference Harley2000). Future research could examine whether and to what degree providing CF in tandem with instruction emphasizing bottom-up processing can promote learners' knowledge of vocabulary items and ultimately enhance their global listening skills.

7.4 Pragmatic targets

A general consensus has been reached in the area of interlanguage pragmatics according to which explicit instruction is more facilitative than implicit instruction for improving learners' use of difficult pragmatic structures that include speech acts such as refusals, invitations, requests, offers, suggestions, agreements, complaints, and apologies (see Jeon & Kaya Reference Jeon, Kaya, Norris and Ortega2006 for a meta-analysis; Kasper & Rose Reference Kasper and Rose2002 for a narrative review). Recently, some attempts have been made to investigate how teachers can use a range of CF techniques during form-focused activities to foster learners' emerging L2 pragmatic knowledge.

Takimoto (Reference Takimoto2006) compared the effects of structured input tasks with and without explicit CF on the ability of adult Japanese learners of English to mitigate requests. All participants engaged in structured input tasks requiring them to rate the appropriateness of dialogues in different situations; participants in the CF group also received explicit CF that involved ‘either a metalinguistic question to elicit a correct response or provision of a metalinguistic rule’ (p. 411–412). Both groups made significant progress in the receptive and production tasks, although the CF group had slightly higher scores than the no-CF group on all measures. Again targeting requests, but with university-level English-speaking learners of Spanish, Koike & Pearson (Reference Koike and Pearson2005) investigated the differential effects of explicit CF (i.e., correct response along with metalinguistic information) and implicit CF (i.e., clarification requests). Their study also examined the extent to which explicit instruction before input enhancement activities influences CF effectiveness. Learners who received both explicit instruction and CF showed distinguishable improvement in receptive measures (multiple choice tasks), while those who received clarification requests without explicit instruction outperformed the other groups in production measures (dialogue tasks). In their study of the appropriate use of refusals by university-level EFL learners in Thailand, Nipaspong & Chinokul (Reference Nipaspong and Chinokul2010) compared the effects of explicit CF (i.e., metalinguistic clues followed by the right answer) and prompts (i.e., elicitation, repetition, metalinguistic clues). Unlike the aforementioned studies, both of which implemented brief treatments ranging from only 40 to 60 minutes, Nipaspong & Chinokul employed a longitudinal quasi-experimental design involving ample instruction and practice over a semester (15 hours). The results of the quantitative data (multiple-choice tests) and qualitative data (think-aloud analyses) showed that learners in the prompt group improved their pragmatic awareness of the target feature more than those in the explicit CF group.

Taken together, these findings confirm that focused instructional tasks benefit learners' L2 pragmatic development and suggest that different types of CF contribute differentially to this development. More research is necessary, however, to obtain a better understanding of the relationship between CF and L2 pragmatic knowledge. Useful research in this area would include detailed descriptions of the nature and operationalization of the CF treatments along with theoretical support (see Nguyen, Pham & Pham Reference Nguyen, Pham and Pham2012 for their detailed discussion of various form-focused options in the context of L2 pragmatics teaching). Also useful would be explorations of how factors such as learners' perceptions and responses to CF interact to mediate its effects specifically on L2 pragmatic development. We call for further research to examine the impact of CF on L2 pragmatics according to various independent variables such as different types of CF (as in Nipaspong & Chinokul Reference Nipaspong and Chinokul2010), various task conditions (e.g. deductive vs. inductive tasks: Takimoto Reference Takimoto2008), various output measures (e.g. controlled vs. free constructed responses: Nguyen et al. Reference Nguyen, Pham and Pham2012), and speech acts with different levels of frequency and saliency in the input (Bardovi-Harlig & Dörnyei Reference Bardovi-Harlig and Dörnyei1998).