Studies in Second Language Acquisition , 2013, 35 , 167– 184 .
doi:10.1017/S027226311200071X
© Cambridge University Press 2012 167
COUNTERPOINT PIECE:
THE CASE FOR VARIETY
IN CORRECTIVE
FEEDBACK RESEARCH
Correspondence concerning this article should be addressed to Roy Lyster, Department
of Integrated Studies in Education, McGill University, 3700 McTavish Street, Montreal,
QC H3A 1Y2, Canada. E-mail: roy.lyster@mcgill.ca
Roy Lyster
McGill University
Leila Ranta
University of Alberta
Goo and Mackey (this issue) outline several apparent design fl aws in
studies that have compared the impact of different types of corrective
feedback (CF). Furthermore, they argue that SLA researchers should
stop comparing recasts to other types of CF because they are inherently
different kinds of phenomena. Our response to their article addresses
(a) the claim that the recast-learning relationship has been “settled,”
(b) the misleading representation of our views on uptake, (c) the
characterization of the CF comparison studies as being weak and
invalid, and (d) Goo and Mackey’s recommendations concerning the
most appropriate approach to investigating the effect of feedback on
second language learning.
We welcome this opportunity to address some of the arguments and
claims made by Goo and Mackey (this issue). In their article, they criti-
cally examine the methodological features of studies that have com-
pared the effi cacy of recasts to other forms of corrective feedback (CF).
They argue that the current state of research supporting the benefi ts of
recasts is broad and deep, whereas the research in support of other
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S027226311200071X
Downloaded from https:/www.cambridge.org/core. McGill University Libraries, on 21 Apr 2017 at 16:46:43, subject to the Cambridge Core
Roy Lyster and Leila Ranta
168
types of CF is weak and possibly invalid. Their criticisms of an alleged
“case against recasts” center on several apparent design fl aws in studies
not substantiating “the across-the-board utility of recasts” (p. 135): single-
versus-multiple comparisons, the confounding effect of opportunities
for modifi ed output, the provision of form-focused instruction as a feature
of experimental treatments, failure to control for the prior knowledge of
learners, and the impact of out-of-experiment exposure. This leads Goo
and Mackey to make the rather controversial claim that SLA researchers
should stop making comparisons between recasts and other forms
of CF because different types of feedback are inherently different,
and, therefore, apples are being compared to oranges. There are many
points in the article that we dispute, but due to constraints of time and
space, we limit our response to the following: (a) the claim that the
recast-learning relationship has been “settled,” (b) the misleading rep-
resentation of our views on uptake, (c) the characterization of the CF
comparison studies as being weak and invalid due to design fl aws, and
(d) Goo and Mackey’s recommendations as to how research on feed-
back should be conducted in the future.
ARE RECASTS EFFECTIVE ACROSS THE BOARD?
In attributing the source of “doubt” (p. 135) regarding recast effective-
ness solely to Lyster and colleagues, Goo and Mackey disregard con-
cerns expressed previously by several applied linguists. We trace the
fi rst signs of “skepticism” (p. 136) to Corder ( 1967 ), who wrote, “simple
provision of the correct form may not always be the only, or indeed the
most effective form of error correction since it bars the way to the
learner testing alternative hypotheses” (p. 168). Similarly, in his seminal
review of the literature on error correction, Hendrickson ( 1978 ) stated
that the procedure whereby teachers provide students with correct
forms “is ineffective when helping students learn from their mistakes”
(p. 393). On the basis of data from an early classroom observation
study, Chaudron ( 1977 ) concluded that recasting—what he called
“repetition with change” (but without emphasis or reduction)—was
“especially weak in helping to locate the error” (p. 41).
Goo and Mackey refer to their 2007 meta-analysis as “evidence that
recasts work” (p. 134). What they do not mention, however, is that the
effect sizes were signifi cantly larger for lab studies than for classroom
studies. Nor do they mention having qualifi ed their results as follows:
“We cannot say they [recasts] were more or less helpful than other
sorts of feedback due to the insuffi cient number of studies.” Further-
more, they called for “more studies in order to obtain a more clear-cut
picture of the effectiveness of different types of feedback” (Mackey &
Goo, 2007 , p. 442). It would then seem that Goo and Mackey are now
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S027226311200071X
Downloaded from https:/www.cambridge.org/core. McGill University Libraries, on 21 Apr 2017 at 16:46:43, subject to the Cambridge Core
Variety in CF Research
169
retracting their call for comparative studies, and as such we are left to
wonder if this is because those that followed did not necessarily provide
unwavering support for recasts. For example, Lyster and Saito’s ( 2010 )
subsequent meta-analysis, which examined the effects of different types of
CF specifi cally in second language (L2) classroom settings, found medium
effect sizes for recasts and large effect sizes for prompts, which proved
signifi cantly more effective in the within-group contrasts than recasts.
