1
The Relative Effectiveness of Mixed, Explicit and Implicit
Feedback in the Acquisition of English Articles
Abstract
This article reports on a study that compares the effectiveness
of three oral negative feedback types in the acquisition of
English articles. A mixed feedback treatment, in which
learners' errors were corrected, first, through explicit
feedback, then through implicit feedback, was compared to
implicit-only (i.e., recasts) and explicit-only (i.e., explicit
correction) feedback treatments. The study followed a
pretest/posttest/delayed posttest experimental design. Eighty
English-as-a foreign language learners, whose proficiency level
ranged from beginner to intermediate, were randomly assigned
into explicit-only feedback, implicit-only feedback, mixed
feedback, reduced-explicit feedback and no-feedback control
groups. The learners and researcher met two times for the
treatment sessions. In each treatment session, the learners and
researcher carried out three oral production tasks, where
learner errors on the English indefinite and definite articles
were treated according to their group assignments. The
2
differences between the groups in performance were measured
using oral production tasks. The results of the study revealed
that the explicit-only feedback and mixed feedback groups
outperformed all the other groups on the immediate posttest,
but there was no difference between the explicit-only and mixed
feedback groups on either of the posttests. This result was
interpreted as indicating that mixed-feedback can be as
effective as explicit-only feedback.
3
1. Introduction
In the field of Second Language Acquisition (SLA), recent
years have witnessed increased research interest in the
investigation of the question of what type of negative feedback
is more effective for L2 development. A group of researchers
has advocated for implicit feedback types that are less likely
to interrupt the flow of communication (e.g., Doughty, 2001;
Doughty and Williams, 1998; Long, 2007; Long and Robinson,
1998). Another group (e.g., Carroll, 2001; Lyster, 1998) has
argued in favor of feedback types that are more explicit,
asserting that the relative salience of explicit feedback can
make it easier for learners to recognize the corrective intent
behind them, which is a necessary step to benefit from negative
evidence according to some researchers (e.g., Carroll, 2001).
Previous research has shown that groups receiving explicit
feedback tend to outperform groups receiving implicit feedback
(Author, XXXX; Carroll and Swain, 1993; Ellis, 2007; Ellis,
Loewen and Erlam, 2006; Sheen, 2007). Despite these results
speaking to the general effectiveness of explicit feedback,
relatively little is known as to what extent a feedback
treatment including both explicit and implicit feedback would
4
be effective. An examination of the comparative effects of a
mixed feedback treatment involving both explicit and implicit
feedback to feedback treatments consisting of only one of the
feedback types is important because it could shed light on 1)
the extent to which explicit feedback is necessary to invoke
the desired level of salience, and 2) whether the salient
nature of explicit feedback can help learners interpret
subsequent implicit feedback more efficiently.
2. Literature Review
Disagreement exists about the role of negative feedback in
second language (L2) acquisition. Some researchers (e.g.,
Schwartz, 1993; Truscott, 1996) have argued that negative
feedback can only produce a superficial form of L2 knowledge,
not affecting learners’ competence. The majority of researchers
(e.g., Gass, 1991; Long, 2007), however, does believe that
negative feedback has potential for affecting learners’
competence. Previous meta-analyses comparing groups that
received oral negative feedback to groups that did not receive
feedback have provided supportive evidence for this second
position by showing that negative feedback has a medium-to-
large effect on L2 acquisition (Li, 2010; Lyster and Saito,
5
2010; Mackey and Goo, 2007). Aside from the absolute effect of
negative feedback, researchers have also been interested in the
relative effectiveness of different feedback types. Although
there are different ways to classify feedback types (e.g.,
Lyster and Saito, 2010), a widely accepted classification
method involves deciding where along an explicit/implicit
continuum feedback types fall (Ellis, 2001). Two features,
metalinguistic information and directness, play role in the
decision of whether a feedback type leans towards the explicit
or implicit end of the continuum. Feedback types conveying
metalinguistic information (e.g., metalinguistic feedback)
and/or directly informing learners about the accuracy of their
utterances (e.g., explicit correction) have been considered
explicit. The ones, however, lacking metalinguistic
information and/or directness (e.g., clarification requests and
recasts) have been considered implicit.
Recasts, broadly defined as targetlike reformulations of
leaners’ non-targetlike utterances, are considered a form of
implicit feedback. Descriptive studies have shown that recasts
are the most frequently occurring feedback type in language
classrooms (Ellis, Basturkmen and Loewen, 2001; Lyster and
6
Ranta, 1997; Panova and Lyster, 2002; Sheen, 2004). For
example, Sheen (2004) reported that recasts accounted for 60%,
whereas explicit feedback (i.e., metalinguistic or explicit
correction) accounted for between 5.8% and 7% of all the
feedback moves provided in different contexts (i.e., second
language, foreign language, immersion). This difference
between implicit and explicit feedback in frequency could be
attributed to teachers’ feedback preferences. Seedhouse (2001)
carried out a conversation analytic study on a database of 330
lessons or fragments of lessons and concluded that teachers
preferred using non-threatening and mitigated implicit feedback
such as recasts.
Despite being a preferred form of feedback, recasts have
been found to be less effective than more explicit feedback
types. Several descriptive studies have carried out a frequency
analysis of learners’ reactions to teachers feedback moves,
i.e., uptake, as a measure for feedback effectiveness. Although
it has been shown that learner uptake and the extent to which
learners benefit from feedback (as measured by posttests) might
not necessarily be related (Mackey and Philp, 1998), the
presence of uptake, at least, indicates that the learner has
7
noticed the corrective intent behind the feedback (Mackey, Gass
and McDonough, 2000). Sheen (2004) reported that the average
uptake rates for explicit feedback types (explicit correction,
63%, and metalinguistic feedback, 83%) were considerably higher
than the average uptake rate for recasts (48.2%).
