Académique Documents
Professionnel Documents
Culture Documents
Article
RELC Journal
43(2) 187201
The Author(s) 2012
Reprints and permission: sagepub.
co.uk/journalsPermissions.nav
DOI: 10.1177/0033688212450497
rel.sagepub.com
Mohammad Khatib
Majid Nikouee
Abstract
The present study is premised on Andersons ACT model that proposes declarative knowledge
is automatizable through practice (1982). The research examined the extent to which declarative
knowledge of one morphosyntactic structure, namely present perfect, can be automatized 2
days after practice and can be retained 2 weeks after practice. Twenty intermediate students
participated in the study. They were randomly assigned to experimental (G1) and comparison
(G2) groups. G1 received rule explanation, mechanical practice, meaningful practice, and planned
communicative practice. G2, however, received rule explanation, mechanical practice, and
meaningful practice. Overall, the results indicated more automatization measured in terms of
reduction in error rate and reaction time on the part of G1than G2. The results obtained two
weeks later showed the superiority of G1, too. The findings can be insightful for teachers who
wish to have a communicative class not at the expense of discarding a focus on grammar.
Keywords
Automatization, Procedural Knowledge, Practice, Planned Focus on Form
Introduction
FonF refers to a type of formal instruction in which without any deviation from meaning
attention is drawn to formal features. A body of research confirms the effectiveness of
FonF, either explicit or implicit, in SL/FL development (e.g. de Graaff, 1997; Dekeyser,
1997; Loewen, 2005; Mackey, 2006; Abu Radwan, 2005; Williams, 2001; Ziemer
Corresponding author:
Majid Nikouee, Department of English Language and Literature, Allameh Tabatabai University, Tehran, Iran.
Email: mnikouee@gmail.com
188
Andrews, 2007). Nonetheless, studies carried out in this line do not agree on the most
effective and efficient approach to grammar instruction. This study is, in fact, an investigation of FonF where form complements meaning and is based on Andersons adaptive
control of thought (ACT) model of skill acquisition theory (1982; see Anderson et al. for
a recent version of ACT and ACT-R, 2004). Summarizing Andersons ACT model,
Dekeyser pointed out that knowledge starts as factual information, i.e. knowledge that,
turns into procedural knowledge, i.e. knowledge how, through practising instances of
behaviour, and finally becomes fine-tuned or automatised gradually (1998, 2007).
Ellis has referred to this kind of FonF that is an attempt to operationalize ACT model of
skill acquisition theory as planned FonF where certain forms are pre-selected to be presented and practised to be automatized (2001).
189
In the P-P-P model (Byrne, 1986, cited in Dekeyser, 2007), the first P signifies the
presentation of the new material, the second P the practice of the new material, specifically mechanical and meaningful practice (Paulston and Bruder, 1976, in Dekeyser,
2007), and the third P the production of the presented and subsequently practised material, specifically communicative practice (Paulston and Bruder, 1976). Considering the
P-P-P model of teaching and ACT model of skill acquisition theory, Dekeyser (2007)
pointed out that the second P results in Andersons procedural knowledge and the third
P results in Andersons automatic knowledge.
The present study is, in fact, an attempt to apply skill acquisition theory in a tasksupported context which is based on the principles of communicative teaching (Ellis,
2003). In other words, the potential effects of communicative practice in comparison to
those of mechanical and meaningful practice on the extent to which a morphosyntactic
feature, namely present perfect, can be automatized, automatization being operationalized as reduction in reaction time and in error rate, have been investigated. To this end,
the P-P-P model of L2 teaching has been juxtaposed with a Present, Practise (P-P)
model of L2 teaching to show whether communicative practice (i.e. the third P in the
P-P-P model) makes any difference in the automatization of the targeted form or not.
190
Research Questions
1. Does the P-P-P model of teaching forms augment the Iranian intermediate EFL
learners degree of automatized procedural knowledge of present perfect more
than the P-P model does two days after the treatment?
