Recast and Explicit Feedback to Young Language Learners: Impacts on Grammar Uptake and Willingness to Communicate

It is generally believed that students learn a second language (L2) best through meaningful interaction with others. However, many studies are indicative of the fact that exposing learners to mere communicative activities is not as effective as enriching communicative activities with consciousness raising activities such as error correction (e.g., Batstone & Ellis, 2009; Hawkes, 2012; Leeman, 2003; Nassaji & Fotos, 2011; Shirazi & Sadighi, 2012). The latter situation enhances learners’ metalinguistic awareness (Serrano, 2011; Varnosfadrani & Basturkmen, 2009) and offers them the chance to notice the gap between their erroneous utterances and target-like utterances and to make modification to their ungrammatical forms (Gass & Lewis, 2007; Van Beuningen, De Jong, & Kuiken, 2012).

Teachers make use of corrective feedback (CF) as a consciousness raising technique to draw students’ attention to their errors while communication. In recent years, there has been a growing body of research on CF including observational and experimental studies into the teachers’ most frequent feedback types in language classrooms (e.g., Panova & Lyster, 2002; Tsang, 2004), teachers’ and students’ preferred corrective practices (e.g., Ghahari & Farokhnia, 2017a, 2017b; Lee, 2013; Park, 2010), and the effectiveness of different types of CF (Ellis, Loewen, & Erlam, 2006; Rassaei, 2013; Varnosfadrani & Basturkmen, 2009). What gives legitimacy to the increasing line of research in this area is that, according to Choi and Li (2012), feedback is a dynamic construct interacting with many contextual and individual factors. The extent to which a particular CF bears the expected results is partly determined by learner-internal factors such as age, gender, motivation, and personality traits as well as learner-external variables such as task type, target features, level of proficiency, and type of instruction (Ammar & Spada, 2006; Li, 2010; Lyster & Saito, 2010; Sheen, 2010). Depending on certain factors including pedagogical purposes and students’ age, teachers might decide to use different types of CF. Lyster and Ranta (1997) divided CF types into six categories of recasts, explicit correction, clarification requests, metalingustic feedback, elicitation, and repetition. The first two categories of recasts and explicit correction are considered in the present study in order to compare the effect of implicit and explicit CF on young L2 learners’ L2 progress.

One of the leading individual characteristics in language learning is willingness to communicate (WTC). WTC is defined as “a readiness to enter into discourse at a particular time with a specific person or persons using an L2” (MacIntyre, Clément, Dörnyei, & Noels, 1998, p. 547). Some factors influence L2 learners’ WTC including motivation (Hashimoto, 2002), group size and familiarity with the interlocutors (Cao & Philp, 2006), the topic of conversation and teachers’ wait time (Cao & Philp, 2006; Zarrinabadi, 2014), and error correction (Kang, 2005; MacIntyre et al., 1998). Following this line of research, the present study aimed at exploring the psychological as well as educational dynamics of two error treatment techniques, with one lying at the explicit end of CF continuum and the other at the implicit end, for a cohort of young L2 pupils.

LITERATURE REVIEW

Recast is the reformulation of all or part of an utterance by the teacher. It is the most commonly studied type of corrective feedback (Ellis & Sheen, 2006) possibly due to the high frequency with which it is used by language teachers (e.g., Kang, 2008; Lyster & Ranta, 1997; Panova & Lyster, 2002). Besides, it is unobtrusive and does not interrupt the flow of communication (Lyster, 1998b; Trofimivich, Ammar, & Gatbonton, 2007). For this latter reason, recasts are considered an implicit feedback in L2 learning.

In Doughty and Varela’s (1998) study on young learners in a content-based classroom, learners who received recasts outperformed the group who received no feedbacks. Mori (2002) found that his Japanese immersion students produced the highest percentage of uptakes (61%) after recasts. Perdomo (2008) examined learners’ acquisition of auxiliary verb to have and the use of past participle in the present perfect tense. The results indicated that the recast group outperformed the other group who were provided with explicit negative evidence. Al-Surmi (2012) concluded that morpho-syntactic recasts during interaction led to more learners’ subsequent recognition of such recasts. Also in Lee’s (2013) study, recasts produced a high amount (92%) of student repair. According to some studies, different kinds of recasts bring about different results. For example, partial recasts or recasts in which the correction is stressed by the teacher are considered more constructive than recasts in which the entire utterance is reformulated (Doughty & Varela, 1998; Loewen & Philp, 2006).