Experimental and quasi-experimental studies investigating
the relative effectiveness of feedback types that differ in
metalinguistic information have revealed that whenever there is
a statistical difference between recasts and metalinguistic
feedback, this difference is in favor of metalinguistic
feedback (Carroll and Swain, 1993; Ellis, 2007; Ellis et al.,
2006; Sheen, 2007). This result, however, should be interpreted
cautiously in light of the following two observations. First,
the findings were not consistent across studies. For example,
Loewen and Nabei’s study (2007), where 66 EFL learners received
recasts, metalinguistic feedback, or clarification requests,
did not show any differences between the groups. Second, there
was considerable methodological variability among the studies:
treatment tasks employed, operationalizations of feedback
types, outcome measures, target structure choice, design
features (e.g., whether learners are assigned randomly to
8
experimental groups), number of treatment sessions, duration of
the treatment sessions, consistency in providing feedback for
each error (see Author, XXXX and Ellis et al., 2006 for an
extended list of features).
Much less research has been conducted to compare the
relative effectiveness of oral feedback types that differ in
directness. The only study to date is that by Author (XXXX),
who investigated the relative effectiveness of explicit
correction (direct) versus recasts (indirect), using an
experimental design with two posttests (immediate and delayed).
Forty-eight participants self-selected themselves for the study
on the basis of (1) being a native speaker of English, (2) not
having taken any linguistics course, and (3) not having been
exposed to Turkish. The participants were asked to learn 50
Turkish words in order to carry out the rest of the study.
After learning the words, they carried out two communication
games with the researcher. Learner errors on two Turkish
morphemes (i.e., locative and plural) were corrected according
to group assignment. Oral production, comprehension and
recognition tests were administered to measure learners’
resulting performance. Results revealed that explicit
9
correction was more effective than recasts on the oral
production and comprehension tasks. When these results are
considered together with the results of the previous studies
that compared metalinguistic feedback versus recasts, it can be
claimed that not only explicit feedback in the form of
metalinguistic feedback but also explicit feedback in the form
of explicit correction is more effective than recasts.
Researchers have attributed the positive results for
explicit feedback to its relatively more salient nature.
Carroll (2001), among others (e.g., Lyster and Ranta, 1997),
has argued that in order for feedback to be beneficial for
learners, learners should recognize the corrective intent so
that they could shift their attention from meaning to form.
Carroll hypothesized that the recognition of the corrective
intent is facilitated by a failure to find relevance between
the feedback and the ongoing topic of the conversation. Drawing
on Carroll’s ideas, Author (XXXX) pointed out that the explicit
correction used in his study, which included explicit verbal
indicators (i.e., x is wrong, you should say y), could have made
the corrective intent more salient because it was not relevant
to the ongoing talk.
10
In sum, research on the relative effects of explicit and
implicit feedback has shown that explicit feedback types are
generally more effective and that this effect might be due to
the way explicit feedback makes the corrective intent salient
for the learner. However, what previous research has not been
able to show so far is the extent to which each feedback
instance in a feedback treatment should be explicit in order to
reach the desired level of effectiveness. Should an learners
receive explicit feedback whenever they produce erroneous
utterance? Or would it be sufficient for learners to receive a
few initial explicit feedback instances and then implicit
feedback instances? One way of seeking answers to these
questions is using a research design where a mixed feedback
treatment including both explicit and implicit feedback can be
compared to single feedback type treatments (i.e., explicit-
only or implicit-only). So far, no attempt has been made to
investigate the effect of a mixed feedback treatment on L2
development by treating it as part of a feedback type variable.
Previous experimental and quasi-experimental studies have
investigated the relative effectiveness of the feedback types
by exposing learners to a single feedback type. Each learner
11
throughout these studies received either explicit or implicit
feedback depending on their experimental group. To address this
gap in the negative feedback research, the present study aims
to examine the effectiveness of a mixed feedback (explicit +
implicit) treatment relative to implicit-only and explicit-only
feedback treatments. In such an investigation, it is important
to tease apart whether any effect for the mixed treatment is
attributable to explicit feedback only or to both explicit and
implicit feedback. For this reason, the use of a fourth group
(i.e., reduced-explicit group) receiving as many explicit
feedback instances as the mixed group without any implicit
feedback is necessary. If one can show that the mixed group
performs significantly better than the reduced-explicit group,
one can deduce that explicit feedback alone cannot be
responsible for the overall effectiveness of the mixed feedback
treatment. The research questions were the following:
1- Is there a difference between the oral performance of
the group that receives explicit feedback (i.e., explicit-
only group) and the oral performance of the group that
receives implicit feedback (i.e., implicit-only group)?
12
2- Is there a difference between the oral performance of
the group that receives both explicit and implicit
feedback (i.e., mixed group) and the oral performance of
the implicit-only group?
3- Is there a difference between the mixed and explicit-
only groups in oral performance ?
4- Is the oral performance of the mixed feedback group
attributable to the explicit feedback instances provided
during the mixed feedback treatment?
3. Method
3.1. Participants and Context
One hundred twenty students EFL students who were taking
either a 3-hour general English course or a 6-hour English
Speaking course with the researcher volunteered for the study.
The general English course is a requirement for all first-year
university students in Turkey. The course targeted beginner
level students and aimed to improve their vocabulary knowledge
and conversational skills. The English Speaking course was one
of the four courses that the preparatory year students in the
English Language Teaching program are required to take.
13
Throughout this course, the learners were engaged in group and
whole-class discussions, individual and group presentations as
well as communicative tasks on topics presented by the course
book. Neither course included any formal grammar teaching. Out
of 120 students who volunteered to participate, 40 students
were screened out because they scored higher than the preset
criterion (20%) on the pretest. The average age of the
remaining 80 participants (28 male and 52 female) was 19.77 (SD
= 1.01). Participants took 7.88 years of English language
instruction on average (SD = 1.82). To measure learners’ general
level of English, the Michigan English Placement Test (EPT) was
administered. The results of the EPT revealed that the levels
of the participants ranged between beginner and intermediate (M
= 35.42, SD = 9.49). Participants were pursuing undergraduate
degrees in math, science, preschool education or English
language teaching.