2. Do the Iranian intermediate EFL learners exposed to the P-P-P model possess a
higher level of automatized procedural knowledge of present perfect more than
those exposed to the P-P model two weeks after the treatment?
Method
Participants
For about a week a notice was put in five language schools in Tehran calling upon students who were at intermediate level to participate in this study. Because the students
were required not to take part in other English classes during the study, only 30 agreed to
participate. To add insult to injury, 10 students gave up gradually and just 20 remained
loyal until the end of the study. Hence, only the data obtained from the remaining students were used. Each participant was paid three dollars worth of a book at the end of
the treatment. All of the participants were speakers of Persian as first language. Their age
range was between 16 and 26 years old, and their average previous experience with
English was 4.35 years. One of the participants was a post-graduate student, one was a
university graduate student, two were university students, two were high school graduate
students, and the rest were senior high school students. They were randomly assigned to
an experimental group (G1) and a comparison group (G2).
Morphosyntactic-Feature
The inclusion of present perfect in this study can only be subjectively justified as it was
selected on the basis of the personal experience of the researchers and their colleagues.
Teaching English for 25 years for one of the researchers and seven years for the other, the
researchers both agreed that one of the most problematic structures Iranian students of
English, especially students of intermediate level, had to tackle was present perfect.
191
Besides, interviewing a number of colleagues regarding this issue, the researchers could
fortify their consensus.
192
present test can be claimed to have construct validity because it was designed in a way
that the participants reaction time and error rate could be recorded.
Untimed grammaticality judgement test.An untimed grammaticality judgement test
(UGJT) containing four grammatical items for the targeted structure was used as a
measure of automatized procedural knowledge. The items were all grammatical to
make the test more a measure of implicit knowledge rather than explicit knowledge as
automatized procedural knowledge is a means of achieving implicit knowledge. This
was done on the basis of Elliss (2009a) finding that grammatical items in a UGJT correlated more with the tests of implicit knowledge than with the tests of explicit
knowledge.
Eight items on four other grammatical features, for each feature one grammatical
and one ungrammatical, were added to the items to militate against the participants
finding a pattern since all of the items on the targeted feature were grammatical. The
non-target features were reported speech, subject-verb agreement, present wish
clauses, and position of frequency adverbs. All of the items were randomly ordered and
were computer-delivered in one-to-one meetings between the individual participants
and the researcher. Although the participants were not pressed for time, they were
alerted that their response time would be recorded. A mark of 1 was assigned to the
items that were judged grammatical and a mark of 0 was assigned to the items that
were judged ungrammatical. This test, too, is expected to have construct validity as it
measures two features of automatized procedural knowledge, namely drop-off in reaction time and in error rate.
Procedure
The present study is a quasi-experimental one based on a pre-test/post-test design. Two
days prior to the start of the treatment, the participants took a pre-test of the automatized
procedural knowledge of the targeted feature and following that the test of metalinguistic knowledge. One day prior to the pre-test phase, the participants were given a list of
all of the verbs included in all of the tests besides 50 other verbs not included in the test
with their past and past participle forms and their Persian equivalents. They were asked
to study them before the first testing phase, and to review them throughout the programme. It was done to make sure that the verbs included in the test were not unknown
to the participants; thus, controlling for the effect of the participants prior familiarity
with the verbs on their accuracy and response time. Being randomly assigned to the
experimental and comparison groups, they started practising. The treatment lasted for 3
sessions, 5 hours, spread over a week. Two days after finishing the treatment, the participants took an immediate post-test, and two weeks after the treatment they took a
delayed post-test.