Despite the above-mentioned advantages, recasts suffer from certain drawbacks. They are assumed to be ambiguous since learners often fail to notice and distinguish them from non-corrective repetitions (Ellis, 2007; Lyster, 1998a; Sheen, 2007). Still other studies have indicated that learners’ interpretation of recasts as feedback is more noticeable for lexical and phonological than morpho-syntactic errors (Egi, 2007; Gass & Lewis, 2007; Lyster, 2001; Mackey, Gass, & McDonough, 2000).

At the other end of the CF spectrum lies explicit feedback. Explicit feedback refers to the explicit indication that a form is incorrect and the "provision of the correct form" (Lyster & Ranta, 1997, p. 46). Some studies have shown that explicit CF is more effective than implicit feedback on learners’ L2 development. For example, in Carroll (2001), direct metalinguistic feedback outperformed other types of correction. Suzuki’s (2004) and Han and Jung’s (2007) studies revealed that explicit correction resulted in better rates of learner repair. Ellis et al. (2006) compared the effectiveness of implicit and explicit CF on low intermediate learners’ performance. Their findings indicated the superiority of explicit feedback over the implicit type for both delayed imitation and grammaticality judgment tests. Varnosfadrani and Basturkmen (2009) compared the effectiveness of explicit and implicit error correction for intermediate-level learners. The results showed higher scores for explicitly corrected learners. Their further analysis showed that developmentally early features (e.g., irregular past tense) were learnt better with explicit correction and developmentally late features (e.g., regular past tense) with implicit correction. Rassaei (2013) found that learners who received explicit correction outperformed both recasts and control groups in writing, untimed grammatical judgment, and error correction tests.

One reason for the superiority of explicit feedback over implicit CF, particularly for low proficient learners, might be that the learners do not sensitively recognize the gap between their interlanguage and the correct forms of a target language (Lin & Hedgcock, 1996). However, some researchers believe that explicit correction is more obtrusive, anxiety provoking, and prone to negative psychological and educational outcomes (Leeman, 2003; Mak, 2011; Trofimovich, Ammar, & Gatbonton, 2007). Among the most important psychological traits in L2 learning is learners’ willingness to take risks and communicate in classroom activities, which is the secondary objective of this study.

PURPOSE OF THE STUDY

Uptake refers to students’ immediate response to feedback. It might result in learners’ modified output which is often considered as the evidence for the efficacy of CF (Egi, 2010; Mackey et al., 2000). However, as Lightbown (2000) stated, any effect must be shown to exist for an extended period of time. Therefore, in this study a long-term and intensive treatment (19 sessions * 50 mins=950 mins) was practiced to examine the actual learning and recall of the target forms.

Moreover, unlike most previous studies and to motivate future related efforts, the focus in this research has been on young learners. Young learners are defined here as school pupils from around 7 to 13 years of age. According to McKay, children learners “are going through a period of social, emotional, and cognitive growth, they are developing literacy and they are highly vulnerable” (2005, p. 256). Therefore, specialist knowledge and particular language input and instruction are required to fulfill their psychological and physiological demands. For instance, scholars concur that tasks should be interesting and motivating, scaffolded by the teacher, of a moderate level of difficulty, and considerably face-saving given the vulnerable psyche of children language learners (Hasselgreen, 2005). Therefore, the second objective of the study was to address young language learners’ WTC before and after feedback provision in order to detect the CF which can most properly minimize their senses of insecurity and frustration.

This study adopted a quasi-experimental design involving a pretest, treatment, and a post-test. It was a mixed-method research in that, in addition to a self-report scale, students’ WTC fluctuations were formatively tracked through the teacher’s observation and completion of a tally chart. CF with two subscales of recast and explicit feedback was the independent variable, while grammar uptake and WTC represented the dependent ones. Age and level of proficiency were controlled, but learners’ individual differences, observer bias (halo effect), and out-of-class-learning experiences were potential confounding variables. The following questions guided the present study:

1. Is there any significant difference between recasts and explicit feedback on young L2 learners’ grammar uptake?

2. Is there any significant difference between recasts and explicit feedback on young L2 learners’ willingness to communication?