3.2. Target structure
The English definite and indefinite articles were selected
as the target linguistic structures for the present study. The
motivation for this selection was that English articles have
been considered an important learning challenge for learners
14
whose first languages lack articles (García Mayo and Hawkins,
2009). The L1 of the participants in this study was Turkish,
which does not have morphological determiners (Öztürk, 2005)i.
Master (2002) discusses three factors contributing to this
challenge. First, the fact that the articles occur very
frequently in English makes the consistent application of
conscious rules very difficult. Second, noticing the articles
in the spoken mode is challenging because they are unstressed.
Finally, articles resist learners’ tendency to look for a one-
to-one overlap between form and function because they serve for
multiple functions.
Because the choice of an article depends on multiple
factors (e.g., linguistic and pragmatic factors), following
Sheen (2007), the current study has limited the obligatory
contexts for each article to one major function of the
indefinite and definite articles: a as first-mention and the as
anaphoric reference (e.g., we made a cake yesterday. My brother
liked the cake very much). Simply put, the rule can be
expressed as follows. Noun phrases referring to entities that
are introduced in the discourse are preceded by an indefinite
article (first-mention), whereas a particular entity that has
15
already been introduced in the discourse is preceded by a
definite article (anaphoric reference).
The learners participating in the study did not receive
explicit instruction or corrective feedback on English articles
in their regular classrooms during the semester they
participated in the study.
3.3. Research design
The study followed a pretest/posttest/control group
between-subjects design with a delayed posttest. Participants
were randomly assigned to one of five conditions: Explicit-only
(N = 16), implicit-only (N = 16), mixed (explicit + implicit;
N = 16) reduced-explicit feedback (N = 16) and no-feedback
control (N = 16) groups. The independent variable was feedback
group, and the dependent variable was oral performance as
measured by oral production tasks.
3.4. Materials
Different versions of an Oral Production Elicitation Tool
(OPET) were used in the tests and feedback sessions. Each OPET
includes three distinct oral production tasks. It has been
argued that learner performance in articles varies depending on
the task used (Muranoi, 2000). Therefore, as an attempt to
16
obtain a more reliable representation of learners’ oral
language performance on the target structures, the study
measured oral performance in as many different contexts as
feasible for the study. In order to form different OPETs,
different versions of each individual oral production task were
developed. These oral production tasks were a guided oral
production task, a spot-the-difference task and a story-
retelling task. Task stimuli in each of these individual tasks
were presented through computers.
In the guided oral production tasks (GOP), learners
completed a communication game with the researcher. In order to
fill in the information missing from his task sheet, the
researcher had to elicit some answers from the learner. The
learner’s task was to provide the researcher with the necessary
information by looking at a series MS Power Point slides
containing different objects. The learners saw 24 objects in
two different contexts. They were instructed to name the object
when they saw it for the first time and to talk about one of
the three features (i.e., color, size and price) of it when
they saw it for a second time. These questions created 24
obligatory contexts for the use of each target structure.
17
In the spot-the-difference task (SD), the learner and
researcher performed an information gap task that was based on
two room pictures with eight objects. The learner was asked to
describe the picture to help the researcher find the
differences between the two pictures. Although spot-the-
difference tasks have been generally considered two-way
information gap tasks in the SLA literature (e.g., Gass, Mackey
& Ross-Feldman, 2005) in this study, the information flow was
kept one way in order to prevent learners from receiving
positive evidence. As a result, the researcher’s task was to
identify whether the dyad had the same objects in their
pictures, and if they did, whether their locations were the
same. The SDs created eight obligatory contexts for the use of
each target structure.
The story retelling task (RET) asked learners to retell a
story after reading its L1version on a computer screen for ten
seconds. A pilot testing revealed that this time limit was long
enough for learners to comprehend the story and short enough to
keep them away from memorizing the story word for word. Then,
they were asked to retell the story in English using a list of
18
English words. Each story created 4-5 obligatory contexts for
the use of each article. Each OPET included two stories.
3.5. Procedure
To prevent lack of vocabulary knowledge from interfering
with task performance, learners were provided with a list of 90
English words prior to the study and asked to study it. No
attempt was made to include words that were relevant for the
courses that the learners were enrolled in. The words were
retrieved from MRC Psycholinguistic Database (Coltheart, 1981)
using the parameters for the following word properties: number
of letters, 3 – 8; number of syllables, 1-2 ; Brown verbal
frequency, 1-1500; concreteness, 600-700; imageability, 600-
700. These properties were selected to ensure that the words
can easily be depicted and to prevent the difficulty in
learning or producing the words from interacting with learners’
performance on the target structures. At the beginning of each
meeting, learners were asked to say the English name of the
objects that were shown to them through a PowerPoint slide
show. Learners could proceed to the next stage if they were
100% accurate in this task. In the first meeting, following the
vocabulary test, the learner and researcher performed two
19
OPETs, one for the pretest and one for the first treatment
session. The dyad met for a second time 7-10 days after the
first meeting to perform two more OPETs (one for the second
feedback session and one for the immediate posttest). Two
months after the second meeting, the dyad met for the third
meeting and carried out another OPET as the delayed posttest.
Also in the third meeting, learners filled out a background
questionnaire.
During the OPETs that were administered as tests (i.e.,
pretest, immediate and delayed posttests), the dyad interacted
as much as necessary, but no feedback was provided. During the
OPETs that were used as feedback treatment, the learners in the
explicit-only, implicit-only, mixed and reduced-explicit groups
received negative feedback when they produced non-targetlike
utterances in article use. The feedback provided to these
learners was of two types: explicit or implicit. Explicit
feedback was operationalized through explicit correction.
Explicit correction was defined as the presentation of the
targetlike reformulation of the erroneous segment of the
learner’s production in a direct manner. In explicit
correction, the incorrect segment of the learner’s production
20
is, first, directly rejected together with the noun following
it, and then, the correct alternative is directly indicated
together with the noun following the article (see Table 1). The
learners’ L1 was used in the explicit correction because of the
learners’ low proficiency level. Implicit feedback was
operationalized through recasts. A recast was defined as the
targetlike reformulation of the erroneous segment of the
learner’s production . The reformulated utterance also
contained the noun following the article. This recast type has
been referred to as partial recasts in the negative feedback
literature (Loewen and Philp, 2006). Feedback was provided
without any special intonation.