Testing sessions. Three versions of the test of automatized procedural knowledge were
designed. One was for the pre-test phase, one for the immediate post-test phase, and one for
the delayed post-test phase. The difficulty and the length of the items in each version, however, were not equal. Items in each version, in fact, contained different words with different
193
lengths that were likely to impact on the participants fluency and accuracy. Considering
this issue, it was decided to counterbalance the test versions and administer a different
sequence to each participant (Dekeyser and Sokalski, 1996). For the pre-test, the participants names in each group were listed and the test versions (version 1, 2, or 3) were
assigned to the names in order. For the immediate post-test, one of the versions not taken
by the participants on the previous testing session was randomly assigned to the participants. Finally, the remaining test version was assigned to the participants for the delayed
post-test.
The data on the measure of automatized procedural knowledge were collected via
one-to-one meetings between one of the researchers and individual participants. Each
meeting took between 15 and 20 minutes.
After taking the pre-test, each participant was given the test of metalinguistic knowledge to see if the participants started from scratch or with some prior explicit knowledge
of the targeted feature. The reason why the test of automatic procedural knowledge was
administered before the test of declarative knowledge was to avoid the learning effects
of the declarative measure on the automatic procedural measure (Webb, 2007).
Treatment sessions. At the outset of session 1, lasting for 120 minutes, both groups were
given the same handout on present perfect. Both groups self-studied their handouts and
discussed the main points in class. This was the presentation phase. For the rest of the
session, both groups did some mechanical and meaningful practice. At the end of the
session G1 did one dyadic controlled communicative task whereas G2 was deprived of
this task.
During sessions 2 and 3, G1 and G2 practised present perfect in different ways.
Whereas G1 did free communicative tasks, G2 continued doing mechanical and
meaningful activities. Session 2 lasted for 120 minutes and session 3 lasted for 60
minutes.
Data Collection and Transcription. Data collection on the pre-test, immediate post-test and
delayed post-test phases was carried out via Windows Vistas Sound Recorder. On the test
of automatized procedural knowledge the participants saw the items one by one on a
computer screen and as they responded their voice was recorded on another computer
controlled by an assistant. The participants were supposed to say pass as they answered
each item to mark that they had finished answering the item. On the UGJT, however, the
participants did not say pass. They just said yes or no for grammatical and ungrammatical sentences, respectively. The assistant stopped the sound recorder as soon as he
heard the word pass (yes or no in the case of the UGJT).
Data transcription was conducted by a piece of software named AVS Audio Editor.
Because reaction time was an essential component of automatized procedural knowledge, this software was used to be able to record tenths of a second. Another issue
regarding the participants reaction time that needed to be taken into account was an
inevitable gap between the participants saying pass (yes or no in the case of the
UGJT) and the assistants stopping the recording. It was decided to ignore the gap
provided that it was less than one second. Whenever the gap was greater than a second,
however, the recorded reaction time was exactly when the word pass (yes or no in
194
the case of the UGJT) was said. Thanks to AVS Audio Editor, it was possible to find
when the word pass (yes or no in the case of the UGJT) was completely expressed.
Fortunately, on each testing session only a few more-than-one-second gaps were
observed.
Statistical Procedure. Test of distributional normality revealed that the data distribution
was not normal obviously owing to the small samples. Hence, nonparametric tests
were applied. For all between-groups comparisons Mann-Whitney U test and for all
within-groups comparisons Wilcoxon Matched-Pairs Signed-Ranks test were used.
Comparisons showing a significant difference was followed by an analysis of the
strength of association (eta) to find out how much of the difference was attributable to
the treatment. All of the statistical tests were conducted using SPSS 18.
Statistics
Mann-Whitney U
Wilcoxon W
Z
Asymp. Sig. (2-tailed)
Exact Sig. [2*(1-tailed Sig.)]