METHOD

Participants

The participants were some young L2 learners in two intact classes (19 in the explicit group and 18 in the implicit) at a major language institute in Kerman (Iran). They included 14 males and 23 females with an average age of 10.32 and had already received two years of English instruction at the institute. At the time of data collection, they were studying Level 3 of Up and Away in English series (Crowther, 1997), which corresponds to pre-intermediate level of proficiency. In order to protect the participants’ anonymity, their self-select pseudonames were used and reported all along the study.

The initial number of the students was 37. After data collection, those students with more than four classroom absences as well as those who had failed to answer the tests fully and/or accurately were excluded. As the instruments (see below for the full description of the instruments) were to be studied and analyzed separately of each other, the data screening was also done independently for each target domain; that is to say, if a student, for instance, failed to do the grammar post-test, but filled out the WTC scale fully and accurately, his/her score in the first test was removed from the data pool, while the second one was considered in further analyses. Therefore, upon pruning, the number of the implicit group dropped to 17 and 15 considering its WTC and grammar scores, respectively. The number of the explicit group, however, remained unchanged in both areas. In the qualitative phase (observed WTC) too, after applying class attendance and tests submission criteria, 16 participants in the explicit group and 17 in the implicit remained for further examinations. The remaining sample size met Fraenkel and Wallen’s (2003) and Mackey and Gass’s (2005) guideline which recommends a minimum of 15 to 30 per group in experimental studies.

Instrumentation

For both pretest and post-test sessions, two sets of instruments were used, a grammaticality judgment test (GJT) and a willingness to communicate (WTC) scale. A WTC checklist was also used during class observations, which is described below (The instruments will be available upon request).

GJT and Target Features

Two 30-item parallel grammaticality judgment tests were designed by the researchers. They examined students’ morpho-syntactic knowledge including tenses, auxiliary verbs, prepositions, articles, third person s, and singular/plural verbs. The selection of the target features was guided by two criteria: 1) Corpus analyses have revealed that these target features are among the most problematic areas for Iranian language learners and highly frequent in their productions (e.g., Hayati, Jalilifar, & Bardideh, 2011; Tahririan, 1986); and (2) They were all from the textbooks the learners had practiced in previous semesters and therefore already known to them. Of the thirty sentences in each test, twelve were correct and the rest contained an error. The students were asked to identify the erroneous sentences and provide the correct forms. No time limitation was considered for them to complete the test. The KR-21 measure yielded reliability indices of .69 and .73 for the pre- and post-tests, respectively.

WTC Scale (WTCS)

A 15-item WTC scale was adapted from Xie (2011), which was originally derived from Weaver’s (2005) study. The items concerned students’ perceived willingness to initiate a conversation and engage in communicative tasks on a percentage scale (0-100%). Considering the age and proficiency of the groups, the scale was translated to their native language (Persian) and then back translated by another bilingual person to English. The original and back translated versions were compared and inconsistencies in meaning were discussed until agreement was reached. Using the KR-21 formula, the reliability of the scale was estimated to be .77.

WTC Observation Checklist (WTCOC)

Students’ WTC behaviors were also recorded using a checklist from Cao and Philp’s (2006) study during classroom observations. It included items about the extent to which each student (a) volunteered an answer by, for instance, raising a hand, (b) answered the teacher’s (directed or otherwise) questions, (c) asked questions from the teacher, (d) guessed the meanings of unknown words, (e) tried out a difficult form (lexical or syntactic) in the target language, (f) commented or expressed personal opinions in class, and (g) volunteered to participate in class activities. The checklist was filled out by the teacher using a tally chart worksheet.

Data Collection Procedure

The study was conducted during a whole semester of 19 sessions from July 2014 up to September 2014. The two classes met twice a week, with each session lasting one hour and a half. One of the researchers served as the classroom teacher for both groups. Following a traditional PPP (presentation-practice-production) teaching approach, each session was divided into 4 sections of reviewing, teaching the new lesson, pattern practicing, and homework checking and assigning. In the first section, the previous lessons were reviewed in the form of teacher-initiated questions to be answered by the students. The students received CF for their errors in this phase which took nearly half an hour.