<Insert Table 1 about here>
The feedback groups received either only one type of
feedback, as in the explicit-only, implicit-only and reduced-
explicit groups, or they received a combination of both as in
the mixed group (see Table 3). Below is a description of each
condition.
Explicit-only. The learners in the explicit group received
explicit feedback in the form of explicit correction whenever
they produced a non-targetlike utterance throughout the tasks
21
<Insert Table 2 about here>
Implicit-only. The learners in the implicit group received
implicit feedback in the form of recasts whenever they produced
a non-targetlike utterance throughout the tasks.
Mixed. Mixed feedback was defined as the combination of
explicit and implicit feedback, where implicit feedback
followed explicit feedback (Table 3). Learners’ non-targetlike
utterances in the first two obligatory contexts for each
structure in each individual task were corrected through
explicit feedback, whereas learners’ non-targetlike utterances
in the remaining obligatory contexts were corrected through
implicit feedback. The decision of providing explicit feedback
only in the first two obligatory contexts was based on the
prediction that two initial instances of explicit feedback
would be enough to obtain the desired level of salience.
Reduced Explicit. The learners in the reduced-explicit group
received explicit feedback on their non-targetlike utterance in
the first two obligatory contexts in each task (Table 3). The
remaining non-targetlike utterances were left uncorrected.
22
No Feedback. In the no-feedback group, the learners
received no feedback but performed the same tasks as the
learners in the other groups. All sessions were audio recorded
for analysis.
<Insert Table 3 about here>
3.6. Scoring
Each task was coded for obligatory contexts, correct
response in obligatory contexts, and suppliance in non-
obligatory contexts. Another rater listened to and coded 10% of
the recorded data from each task. Interrater reliability was
measured by both percentage agreement and Cohen’s Kappa
coefficient (where applicable). Interrater reliability was
acceptable (see Table 4). Next, a target-language use (TLU)
score for each learner was calculated for each target structure
and test following the formula explained in Pica (1983). When
calculating the scores, the individual tasks constituting OPETs
were considered together. Accordingly, correct responses were
added, and then, the sum was divided by the sum of number of
obligatory contexts and suppliance of the target structure in
23
non-obligatory contexts, in which the use of the target
structure was inappropriate.
<Insert Table 4 about here>
4. Results
The descriptive statistics, as shown in Table 5 below,
revealed that the mean definite and indefinite scores for each
treatment group were close to each other on the pretest. The
scores of the feedback groups started to differ on the
immediate posttest. On the delayed posttest, there is a
noticeable decrease in the performance of all the feedback
groups. The highest scoring groups were the mixed and explicit-
only in both structures and on both posttests. Before running
the necessary tests to find out whether there were statistical
differences between the groups, pretest, immediate and delayed
posttest scores on both structures were checked for normality.
Pretest scores failed to conform to normality as indicated by
the results of a series Kolmogorov-Smirnov (KS) tests for both
the definite (explicit-only, Z(16) = 1.91, p < . 01; implicit-
only, Z(16) = 2.07, p < . 01; mixed, Z(16) = 1.92, p < . 01;
no-feedback, Z(16) = 1.92, p < . 01; reduced-explicit, Z(16) =
1.57, p = . 02) and the indefinite (explicit-only, Z(16) =
24
2.07, p < . 01; implicit-only, Z(16) = 2.15, p < . 01; mixed,
Z(16) = 2.15, p < . 01; no-feedback, Z(16) = 2.04, p < . 01;
reduced-explicit was constant). Therefore, a Kruskal-Wallis
test, a non-parametric equivalent of a one-way ANOVA, for each
structure was carried out to check whether groups differed on
their pretest scores. Neither tests revealed a statistical
difference among groups, indefinite, χ2(4)=2.63, p = .62,
2= .03; definite, χ2(4)=2.32, p = .68, 2= .03. These results
indicate that all groups began on an equal footing with respect
to their pretest scores on both target structures.
<Insert Table 5 about here>
Next, immediate and delayed posttest scores on each of the
target structures were checked for normality. It was found that
data largely conformed to normalityii. Then, two separate
Bonferroni-adjusted, one-way ANOVAs for each target structure
were run. The ANOVA with the immediate posttest indefinite
scores as the dependent variable and Group as the independent
variable revealed an effect for Group F(4, 75) = 26.88, p
< .001. Tukey’s post hoc multiple comparisons revealed that the
explicit-only and mixed groups outperformed all the other
groups with large effect sizes (ranging from d = 1.05 to d =
25
5.03), but the difference between the two was not statistical
and small in effect size, d = .24 (see Table 6 for the post hoc
results). It was also shown that the implicit-only and reduced-
explicit groups outperformed the no-feedback group with large
effect sizes (implicit-only, d = 2.05; reduced-explicit, d =
1.98), but there was no statistical difference between the
implicit-only and reduced-explicit groups (in fact, the mean
scores of these groups were the same). The ANOVA with the
delayed indefinite scores as the dependent variable and Group
as the independent variable revealed an effect for Group F(4,
75) = 9.06, p < .001. Tukey’s post hoc multiple comparisons
showed that the explicit-only, mixed, implicit-only and
reduced-explicit groups outperformed the no-feedback group with
large effect sizes (ranging from d = 1.63 to d = 2.91). The
scores of the explicit-only and mixed groups were not
statistically higher than the reduced-explicit and implicit
groups at this time even though the mean scores of the
explicit-only and mixed groups were still higher than the mean
scores of the reduced-explicit and implicit-only groups. The
magnitude of the difference between the mixed and reduced-
explicit as well as the magnitude of the difference between the
26
mixed and implicit-only groups were noteworthy as shown by
medium effect sizes (d = .55 and d = .72, respectively).
<Insert Table 6 about here>
The ANOVA with the immediate posttest definite scores as
the dependent variable and Group as the independent variable
revealed an effect for Group, F(4, 75) = 45.01, p < .001.