43.000
98.000
.601
.548
.631
After taking the test of metalinguistic knowledge, G1 and G2 took the test of automatized procedural knowledge and the UGJT. To compare the groups initial automatized
procedural knowledge to find out if they were at (almost) the same level of knowledge
before receiving the treatment, the mean performance of the groups on the tests was
compared via the Mann-Whitney U test. Note that the mean performance in this study
195
Error Rate
Test of Automatized
Procedural Knowledge
UGJT
Mean rank
Sum of
ranks
MannWhitney U
Exact sig
(2-tailed)
G1:9.10
G2:11.90
G1:9.90
G2:11.10
G1:91.00
G2:119.00
G1:99.00
G2;111.00
36.000
1.058
44.000
.454
.315
.684
G1:9.50
G2:11.50
G1:10.30
G2:10.70
G1:95.00
G2:115.00
G1:103.00
G2;107.00
40.000
.787
48.000
.163
.481
.912
had two aspects one was the mean reaction time and the other was the mean error rate.
To save space, the results of reaction time and error rate comparisons in the pre-test
phase are summarized in Table 2.
Table 2 does not show any significant difference between the groups reaction times
at the probability level of .05. On both tests it was G1 that outperformed G2. In terms of
error rate, no significant difference between the groups could be found at the probability
level of .05.
The next step was to compare the groups performance in the immediate post-test
phase. Table 3 presents the groups reaction times and error rates in the immediate
post-test phase which were compared to evaluate the effectiveness of the treatment
phase.
Table 3. Between-groups Comparisons of Experimental Groups (G1) and Comparison
Groups (G2) Reaction Time and Error Rate on the Test of Automatized Procedural Knowledge
and UGJT in the Immediate Post-test Phase
Test
Reaction Time
Test of Automatized
Procedural Knowledge
UGJT
Error Rate
Test of Automatized
Procedural Knowledge
UGJT
Mean rank
Sum of
ranks
MannWhitney U
Exact
sig(2tailed)
Eta
G1:7.80
G2:13.20
G1:8.70
G2:12.30
G1:78.00
G2:132.00
G1:87.00
G2:123.00
23.000
2.041
.043
32.000
1.361
.190
.21
-----
G1:10.00
G2:11.00
G1:8.30
G2:12.70
G1:100.00
G2:110.00
G1:83.00
G2;127.00
45.000
.389
.739
28.000
2.063
.105
196
Comparison of G1 and G2s reaction times, as shown in Table 3, did not show any
significant differences two days after the treatments except for the comparison of the
groups on the test of automatized procedural knowledge at p < .043, with G1 significantly outperforming G2 (Z = 2.041, h = .21). With regard to error rates in the
immediate post-test phase, comparison did not evince any significant difference at
p < .05.
To further evaluate the effectiveness of the treatment, each groups performance in the
pre-test and the immediate post-test phases were compared via Wilcoxon Matched-Pairs
Signed-Ranks test. The results of these within-groups comparisons are all summarized in
Tables 4 (G1) and 5 (G2).
As Table 4 indicates, there was no significant within-groups difference for G1 in
terms of reaction time. But in terms of error rate, according to Table 4, a significant
within-groups difference was observed on the test of automatized procedural knowledge (Z = 2.701d, p < .007) and the UGJT (Z = 2.449d, p < .014) of present
perfect.
Ties: 0c
Error Rate
Test of Automatized
Procedural Knowledge
Ties: 1c
UGJT
Ties: 4c
Number
of ranks
Mean rank
Sum of
ranks
-----
------
.81
.66
Note. NNR = Number of negative ranks; NPR: Number of positive ranks; a = Post-test < Pretest; b = Post-test
> Pretest; c = Post-test = Pretest; NMR = Negative mean ranks; PMR = Positive mean ranks;
NSR = Negative sum of ranks; PSR = Positive sum of ranks; d = Based on positive ranks.
Table 5, presenting the results of comparing the performance of G2 from the pre-test
phase to the post-test phase, does not show any significant difference on any of the measures in terms of reaction time and error rate.