In the second phase, the new lesson was introduced by the teacher getting the students to listen to audio-recorded conversations, practice repetition, and accomplish reading activities and drills. The students were then encouraged to ask the teacher or peers questions regarding the pictorial information in the new chapter. In this part too, the teacher could supply feedback to the students’ erroneous utterances. The last quarter of the class time was devoted to homework evaluation and assignment. Therefore, CFs were provided in two sections of the class (see Table 1 for a summary of class activities per session).

Table 1: Class organization and lesson plan per session

In one group, the students received recasts whenever they used the target structures incorrectly, while the students in the other group received explicit CF. The two error treatments are illustrated in the table below. After reviewing the transcripts, the number of the teacher corrections at the beginning and final sessions was tabulated. As can be seen in Table 2, the frequency was roughly consistent in the explicit group throughout the course, ranging from 49% at early sessions to 44% at later ones. In the recast group, however, the number decreased from an estimate of 62% in the first half to some 37% in the second half of the semester. Another interesting observation was that recasts were generally more frequently distributed (f ≈ 690) than the explicit CF (f ≈ 510) over the whole semester.

Table 2: Groups’ composition in terms of CF provision

The researcher-made GJTs (original and parallel versions) and the WTCS were administered once prior to the main study and once after its completion. In answering the WTCS, considering the age and limited experience of the sample, the researcher offered explanations about the statements by making examples or illustrations. All the sessions were audio recorded and transcribed for subsequent analysis. Students’ interest and participation in classroom activities were further checked through the researcher’s observation and recordings. For this purpose, the teacher recorded each student’s participation (of the kinds expatiated in the instrument section) per session by filling out the tally chart in the WTCOC. She then added up the totals for each student and scored them.

Data Analysis

According to the Kolmogorov–Smirnov normality test, the data distribution met the assumption of normality (p > .05), thereby allowing the use of parametric statistics for quantitative data analyses. Several paired samples t-test and independent samples t-test analyses were conducted for within-group and between-group comparisons, respectively.

In the observation phase, the WTCOC was filled by the teacher (and researcher) in situ (during each session) and then rechecked through listening to the class recordings. The checklists included seven categories representing students’ frequency of participation in class activities. Their total contribution was computed by adding up the number of individual participations. Any disagreement and/or ambiguity was discussed by the researchers until consensus was achieved.

RESULTS

Within-Group Comparisons of WTC

Table 3 below represents the descriptive statistics of the sample and their WTC performance. Of the 19 learners in the explicit group, the minimum scores on WTC were 59 and 47 in pre- and post-intervention conditions, respectively. The maximum score obtained in both tests was 144. The mean performance of the explicit group experienced a drop from 107.42 (SD = 25.70) in the pretest to 106.05 (SD = 29.31) in the post-test.

The composition of the recast group yielded that the minimum WTC score in the pretest was 55 and equal to 92 in the post-test. The maximum scores achieved on pre- and post-tests were 137 and 141, respectively. In contrast to the explicit group, the mean of the implicit group showed an increase from 112.87 (SD = 18.73) in the pretest to 122.27 (SD = 14.99) in the post-test condition.

Table 3: Composition of the sample and their WTC performance

The paired t-test for the explicit group revealed a non-significant difference between their pre- and post-test performances on WTC (t = .310, p > .05). It follows that the WTC of this sample did not change any differently across the two conditions. One interesting finding was that the treatment in this group led to a decreased WTC in the post-test, although the change was found to be statistically non-significant (Table 4).

Table 4: WTC of the groups across testing conditions

Note. SEM = Standard error mean, p > .05

In the recast group, unlike the competing group, WTC increased from pre- to post-test condition. The t-test results, however, yielded that the change in their mean scores was not a significant one (t = -1.810, p > .05). The finding implies that implicit CF could not encourage learners’ willingness to take part in classroom interactions either.

Within-Group Comparison of Grammar Uptake

In order to assess the effect of the intervention on their uptake, both groups sat for two grammar judgment pre- and post-tests. The mean performance of the explicit group raised from some 12.11 (SD = 2.15) to 13.05 (SD = 3.85) across the two testing conditions. This increase was relatively more drastic in the recast group, where the pretest mean score was 12.94 (SD = 2.48) and the post-test score was 16.24 (SD = 4.89) in the post-test (Table 5).