Tukey’s post hoc multiple comparisons (see Table 6) showed that
the explicit-only and mixed groups outperformed all the other
groups with large differences between the scores (effect sizes
ranging from d = 1.17 to d = 14.41), but there was no
statistical difference between the two, with the magnitude of
the difference being small (d = .15). The implicit-only and
reduced-explicit groups outperformed the no-feedback control
group with large effect sizes (implicit-only, d = -3.15;
reduced-explicit, d = 2.29), but there was no statistical
difference between the two, with the magnitude of the
difference being small (d = .20). The ANOVA with the delayed
posttest definite scores as the dependent variable and Group as
the independent variable revealed an effect for Group, F(4, 75)
= 9.70, p < .001. Tukey’s post hoc multiple comparisons showed
that the explicit-only, mixed, implicit-only and reduced-
27
explicit groups outperformed the no-feedback group with large
effect sizes (values ranging from d = 1.22 to d = 2.54). On the
delayed posttest, the scores of the explicit-only and mixed
were not statistically higher than the scores of the reduced-
explicit and implicit-only. However, the mean scores for the
explicit-only and mixed groups were still larger than the mean
scores for the reduced-explicit and implicit-only groups, and
the magnitude of these differences were noteworthy as shown by
medium effect size values (ranging from d = .39 to d = .75).
Finally, the difference between the implicit-only and reduced-
explicit groups was not statistical and small in effect size, d
= .29.
<Insert Figure 1 about here>
<Insert Figure 2 about here>
5. Discussion
The results of the study revealed that all feedback groups
outperformed the no-feedback control group in both structures
on both posttests with large effect sizes (d values ranging
from 1.22 to 14.41). This finding is in line with previous
findings that negative feedback groups consistently performed
28
better than no-feedback control groups (Author, XXXX; Ammar and
Spada, 2006; Carroll, 2001; Carroll and Swain, 1993; Doughty
and Varela, 1998; Ellis et al., 2006; Loewen and Nabei, 2007;
Mackey and Philp, 1998; Sheen, 2007). The first research
question of the study was about the relative performance of the
explicit and implicit feedback groups. Results showed that the
explicit feedback was more effective than the implicit feedback
on the immediate posttest, but not on the delayed posttest. The
magnitude of the difference between the explicit and implicit
feedback (d= 1.07) was large. This finding is in line with the
findings of the previous studies that have shown that learners
benefit more from explicit feedback than implicit feedback
(Author, XXXX; Ellis, 2007; Ellis et al. 2006; Sheen, 2007).
However, as mentioned earlier, most of the previous studies
(Ellis, 2007; Ellis et al., 2006; Sheen, 2007) operationalized
explicit feedback through metalinguistic feedback in their
comparisons. Therefore, the consistency between the findings of
AUTHOR (XXXX) and the findings of the current study becomes
especially important because although these two studies
targeted different structures and languages, they
operationalized explicit and implicit feedback in the same way
29
(e.g., explicit correction and recasts). In addition, the fact
that the magnitude of the difference between the feedback types
in AUTHOR’s study (oral production, d = 1.00; recognition, d
= .87) was considerably close to the one found in this study (d
= 1.07) adds validity to the findings of these studies.
The second and third research questions asked about the
performance of the mixed feedback relative to the implicit and
explicit groups, respectively. It was found that the mixed
group performed significantly better than the implicit-only on
the immediate, but not on the delayed posttest. In addition,
results revealed that there was no statistical difference
between the mixed and explicit-only groups. Based on these
results, it can be argued that the mixed group performed as
well as the explicit-only group. Both groups performed better
than the implicit-only on the immediate posttest, and there was
no statistical difference between the two on both posttets. The
fourth research question asked whether the oral performance of
the mixed group could be attributed to the explicit feedback
instances provided during the mixed feedback treatment. Because
the mixed group outperformed the reduced-explicit group, it is
possible to claim that the resulting performance of the mixed
30
group could not be due to the few instances of explicit
feedback that the reduced-explicit and mixed groups had in
common. As Appendix D shows, the learners in the mixed group
continued to make errors after receiving explicit feedback and
received implicit feedback on these errors ( indefinite, M =
10.69; definite, M = 9.31). It seems, then, most probable
that the resulting performance was due to the combined effect
of explicit and implicit feedback instances.
A possible reason for this finding may be that the
salience of the implicit feedback (i.e., recasts) in the mixed
feedback treatment increased due to the salience of the
explicit feedback. When salience is viewed as the ease of
noticing the negative evidence and the correct form conveyed
through feedback, explicit feedback has a clear advantage over
implicit feedback as evidenced by the superior performance of
the explicit-only group over the implicit-only group. Part of
this advantage might have come from the overt indicators (e.g.,
X is wrong, you should say Y) used in the explicit correction.
These indicators might have helped learners realize that the
feedback was irrelevant to the topic of the ongoing
conversation, which is considered a necessary step by Carroll
31
(2001) in order for learners to interpret the feedback as a
comment on the form of the utterance. It could be that once
learners came to be aware, through the initial explicit
feedback instances, that comments coming from the researcher
were not related to the topic of the task but about the
targetlikeness of their utterances, then, this awareness might
have facilitated the interpretation of the recasts provided
afterwards. In other words, recasts could have worked like an
elliptical discourse device. In ellipsis, some part of a
sentence is left out, but this does not interfere with its
comprehension because the learner can fill in the missing
pieces. Therefore, although the implicit feedback lacked the
directness of the explicit feedback, previously provided
explicit feedback instances might have caused learners to gain
the maximum benefit out of the implicit ones. One could argue,
then, that the effect of explicit feedback on recasts in a
mixed feedback treatment is similar to the effect of several
previously reported factors on recasts. Previous research on
recasts has shown that factors such as length, number of
changes, type of intonation, communication mode moderate the
effectiveness of recasts by increasing their salience (Author,
32
XXXX; Egi, 2007; Loewen and Philp, 2006; Sheen, 2006). Viewed
from this perspective, the provision of explicit correction
before recasts can be considered another factor that increases
the salience of recasts.