Several reviews of recast studies are available that put recast effec-
tiveness into perspective through reference to a range of linguistic,
pragmatic, cognitive, and contextual constraints (e.g., Ellis & Sheen,
2006 ; Nicholas, Lightbown, & Spada, 2001 ; Sato, 2011 ; Sheen, 2011 ). For
example, the positive effects for recasts appear to be reserved for
learners in form-oriented settings (Nicholas et al., 2001 ) and for those
with high literacy levels (Bigelow, delMas, Hansen, & Tarone, 2006 ),
developmental readiness (Mackey & Philp, 1998 ), high working memory
capacity (Mackey, Philp, Egi, Fujii, & Tatsumi, 2002 ), and high phonological
memory, attention control, and analytic ability (Trofi movich, Ammar, &
Gatbonton, 2007 ). Goo and Mackey acknowledge these and other con-
straints but then ignore them as they claim instead that research has
substantiated “the across-the-board utility of recasts” (p. 135).
For the most part, research demonstrating the effectiveness of
recasts has been conducted in laboratory settings where CF can be
delivered intensively in consistent ways on specifi c linguistic targets.
These laboratory studies have generally yielded positive results for
recasts, but many have methodological problems—such as no control
group (e.g., Ishida, 2004 ) or a control group receiving no CF (e.g., Han,
2002 ; McDonough & Mackey, 2006 ; Sagarra 2007 )—as well as other limi-
tations, as we will point out, akin to those that Goo and Mackey asso-
ciate only with the alleged “case against recasts.”
Whereas the lab studies to which Goo and Mackey appeal for support
fall short of substantiating their claim for across-the-board benefi ts of
recasts, the classroom studies to which they refer (i.e., Doughty & Varela,
1998 ; Goo, 2012 ; Loewen & Nabei, 2007 ) provide even less support, as
we explain next. First, in Doughty and Varela’s classroom study com-
paring CF to no CF, recasts were used solely as secondary moves in the
event that the primary move, which was a prompt that repeated verba-
tim the learner’s error, failed to elicit self-repair. Thus, the fact that Goo
and Mackey refer to this study as evidence of recast effectiveness is
misleading. Also worrying is their lack of criticism of this study for using
multiple feedback types and not controlling for modifi ed output, a criti-
cism they seem to apply only to studies that appear unsupportive of
recast effectiveness. Second, Goo and Mackey classify Loewen and
Nabei ( 2007 ) as a classroom study, but it was not conducted in intact
classrooms. Instead, it was conducted outside of the classroom, with a
researcher who interacted with small groups of four learners, and so it
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S027226311200071X
Downloaded from https:/www.cambridge.org/core. McGill University Libraries, on 21 Apr 2017 at 16:46:43, subject to the Cambridge Core
Roy Lyster and Leila Ranta
170
may be better classifi ed as a lab study. As with other lab studies com-
paring different types of CF (e.g., Lyster & Izquierdo, 2009 ; McDonough,
2007 ), Loewen and Nabei found no difference across CF types.
Third, the treatment in Goo’s ( 2012 ) study was a form-oriented task
requiring learners to ask the teacher a question in order to complete two
fi ll-in-the-blank exercises, each with 15 blanks. Whereas the 14 participants
in the recast group repeatedly heard positive exemplars of the target form
each time an error was made, the 32 participants in the metalinguistic
group heard a rule, such as, “Don’t use the conjunction ‘that’ when
you ask about the subject of the subordinate clause” (p. 455), without
opportunities for self-repair. We have never heard of rule repetition as
a potentially effective type of CF and fi nd it surprising that the recast
group was unable to outperform the metalinguistic group. Moreover,
that both groups outperformed the control group does not provide any
support for recast effectiveness because the control group did not par-
ticipate in the same exercises and so was unable to benefi t from similar
exposure to the target form. A control group in a CF comparison study
needs to participate in the same treatment tasks, but without CF, as was
the case in Loewen and Nabei ( 2007 ), Lyster ( 2004 ), and Yang and Lyster
( 2010 ), so that all groups have similar exposure to the target forms,
with and without CF (Lyster, Saito, & Sato, 2013 ).