Table 5: Composition of the sample and their GJT performance

Comparison of the mean difference within the explicit group showed a non-significant change across the two tests (Table 6). As illustrated in Table 5, the explicit instruction has led to an increase in their GJT performance. But further inferential analysis (Table 6) failed to support a significant difference (t = -1.093, p > .05).

Another paired samples t-test suggested a significant within-group mean difference in the recast group (t = -3.967, p < .01). It can be concluded then that implicit CF has successfully enhanced grammar uptake of the learners.

Table 6: Grammar uptake of the groups across testing conditions

Note. EM = Standard error mean, p < .01

In sum, based on within-group comparisons, no significant changes were observed in the WTC of the two participating groups. Regarding their grammar uptake, the explicit group showed no significant changes across the two testing conditions. The recast group, however, significantly outperformed in the post-test. In order to get a clearer picture of the two groups’ performances on WTC and grammar tests, some between-group analyses were in order (see below for the results).

WTC and Grammar Uptake: Intergroup Comparisons

Table 7 summarizes the results of an independent-samples t-test. The Levene’s test showed that the variance between the two groups’ mean scores could be assumed as equal (F = 1.105, p > .05). Further analysis of the two groups’ grammar post-tests revealed that the difference was a significant one (t = 2.180, p < .05). Therefore, comparing the two groups’ post-test performances yielded that the implicit group made more significant grammar gains than the competing group.

Table 7: Cross-comparison of the two groups’ GJT and WTC post-tests

Note. SED = Standard error difference, * p < .05

The difference between the WTC scores of the two groups was also examined. The preliminary Levene’s test disproved the equality of the variances of two treatment groups (F = 9.871, p < .05). The results showed that the implicit group’s WTC significantly outweighed that of the contrast group (t =2.090, p < .05). It implies that, all other things being equal, those who received recasts proved in the end to be more cooperative and willing to communicate in class.

In conclusion, between-group comparisons could disclose findings not achievable through the other analysis expatiated in the previous section. It was demonstrated that the recast group outperformed in both areas of WTC and grammar uptake.

Observation Results

The graphs and tables below show the students’ amount of participation during the semester. As the intermediate sessions (Sessions 9 up to 11) in both groups were allotted to practice activities and midterm exam, in which classroom participation was either minimal or equally distributed among the students, they were opted out of further analysis.

Table 8 depicts the classroom performance of each explicit group member over 19 sessions. The absent sessions were assigned missing values in the computations.

Table 8: Explicit group’s WTC across the sessions

Note. a=absent

The graph for the explicit group (Figure 1) turned out to be a bimodal curve. There are a few outliers with scores above 20 at around sessions 5 through to 7 and from session 12 up to 14, suggesting a sort of inconsistency across the sessions. Moreover, as the figure represents, the group’s participation and contribution to classroom activities has dropped towards the end of the course, most likely as a result of the corrective feedback they have received along the instructional sessions.

Figure 1: Explicit group’s WTC pattern

Table 9 below presents the recast group’s WTC fluctuations over the 19 sessions with the middle sessions (highlighted area) being excluded from further analysis.

Table 9: Recast group’s WTC across the sessions

Note. a=absent

Figure 2 schematically presents the implicit group’s WTC rate, where a comparatively more consistency is observed in classroom participations throughout the course. The graph is a unimodal one, suggesting that for a certain while (from session 3 up to around 14) the subjects have performed systematically and equally well (above the average). Considering the fact that classroom opportunities and requirements vary to a degree from one session to another, these slight fluctuations do not seem unexpected.

Figure 2: Implicit group’s WTC pattern

Overall, then, the results of the qualitative phase were in line with the quantitative analyses. The recast group outweighed the explicit group in its tendency towards class participation in both phases. Caution, however, should be practiced in making claims about the strength of this difference.

DISCUSSION

This study examined the effect of explicit and recast corrective feedbacks on young L2 learners’ grammar uptake and willingness to communication. The results of the grammar post-tests revealed that implicit CF was more effective than explicit correction on the development of L2 knowledge. This finding is on par with those of Lyster (2001), Lyster and Ranta (1997), Perdomo (2008), and Sheen (2004). It contradicts, however, with the findings of some other researchers like Ellis et al. (2008), Varnosfadrani and Basturkmen (2009), and Rassaei (2013).