The finding that mixed feedback is comparable to explicit
feedback in effectiveness has an important pedagogical
implication. Previous descriptive classroom studies have
reported that recasts are the most frequent feedback type,
whereas explicit correction is one of the least frequent
(Ellis, et al., 2001; Lyster and Ranta, 1997; Oliver and
Mackey, 2003; Panova and Lyster, 2002; Sheen, 2004). A
potential underlying factor for this difference in frequency
could be teachers’ feedback type preferences. For example, some
teachers are reported to avoid the use of explicit feedback not
to undermine learners` confidence because they consider
explicitly pointing out errors intimidating (Yoshida, 2010).
Ironically, the most salient and effective feedback types are
the ones that are the least frequent ones (as predicted by
Carroll, 2001). To put it differently, the feedback types
avoided by some teachers because of their obtrusive and
intimidating nature are the most effective ones. This means
33
that no matter which type of feedback is chosen to be used
exclusively, the decision will entail a compromise, either from
teacher preferences or from salience. The effectiveness of
mixed feedback in the present study, however, shows that it is
possible to provide negative evidence through feedback
treatments that are closer to what learners and teachers are
already used to without compromising salience. In other words,
the study has shown that it is not necessary to correct each
learner error with explicit correction in order to obtain the
desired level of salience.
Finally, results showed that the statistical differences
between the feedback group pairs (e.g., explicit vs. implicit)
observed on the immediate posttest vanished on the delayed
posttest. This, however, should not be taken to mean all
feedback groups performed equally on the delayed posttest. A
glance at the mean scores reveals that the explicit-only and
mixed groups performed better than the implicit-only and
reduced-explicit groups even on the delayed posttest, which was
two months after the treatment. Future research should
investigate the question of whether the differences between the
feedback groups are durable beyond immediate posttests in order
34
to rule out the possibility that observed differences between
the feedback groups in this study are probabilistically likely.
6. Limitations, Future Research, Conclusions
Several limitations should be born in mind when
generalizations are made from this study. Unlike the majority
of previous studies that operationalized explicit feedback
through metalinguistic feedback, this study operationalized
explicit feedback through explicit correction. This means that
the explicit and implicit feedback compared in the current
study differed in directness, not in metalinguistic
information. Obviously, then, the hybrid feedback treatment
used in this study was a mixture of the two feedback types that
differed in directness. Thus, results should be seen as the
outcome of these specific feedback types. Future research can
combine metalinguistic feedback and recasts and investigate the
extent to which this combination would be effective in
comparison to metalinguistic-only and implicit-only groups.
Furthermore, the study has a limited focus with respect to
the linguistic target. Therefore, claims about the relative
effectiveness of the feedback types on target structures that
35
were not the focus of this study should be made with caution.
Again related to the target structures, this study has reduced
the complexity of English articles by limiting the use of each
article to one major function, i.e., first-mention and
anaphoric reference, for the sake of experimental control. This
approach might have artificially boosted the salience of the
target structures. Therefore, it is essential for future
research to represent a broader array of functions of the
articles.
Finally, the present study has used intensive feedback
targeting pre-determined linguistic structures, as has been the
norm in laboratory studies. Therefore, the results of the study
can only be generalized to one-to-one teaching contexts (i.e.,
tutoring). In tutoring lessons aiming to improve learners’
accuracy on particular targets, instructors can implement the
mixed feedback technique by providing explicit feedback
initially and then keeping on with implicit feedback. The
extent to which the findings of this study informs the
classroom settings where feedback is incidental and provided on
a variety of structures (i.e., extensive) may be limited. In
order to determine whether the findings of this study are
36
applicable to these settings, researchers can investigate the
research questions of this study using designs that include
intact classes. In addition, future classroom studies can vary
the number of explicit feedback instances in a mixed feedback
treatment in order to find the mixture that is optimally
effective and practical.
The present study has made a unique contribution to the
research literature by investigating whether a mixed feedback
treatment would be as effective as an explicit-only feedback
treatment. The results of this study have provided support for
the emerging consensus in the field of SLA that explicit
feedback is more effective than implicit feedback in the short
term. The study has also revealed that a mixed feedback
treatment can be as effective as an explicit feedback
treatment. Because this study is the first to show that a mixed
feedback treatment could be as effective as explicit feedback
treatment, further research is necessary to confirm or
disconfirm this finding. Finally, this study targeted one small
aspect of the English language system (two uses of articles),
future research is necessary to investigate whether the
38
7. References
Ammar, A., Spada, N., 2006. One size fits all?: Recasts,
prompts, and L2 learning. SSLA. 28 04, 543–574.
Carroll, S., 2001. Input and evidence: The raw material of
second language acquisition. John Benjamins, Philadelphia
Carroll, S., Swain, M., 1993. Explicit and implicit negative
feedback: an empirical study of the learning of linguistic
generalizations. SSLA. 153, 357–386.
Coltheart, M., 1981. The MRC psycholinguistic database. Quar.
Jour. of Exp. Psyc. 33A, 497-505.
Doughty, C., 2001. Cognitive underpinnings of focus on form,
in: Robinson, P. (Ed.), Cognition and second language
instruction. CUP, Cambridge, pp. 206–257.
Doughty, C., Varela, E., 1998. Communicative focus on form, in:
Doughty, C., Williams, J. (Eds.), Focus on form in
classroom second language acquisition. CUP, New York, pp.
114–138.
Doughty, C., Williams, J., 1998. Focus on form in classroom
second language acquisition. CUP, New York.
Egi, T., 2007. Interpreting recasts as linguistic evidence: The
roles of linguistic target, length, and degree of change.
39
SSLA. 29 4, 511-537.
Ellis, R., 2001. Introduction: Investigating form-focused
instruction. Lang. Learn. 51 s1, 1-46.
Ellis, R., 2007. The differential effects of corrective
feedback on two grammatical structures, in: Mackey, A.
(Ed.), Conversational interaction in second language
acquisition. OUP, New York, pp. 339–360.