197
Ties: 0c
Error Rate
Test of Automatized
Procedural
Knowledge
Ties: 1c
UGJT
Ties: 7c
Number
of ranks
Mean rank
Sum of
ranks
NNR: 4a
NPR: 6b
NMR: 6.00
PMR: 5.17
NSR: 24.00
PSR: 31.00
.357e
NNR: 6a
NPR: 4b
NMR: 6.67
PMR: 3.75
NSR: 40.00
PSR: 15.00
1.274d
NNR: 8a
NPR: 1b
NMR: 4.81
PMR: 6.50
NSR: 38.50
PSR: 6.50
1.919d
NNR: 2a
NPR: 1b
NMR: 2.50
PMR: 1.00
NSR: 5.00
PSR: 1.00
1.089d
Asymp. sig
(2-tailed)
.721
.203
.055
.276
Note. NNR = Number of negative ranks; NPR: Number of positive ranks; a = Post-test < Pretest; b = Post-test
> Pretest; c = Post-test = Pretest; NMR = Negative mean ranks; PMR = Positive mean ranks; NSR = Negative sum of ranks; PSR = Positive sum of ranks; d = Based on positive ranks; e = Based on negative ranks.
Research Question 1
The first research question was a search for the effect of practice on the participants ability to automatize their declarative knowledge of present perfect two days after the treatment by juxtaposing the Present, Practise, Produce (P-P-P) model and the Present,
Practise (P-P) model. To this end, it was hypothesized that the models did not differ in
terms of effects on the participants ability to automatize their declarative knowledge of
the targeted feature.
Post-treatment between-groups comparison of G1 and G2s reaction times (Table 3)
showed a significant difference on the test of automatized procedural knowledge of present perfect in the immediate post-test phase with G1 having lower reaction time than G2
(Z = 2.041, p < .043). The observed eta value (h = .21) indicated that 21% of the
observed difference in reaction time on the test of automatized procedural knowledge of
present perfect could be attributed to the method applied to G1. Note that G1 had lower
reaction time on this test in the pre-test phase; however, that difference was not significant at p < .05 (Table 3). On the same test in the immediate post-test phase G1 outperformed G2 in terms of error rate in the immediate post-test phase (Z = .389) as it did in
the pre-test phase (Z = .787). The Z value dropped from the pre-test to the immediate
post-test phase suggesting some progress (i.e. automatization) on the side of G2. In both
phases the differences were nonsignificant (Tables 2 and 3).
198
On the UGJT of present perfect G1 and G2s reaction time difference, like that of the
pre-test phase, was not significant at p < .05. As shown in Tables 2 and 3, although in
both phases G1 outperformed G2 on this test, the Z value in the immediate post-test
phase (Z = 1.361) was larger than that in the pre-test phase (Z = 1.058). On the same
test G1 outperformed G2 in the immediate post-test phase as it did in the pre-test phase
in terms of error rate. On this test the Z value rises from the pre-test phase (Z = .163) to
the immediate post-test phase (Z = 2.063) suggesting some progress by G1. Nonetheless,
in neither phase was the difference significant (Tables 2 and 3).
As shown in Table 4, within-groups comparisons suggested that on the test automatized procedural knowledge G1 had a drop-off in reaction time from the pre-test phase to
the immediate post-test phase though not significant (Z = 1.376, p < .169). G2, on the
contrary, showed a rise in reaction time on the same test (Z = .358, p < .721) though the
rise was not significant at .05 (Table 5). In terms of error rate, on the test of automatized
procedural knowledge both groups were able to considerably reduce with G1 producing a
Z score of 2.701 that was significant at p < .007 with the eta value of .81, and with G2
producing a Z score of 1.919 that was nearly significant at p < .055 (Tables 4 and 5).