Regarding the effect of feedback on young L2 learners’ WTC, the findings were also in favor of implicit correction. The t-test results showed that recasts lead to students’ willingness to participation in classroom activities and communicative tasks more than the explicit CF. This finding is compatible to the idea of some researchers who believe that explicit correction is anxiety provoking and might lead to negative psychological outcomes (Leeman, 2003; Mak, 2011; Trofimovich et al., 2007).

One can argue that the findings of the study are also verification to the traditional claim made by Krashen (1982) that learners learn best through unconscious acquisition and that conscious learning does not lead to acquisition. However, this study made use of partial recasts which, as Ellis et al. (2006) argue, may not be considered as an implicit or even a relatively implicit technique. In partial recasts, only the erroneous part of the sentence is repeated by the teacher, which helps learners to locate the error. Partial recasts are considered to be more effective than full recasts by many researchers (Doughty & Varela, 1998; Kim & Mathes, 2001; Loewen & Philp, 2006; Rassaei, 2013).

Another reason for the superiority of the implicit correction in the present study might be the rather long treatment the students experienced (950 mins over 19 sessions). In most studies in which explicit CF had a better effect over the implicit one, the length of the treatment was very short (e.g., Ellis et al., 2006; Rassaei, 2013). Therefore, we may conclude that although explicit correction might lead to immediate L2 development (i.e. intake), as some other studies have shown recasts have a more enduring effect (Rassaei & Moinzadeh, 2014; Shirazi & Sadighi, 2012).

In addition, contrary to most studies which compared the effect of the two feedback types on adults, in this study we had a sample of young learners. Standing at the implicit end of CF continuum, recasts are characterized by saving class time and being less than any other feedback threatening to the learners’ self-confidence (Loewen & Philp, 2006). They are, according to Ellis and Sheen (2006), both communicative and didactic treatments, in that they are comparatively less disruptive to the flow of communication, on the one hand, and are “supportive, scaffolding help”, on the other (Choi & Li, 2012, p. 344). Therefore, given the design and the participants’ characteristics in this study, one plausible explanation for the results could be age factor. Young learners are characterized by being more affectively susceptible, at the early stages of self-empowerment, and in need of motivation especially offered externally by teachers, parents, and peers. These features should in turn affect the “instructional choices of language teachers in young learner programs” (Inbar-Lourie, 2010, p. 351).

Consequently, the effectiveness of recasts, as indicated in this study, could be accounted for (a) linguistically by their inherently unobtrusive and communication-friendly properties and (b) affectively as a result of being less face-threatening and anxiety-provoking than many other corrective moves. There is also some actual evidence in support of this empirical finding: The researcher who served as the classrooms teacher witnessed that some students were uninterested in and complained about being corrected explicitly halfway throughout the course. When asked for the reason, majority of them referred to peer pressure. They contended that their peers mistreated them for such goofs when noticed by the teacher, leading them to feel nervous and frustrated.

CONCLUSION AND IMPLICATIONS

The study aimed at detecting the optimal corrective practice for young language learners. Although both explicit and implicit feedbacks are useful in L2 development, explicit error correction might be discouraging and demotivating for young language learners. This assumption was reinforced in this study since a less direct and more tactful (implicit) corrective move was more fruitful in raising children’s unconscious grammar knowledge and their willingness to participation in classroom interactions. This finding lends further evidence in support of the role of learner characteristics and individual differences (e.g., age) in L2 development. By taking these characteristics into account in educational systems, according to Pawlak (2014) and Ghahari and Ahmadinejad (2016), the number of educational decisions and instructional choices will be considerably increased. As it was indicated here, despite the recurrent contribution of explicit CF to morpho-syntactic development in the literature, it did not work equally well for children learners. Motivated by these findings, teachers of young language learners are encouraged to

(1)   delay or minimize error correction until learners are emotionally and linguistically prepared,

(2)   correct selectively by critically evaluating the weight and salience of the errors before treatment,

(3)   consider the type of activity since error correction during fluency-based tasks (e.g., conversations and conferences) is not as constructive and encouraging as in accuracy-based activities (e.g., drills and exercises), and

(4)   treat errors in a step-wise and developmental manner, moving from implicit CF provision at early stages (children learners) towards explicit correction at later stages (adult learners).

Finally, as a classroom research, this study suffers from some limitations including the use of intact group design and absence of a control group. Future research is invited to control for the intervening impacts of these variables.