Ellis, R., Basturkmen, H., Loewen, S., 2001. Learner uptake in
communicative ESL lessons. Lang. Learn. 51, 281-318.
Ellis, R., Loewen, S., Erlam, R., 2006. Implicit and explicit
corrective feedback and the acquisition of L2 grammar.
SSLA. 28 02, 339–368.
García Mayo, M.P. & Hawkins, R. D. (eds.). 2009. Second
language acquisition of articles. empirical findings and
theoretical implications. Amsterdam: John Benjamins.
Gass S., (1991). Grammar instruction, selective attention, and
learning, in Phillipson, R., Kellerman, Selinker, L.,
Sharwood Smith, M., Swain, M. (Eds.), Foreign/second
language pedagogy research. Multilingual Matters,
Clevedon, pp. 124–141.
Gass, S., Mackey, A., Ross-Feldman, L., 2005. Task-based
40
interactions in classroom and laboratory settings. Lang.
Learn. 55, 575-611.
Goad, H., White, L., 2007. Articles in Turkish/English
interlanguage revisited: implications of vowel harmony,
in: Garcia-Mayo, P., Hawking, R. (Eds.), Second language
acquisition of articles: empirical findings and
theoretical implications. John Benjamins, New York, pp.
201–232.
Li, S., 2010. The effectiveness of corrective feedback in SLA:
A meta-analysis. Lang. Learn. 60, 309-365.
Loewen, S., Nabei, T., 2007. Measuring the effects of oral
corrective feedback on L2 knowledge, in: Mackey, A. (Ed.),
Conversational interaction in second language acquisition.
OUP, New York, pp. 361–377.
Loewen, S., Philp, J., 2006. Recasts in the adult L2 classroom:
characteristics, explicitness and effectiveness. MLJ. 90 4,
536-556.
Long, M. H., 2007. Problems in SLA. Lawrence Erlbaum, Mahwah.
Long, M.H., Robinson, P., 1998. Focus on form: Theory,
research, and practice, in: Doughty, C., Williams, J.
(Eds.), Focus on form in classroom second language
41
acquisition. CUP, New York, pp. 15-41.
Lyster, R., 1998. Recasts, repetition, and ambiguity in L2
classroom discourse. SSLA. 20 01, 51–81.
Lyster, R., Ranta, L., 1997. Corrective feedback and learner
uptake. SSLA. 19 01, 37–66.
Lyster, R., Saito, K., 2010. Oral feedback in classroom SLA: A
meta-analysis. SSLA. 32, 265-302.
Mackey, A., Goo, J., 2007. Interaction research in SLA: A meta-
analysis and research synthesis, in: Mackey, A. (Ed.),
Conversational interaction in second language acquisition.
OUP, New York, pp. 408–452.
Mackey, A., Philp, J., 1998. Conversational interaction and
second language development: Recasts, responses, and red
herrings? MLJ. 82 3, 338–356.
Mackey, A., Gass, S., McDonough, K., 2000. How do learners
perceive interactional feedback? SSLA. 22, 471–97.
Master, P., 2002. Information structure and English article
pedagogy. System. 30, 331-348.
Muranoi, H., 2000. Focus on form through interaction
enhancement: integrating formal instruction into a
communicative task in EFL classrooms. Lang. Learn. 50,
42
617–673.
Oliver, R., Mackey, A., 2003. Interactional context and
feedback in child ESL classrooms. MLJ. 87 4, 519–533.
Öztürk, B., 2005. Case, referentiality and phrase structure.
Amsterdam: John Benjamins.
Panova, I., Lyster, R., 2002. Patterns of corrective feedback
and uptake in an adult ESL classroom. TESOL Quart. 36,
573-95.
Pica, T., 1983. Methods of morpheme quantification: Their
effect on the interpretation of second language data.
SSLA. 6, 69-78.
Schwartz, B. D., 1993. On explicit and negative data effecting
and affecting competence and linguistic behavior. SSLA. 15 2,
147-63.
Seedhouse, P., 2001. The case of the missing "no": The
relationship between pedagogy and interaction. Lang, Lear.
51 s1, 347-385.
Sheen, Y., 2004. Corrective feedback and learners uptake in
communicative classrooms across instructional settings.
Lang. Teach. Res. 8 3, 263-300.
Sheen, Y., 2006. Exploring the relationship between
43
characteristics of recasts and learner uptake. Lang.
Teach. Res. 10, 361–92.
Sheen, Y., 2007. The effects of corrective feedback, language
aptitude, and learner attitudes on the acquisition of
English articles, in: Mackey, A. (Ed.), Conversational
interaction in second language acquisition. OUP, New York,
pp. 301–322.
Truscott, J,. 1996. The case against grammar correction in L2
writing classes. Lang. Learn. 46, 327-369
Yoshida, R., 2010. How do teachers and learners perceive
corrective feedback in the Japanese language classroom?
MLJ. 94, 293–314.
44
Table 1
Feedback Episodes
Table 2
Implicit Feedback Episodes
Episode 1(Story Retelling) Episode 2 (Spot the
difference)
L: The man put the pizza in a
cupboard and leave. Girl is see
everything.
R: A girl?
L: The girl is see everything.
R: What can you see in your
picture?
L: I can see shirt
R: a shirt
L: a shirt and a pencil
R: Where are they?
Explicit Feedback (Explicit
Correction)
Implicit Feedback (Recast)
L: There is a laptop and a
vase.
R: Where are they?
L: Vase is on the coffee table.
R: Vase yanlis, the vase demen
R: What can you see in your
picture?
L: I can see shirt
R: a shirt
45
lazim.
“Vase is incorrect, you
should say the vase.”
R: What else can you see?
L: a shirt and a pencil
R: Where are they?