On the UGJT, according to the within-groups comparisons of the groups reaction
time, G1 and G2 both showed a nonsignificant drop-off in their reaction time (Tables 4
and 5). Although neither showed a significant reduction, G2 decreased its reaction time
(Z = 1.274, p < .203) slightly more than G1 (Z = 1.172, p <.241). In terms of error rate,
both groups performed better in the immediate post-test phase than in the pre-test phase
on the UGJT with G1s significantly reducing its error rate (Z = 2.449, p < .014) with
the eta value of .66 (Table 4). However, G2s reduction of error rate (Z = 1.089,
p < .276) was not significant at p < .05 (Table 5).
In short, in terms of reduction in reaction time on the test of automatized procedural
knowledge G1 significantly outperformed G2 in the immediate post-test phase. Note that
G1 outperformed G2 in the pre-test phase, too, though not significantly. On the UGJT,
G1 outperformed G2 (i.e. more reduction of reaction time) in both the pre-test phase and
the immediate post-test phase. Although G1s outperforming G2 was not significant in
any of the phases, it was larger in the immediate post-test phase. Within-groups comparisons of reaction time showed that while G1 managed to reduce on the test of automatized
procedural knowledge, though not significantly, G2 increased. According to these findings, G1 was partially more successful than G2 in reducing reaction time.
In terms of error rate on the test of automatized procedural knowledge, G1 could considerably, though not significantly, increase the difference from G2 from the pre-test
phase to the immediate post-test phase. On the UGJT, too, G1 could considerably increase
the difference from G2 in terms of error rate from the pre-test phase to the immediate posttest phase. Within-groups comparisons of error rate showed that both G1 and G2 could
reduce error rate on both the test of automatized procedural knowledge and the UGJT.
However, only G1 was able to reduce error rate significantly. These findings, accordingly,
partially suggested that G1 was more successful in reducing error rate than G2.
The results obtained from the test of automatized procedural knowledge and the
UGJT partially evidenced automatization on the part of G1. There was one significant
between-groups difference and two significant within-groups differences that were all
attributable to G1. In nonsignificant cases, it was G1 that surpassed G2 not vice versa.
199
These findings are, overall, reflective of G1s ability to automatize the knowledge of
present perfect more than G2 did. The eta values obtained from the within-groups comparisons of G1 in terms of error rate on the test of automatized procedural knowledge and
on the UGJT (h = .81 and h = .66, respectively), and the eta value obtained from
between-groups comparison of the groups reaction times on the test of automatized
procedural knowledge (h = .21) to some extent signify the robustness of the method
applied to G1 compared to the one applied to G2. However, care must be exercised in the
construal of these results since they do not strongly fit the zone of statistical
significance.
Research Question 2
The second research question asked which group, G1 or G2, possessed a higher level of
automatized procedural knowledge of present perfect two weeks after the treatment. It
was hypothesized that there was no difference between the groups level of automatized
procedural knowledge two weeks after the treatment.
To answer this question, both groups took a delayed post-test two weeks after finishing the treatment. The results of comparing G1s reaction time and error rate with those
of G2 on both measures of the delayed post-test phase, computed via the Mann-Whitney
U test, are presented in Table 6. The only statistically significant difference between G1
and G2s reaction time was on the UGJT (p < .015), with G1 having lower reaction time
(Z = 2.419). This difference existed in the immediate post-test phase, too, although it
was not statistically significant. In terms of error rate, however, none of the differences
between G1 and G2 was statistically significant. In the immediate post-test phase, too,
none of the error rate differences was statistically significant.
Table 6. Between-groups Comparisons of Experimental Groups (G1) and Comparison
Groups (G2) Reaction Time and Error Rate on the Test of Automatized Procedural Knowledge
and UGJT in the Immediate Post-test and Delayed Post-test Phases
Test
Reaction Time
Test of Automatized
Procedural Knowledge
UGJT
Error Rate
Test of Automatized
Procedural Knowledge
UGJT
Mean rank
Sum of
ranks
MannWhitney U
G1:10.00
G2:11.00
G1:7.30
G2:13.70
G1:100.00
G2:110.00
G1:73.00
G2;137.00
45.000
.378
18.000
2.419
G1:9.00
G2:12.00
G1:8.90
G2:12.10
G1:90.00
G2:120.00
G1:89.00
G2:121.00
35.000
1.186
34.000
1.446
Exact
sig(2tailed)
.796
.015
.280
.247
In sum, G1 showed lower reaction time and error rate on both of the tests. These
results are indicative of G1s higher level of automatized procedural knowledge two
weeks after the treatment. However, these results, too, must be considered cautiously.