INSTANCES OF UPTAKE ARE NOT INSTANCES OF LEARNING
Goo and Mackey portray our earlier work on uptake in a particularly
misleading way and fail to acknowledge that Lyster’s program of research
has continued to evolve since the publication of Lyster and Ranta
( 1997 ). In that seminal work, we stated upfront that, because we exam-
ined only immediate uptake, “claims related to language learning remain
speculative and subject to further empirical investigation” (p. 57), and
that our results needed to “be validated by means of coding sets of
classroom data from other contexts” (p. 56). This call for further class-
room research led to several studies that indeed revealed discrepancies
in rates of uptake following recasts across instructional settings, most
notably Ellis, Basturkmen, and Loewen ( 2001 ), Sheen ( 2004 ), and Lyster
and Mori ( 2006 ), as well as other studies that also revealed low rates of
repair: in English immersion classrooms in Korea (Lee, 2007 ) and
English as a foreign language (EFL) classrooms in Hong Kong secondary
schools (Tsang, 2004 ). Goo and Mackey’s reference to studies of dyadic
interaction (Braidi, 2002 ; Oliver, 1995 ) as counterevidence to our fi nd-
ings is unconvincing, especially when we consider Braidi’s ( 2002 ) con-
clusion that, even in a laboratory setting, it was impossible to determine
whether learner responses to recasts addressed form or meaning, as in,
“‘Yes, I recognize that that is the correct form’ versus ‘Yes, that is what
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S027226311200071X
Downloaded from https:/www.cambridge.org/core. McGill University Libraries, on 21 Apr 2017 at 16:46:43, subject to the Cambridge Core
Variety in CF Research
171
I meant to say,’” (p. 31) and Oliver’s ( 2000 ) subsequent observation
that, in contrast to dyadic interaction, “the nature of whole class inter-
actions diminishes the opportunity for students to respond to the feed-
back” (p. 126). The reference to Ohta’s ( 2000 ) classroom study as
evidence “that L2 learners do in fact produce uptake” (Goo & Mackey,
this issue, p. 137) also seems underwhelming, given that it showed only
that four adult learners, who produced private speech with moderate to
high frequency while wearing lapel microphones in a form-oriented for-
eign language classroom, tended to repeat recasts. These references—
taken as evidence to suggest that the rates of uptake observed by Lyster
and Ranta in French immersion classrooms were somehow wrong—
completely miss the point that we now know that rates of uptake vary
from context to context.
As we have always maintained, instances of uptake are not instances
of learning (Lyster, 1998 , 2002 , 2007 ; Lyster & Ranta, 1997 ). Instead, uptake
refers to a range of possible responses made by students following CF.
Goo and Mackey follow Long ( 2007 ) in misrepresenting Lyster’s pro-
gram of research by attributing to it an “emphasis on uptake” and the
“implication that uptake is a measure of learning” (p. 137). For example,
they refer disapprovingly to Panova and Lyster’s ( 2002 ) suggestion that
“if recasts and translations are essentially corrective in purpose, there
is little evidence that L2 learners in the present study processed them
as such” (p. 591), which was based on Panova and Lyster’s observation
of minimal uptake after recasts and translations. Yet the observational
data in Panova and Lyster, similar to those in Lyster and Ranta, indeed
suggested that learners might not be processing the corrective function
of recasts. That responses to CF may be a sign of noticing was claimed
by others as well, including Mackey herself, who suggested that “uptake
may be related to learners’ perceptions about feedback at the time of
feedback” (Mackey, Gass, & McDonough, 2000 , p. 492). Concerning the
value of uptake following recasts, however, Panova and Lyster noted
that “uptake consisting of a repetition may not have much to contribute
to L2 development, because of its redundancy in an error treatment
sequence where the repair is both initiated and completed by the
teacher within a single move” (p. 579).
Goo and Mackey claim in their article to have “considered the issue of
uptake and its meaning” (p. 146), yet they do not explain uptake as a dis-
course phenomenon, and they fail to make an important distinction
between (a) utterances still in need of repair or (b) utterances with repair.
Utterances in need of repair include simple acknowledgements such as
yes , hesitations, off-target responses, partial repair, and occurrences of
either the same or a different error, whereas utterances with repair entail
the correct reformulation of an error and thus differ from modifi ed out-
put, which may or may not be a correct reformulation. Learner repair
can either be a repetition or self-repair, and these different types of
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S027226311200071X
Downloaded from https:/www.cambridge.org/core. McGill University Libraries, on 21 Apr 2017 at 16:46:43, subject to the Cambridge Core