Table 3
Feedback Provision by Group
Group First Two OCsa Rest of the OCs
Explicit-only Explicit Explicit
Implicit-only Implicit Implicit
Mix Explicit Implicit
No Feedback No Feedback No Feedback
Reduced
Explicit
Explicit No Feedback
Note. a = Obligatory Context
46
Table 4
Interrater Reliability
Obligatory
Contexts
Correct Response Suppliance in Non-
Obligatory
Indefin
ite
Defini
te
Indefini
te
Definit
e
Indefin
ite
Definite
κ % κ % κ % κ % κ % κ %
GOP - 99.9 - 100 .9
3
99.9
6
.9
0
99.
95
- 99.9
6
- 100
SD - 100 - 100 .7
8
99.8
9
.6
2
99.
83
- 99.8
4
- 99.87
RET - 100 - 100 .8
6
99.9
3
.9
3
99.
96
- 99.9
7
- 99.96
i It should be noted that for some researchers (e.g., Goad & White, 2009) the unstressed bir is an indefinite article.ii KS test results for the indefinite posttest scores were Explicit-only, Z(16) = .79, p = .57; Implicit-only, Z(16) =.67, p = .76; Mixed, Z(16) =.86,p = .45; no-feedback, Z(16) = 1.91, p < . 01; reduced explicit, Z(16) =.68,p = .74. KS test results for definite posttest scores were Explicit-only, Z(16) = .56, p = .91; Implicit-only, Z(16) = 1.14, p = .15; Mixed, Z(16) = 1.35, p= . 05; no-feedback, Z(16) = 2.15, p < . 01; reduced explicit, Z (16) = .75, p = .63. KS test results for indefinite delayed posttest scores wereExplicit-only, Z (16) = .62, p = .84; Implicit-only, Z (16) =.95, p = .33; Mixed, Z (16) =.87, p = .43; no-feedback, the distribution had no variance,therefore a KS test could not be performed; reduced explicit, Z (16) =.68, p = .75. KS test results for definite delayed posttest scores were Explicit-only, Z (16) = 1.09, p =. 19; Implicit-only, Z (16) =.79, p= .56; Mixed, Z (16) = .94, p = .34; No-feedback, the distribution had no variance, therefore a KS test could not be performed; Reduced explicit, Z (16) =1.32, p = .60.
47
Table 5
Post Hoc Tests
Indefinite Definite
Comparison Immediate Delayed Immediate Delayed
MDb p d MD p d MD p d MD p d
Expc vs. Impd .27
*a
.01 1.07 .12 .73 .38 .26
*
.00 1.35 .17 .52 .51
Exp vs. NFe .66
*
.00 3.75 .45* .00 1.92 .85
*
.00 14.4
1
.59* .00 2.38
Exp vs. REf .27
*
.01 1.05 .09 .89 .25 .32
*
.00 1.37 .27 .09 .75
Exp vs. Mixg .05 .96 .24 -.0
8
.92 -.26 .03 1.0
0
.15 .05 .99 .15
Mix vs. Imp .32
*
.00 1.42 .20 .25 .72 .24
*
.01 1.09 .11 .82 .39
Mix vs. NF .72
*
.00 5.03 .53* .00 2.91 .82
*
.00 6.77 .54* .00 2.54
48
Indefinite Definite
Mix vs. RE .32
*
.00 1.38 .16 .42 .55 .30
*
.00 1.17 .22 .26 .65
Imp vs. NF .39
*
.00 2.05 .33* .01 1.63 .58
*
.00 3.15 .42* .00 1.91
Imp vs. RE .00 1.0
0
.00 -.0
3
1.00 2.94 .06 .91 .20 .10 .87 .29
RE vs. NF .39
*
.00 1.98 .37* .00 1.64 .52
*
.00 2.29 .32* .03 1.22
Note. a= statistically significant at an alpha level of .05. b
= mean difference, c = explicit-only, d = implicit-only, e = no
feedback, f = reduced explicit, g = mixed
49
Table 6
Post Hoc Tests
Indefinite Definite
Comparison Immediate Delayed Immediate Delayed
MDb p d MD p d MD p d MD p d
Expc vs. Impd .27
*a
.01 1.07 .12 .73 .38 .26
*
.00 1.35 .17 .52 .51
Exp vs. NFe .66
*
.00 3.75 .45* .00 1.92 .85
*
.00 14.4
1
.59* .00 2.38
Exp vs. REf .27
*
.01 1.05 .09 .89 .25 .32
*
.00 1.37 .27 .09 .75
Exp vs. Mixg .05 .96 .24 -.0
8
.92 -.26 .03 1.0
0
.15 .05 .99 .15
Mix vs. Imp .32
*
.00 1.42 .20 .25 .72 .24
*
.01 1.09 .11 .82 .39
Mix vs. NF .72
*
.00 5.03 .53* .00 2.91 .82
*
.00 6.77 .54* .00 2.54
Mix vs. RE .32
*
.00 1.38 .16 .42 .55 .30
*
.00 1.17 .22 .26 .65
Imp vs. NF .39
*
.00 2.05 .33* .01 1.63 .58
*
.00 3.15 .42* .00 1.91
50
Indefinite Definite
Imp vs. RE .00 1.0
0
.00 -.0
3
1.00 2.94 .06 .91 .20 .10 .87 .29
RE vs. NF .39
*
.00 1.98 .37* .00 1.64 .52
*
.00 2.29 .32* .03 1.22
Note. a= statistically significant at an alpha level of .05. b
= mean difference, c = explicit-only, d = implicit-only, e = no
feedback, f = reduced explicit, g = mixed
51
Figure 1.Indefinite Scores by Time and Treatment Group
Figure 2. Definite Scores by Time and Treatment Group
52
Appendix D
Number of Feedback Instances by Treatment Group and Target
Structure
Indefinite Definite
Groups M SD M SD
Implicit 24.87 9.32 23.63 10.11
Explicit 15.67 6.09 11.73 5.95
Mixed 15.12
(4.44
+10.69)a
6.17
(1.90+5.4
5)
14.00
(4.69 +9.31)
7.28
(3.07+5.92)
No Feedback - - - -
Reduced
Explicit
6.87 3.25 6.93 3.13
Note. a = The figures in parenthesis show the breakdown of the
related category according to explicit correction (first
figure) and recasts (second figure).
Top Related