200
201
References
Abu Radwan A (2005) The effectiveness of explicit attention to form in language learning. System
33: 69-87.
Anderson J (1982) Acquisition of cognitive skill. Psychological Review 89: 369-406.
Anderson J, Bothell D, Byrne MD, Douglass S, Lebiere C, and Qin Y (2004) An integrated theory
of the mind. Psychological Review 111: 1036-1060.
Anderson JR, Fincham JM (2010) Acquisition of procedural skills from examples. Available at:
ja@cmu.edu
de Graff R (1997) The Experanto experiment: effects of explicit instruction on second language
acquisition. Studies in Second Language Acquisition 19: 249-76.
Dekeyser RM (1996) Exploring automatisation processes. TESOL Quarterly 30: 349-57.
Dekeyser RM, Sokalski K (1996) The differential role of comprehension and production practice.
Language Learning 46: 613-42.
Dekeyser RM (1997) Beyond explicit rule learning: automatising second language morphosyntax.
Studies in Second Language Acquisition 19: 195-221.
Dekeyser RM (1998) Beyond focus on form: cognitive perspectives on learning and practicing
second language grammar. In: Doughty C, Williams J (eds) Focus on Form in Classroom
Second Language Acquisition. New York: Cambridge University Press, 42-63.
Dekeyser RM (2007) Introduction: Situating the concept of practice. In: Dekeyser RM (ed.) Practice
in Second Language: Perspectives from Applied Linguistics and Cognitive Psychology.
New York: Cambridge University Press, 1-18.
Dekeyser RM (2007) Conclusion: The future of practice. In: Dekeyser RM (ed.) Practice in
Second Language: Perspectives from Applied Linguistics and Cognitive Psychology. New
York: Cambridge University Press, 287-304.
Douglas D (2001) Performance consistency in second language acquisition and language testing:
a conceptual gap. Second Language Research 17: 442-56.
Ellis R (2001) Introduction: investigating form-focused instruction. Language Learning 51 (suppl.
1): 1-46.
Ellis R (2003) Task-based Language Learning and Teaching. Oxford: Oxford University Press.
Ellis R (2005) Principles of instructed language learning. System 33: 209-24.
Ellis R (2009a) Measuring implicit and explicit knowledge of a second language. In: Ellis
R, Loewen S, Elder C, Erlam R, Philip J, and Reinders H (eds) Implicit and Explicit
Knowledge in Second Language Learning, Testing and Teaching. Bristol: Multilingual
Matters, 31-64.
Ellis R (2009b) Task-based language teaching: sorting out the misunderstandings. International
Journal of Applied linguistics 19: 221-46.
Loewen S (2005) Incidental focus on form and second language learning. Studies in Second
Language Acquisition 27: 361-86.
Mackey A (2006) Feedback, noticing, and instructed second language learning. Applied Linguistics
27: 405-30.
Robinson P (1997) Generalizability and automaticity of second language learning under implicit,
incidental, enhanced, and instructed conditions. Studies in Second Language Acquisition 19:
223-47.
Webb S (2007) Learning word pairs and glossed sentences: the effects of a single context on
vocabulary knowledge. Language Teaching Research 11: 63-81.
Williams J (2001) The effectiveness of spontaneous attention to form. System 29: 325-40.
Ziemer Andrews KL (2007) The effects of implicit and explicit instruction on simple and complex
grammatical structures for adult English language learners. TESL-EJ 11: 1-15.