System 89 (2020) 102196

Contents lists available at ScienceDirect

System
journal homepage: www.elsevier.com/locate/system

Using model texts as a form of feedback in L2 writing

Eun Young Kang
Carleton University, School of Linguistics and Language Studies, 1125 Colonel by Drive, Ottawa, ON, K1S 5B6, Canada

a r t i c l e i n f o a b s t r a c t

Article history: Recently, researchers have proposed that model texts can be used as an effective feedback
Received 25 April 2019 technique. To contribute to the ongoing research, the study examined the extent to which
Received in revised form 16 December 2019 models can result in the improvement of L2 writers’ subsequent drafts using a three-stage
Accepted 16 December 2019
writing task: (i) writing (pretest), (ii) comparing (treatment), and (iii) rewriting (posttest).
Available online 24 December 2019
Forty Grade 11 EFL students wrote an argumentative essay and then were assigned to
either an experimental (n ¼ 20) or a control group (n ¼ 20). The experimental group
Keywords:
received two model texts to compare their writing against, whereas the control group
Models
L2 writing
(n ¼ 20) was asked to self-correct their initial composition. The results suggested that
Written corrective feedback learners mainly noticed their lexical problems at the initial stage of writing, which drew
their attention to the words included in the models. They subsequently incorporated some
of the words into their rewriting. By incorporating linguistic elements from the model
texts, the participants in the experimental group in general received higher writing scores
on the posttest compared to those in the self-correction group. Model texts were found to
be particularly effective in improving vocabulary and content of L2 learners’ writing.
© 2020 Elsevier Ltd. All rights reserved.

1. Introduction

Written corrective feedback (WCF) has received much attention from researchers over the past few decades. Much of WCF
research focuses on the effectiveness of different feedback methods in improving students’ written accuracy. A variety of
feedback strategies have been researched, such as direct corrective feedback (i.e., provision of the correct form), indirect
corrective feedback (i.e., underlining an error or using codes to show that an error has been made), and metalinguistic
explanation (i.e., providing grammar rules) (Guirao, de Larios, & Coyle, 2015; Kang & Han, 2015). One commonality of these
feedback methods is that they mainly deal with lexical and grammatical errors at the sentence level. More recently, several
researchers have investigated more discursive forms of feedback such as providing model texts (e.g., Coyle & de Larios, 2014;
Guirao et al., 2015; Hanaoka & Izumi, 2012). Model texts are defined as samples of well-written texts tailored to “the content
and the genre of the writing task at hand” (Coyle & de Larios, 2014, p. 453). While standard teacher feedback methods only
present learners with corrections of erroneous forms, model texts can provide them with a variety of suggestions with regard
to relevant content, vocabulary, and organizational structures. It is assumed that learners can notice such information when
they compare model texts against their own writing, consequently incorporating some features from the models into their
subsequent drafts.
This assumption has been tested and partially confirmed by the findings of some recent studies (e.g., Guirao et al., 2015;
Coyle, Guirao, & de Larios, 2018; Coyle & de Larios, 2014; García Mayo & Loidi Labandibar, 2017; Hanaoka, 2007; Hanaoka &

E-mail address: [email protected].

https://doi.org/10.1016/j.system.2019.102196
0346-251X/© 2020 Elsevier Ltd. All rights reserved.
2 E.Y. Kang / System 89 (2020) 102196

Izumi, 2012). In all of these studies, learners completed a written picture description task and then were asked to take notes
on any problems they had during the writing phase (e.g., their inability to use a particular word or a grammatical structure).
Immediately upon the completion of their own description writing and note-taking, model texts were provided. While
comparing their writing against model texts, L2 writers were able to notice solutions to problems they had posed themselves
during writing. Evidence for such noticing was provided through learners’ notes taken during the comparison stage. When
asked to rewrite the same picture description, the learners incorporated linguistic features that they had noticed in the model
writing and thus produced linguistically more accurate drafts, especially in terms of vocabulary.
These studies have provided evidence of the potential benefit of using model texts as a feedback tool. However, re-
searchers have yet to use model texts comprised of more than a single paragraph. Moreover, all of the existing studies have
looked into the role of model texts using a picture-based writing task, so it remains unclear if the positive results found would
be replicated in other studies adopting different genres of writing. In addition, previous studies have yet to systematically
document evidence of improvements in subsequent drafts in terms of specific aspects of writing (e.g., lexis, grammar, content,
and organization). To address these issues and further evaluate if models can efficiently serve as a tool for WCF, the current
study investigated the usefulness of a longer model text consisting of multiple paragraphs within the context of an argu-
mentative writing task, which has long been identified as a crucial genre in academic studies at various grade levels (Qin &
Karabacak, 2010). Furthermore, using an analytic scoring rubric, this study examined whether or how model texts contributed
to improvements in various aspects of writing.

2. Literature review

2.1. Theoretical justification for using models as a feedback tool

The benefits of using model texts to improve L2 learners’ writing can be understood in light of (i) Schmidt’s (1990, 2001)
Noticing Hypothesis and (ii) Swain’s Output Hypothesis (1995). In Noticing Hypothesis, Schmidt (1990, 2001) stresses the role
that noticing plays in the process of second language learning and claims that more noticing could lead to more learning.
Other second language acquisition (SLA) researchers have also agreed that noticing plays a crucial role in language learning
(Hanaoka, 2007). Relatedly, Swain (1995) argues that using the target language during writing and speaking promotes
learners’ noticing of their linguistic problems and draws their attention to subsequent input, which helps them solve the
problems. Specifically, Swain describes four ways through which output promotes noticing and second language acquisition.
The act of producing language provides learners with opportunities to (i) test their hypotheses about accuracy and
comprehensibility; (ii) reflect upon their target language use through metalinguistic analysis; (iii) enhance their fluency; and
(iv) notice their inability to express their intended meaning and subsequently pay closer attention to relevant feedback or
input to resolve their linguistic problems. On the basis of the noticing function of output, it is likely that learners notice their
own linguistic limitations as they engage in writing and that such noticing prepares them to use model texts as a resource to
resolve their problems (Hanaoka, 2007; Hanaoka & Izumi, 2012; Izumi, 2003). While reading the model texts, L2 learners may
notice a gap between their language use and the target language models they receive (Hanaoka, 2007; Hanaoka & Izumi,
2012). It is plausible that the provision of model texts right after learners complete a writing task could efficiently allocate
learners’ attention to alternative vocabulary, grammar, and content provided in the models. In this regard, models can serve
as a corrective feedback tool.

2.2. Empirical studies on the use of models as a feedback tool

In light of Swain’s (1995) theoretical arguments, many studies empirically tested the role of output in L2 learners’ noticing
of their linguistic shortcomings (see Gass & Mackey, 2007 for a review). These studies supported Swain’s claim that output
triggers the noticing of language problems that L2 learners are experiencing and directs their attention to the means by which
they can express their intended meaning. While much of the research on the noticing function of output focuses on oral
interaction, Hanaoka (2007) investigated the issue in the written context, attempting to uncover the relationship between
output, noticing, and exposure to model texts (Kang, 2010). In the study, Hanaoka’s (2007) had 37 adult Japanese-speaking
English-as-a-foreign language (EFL) learners engage in a multi-stage writing task which was designed to have them (a) notice
their linguistic problems while writing a narrative based on a picture prompt (stage 1), (b) notice the difference between their
compositions and two native-speaker models while comparing their drafts with the models (stage 2), and (c) revise their
draft (stage 3). The participants were also asked to take notes in Japanese on whatever they noticed during stages 1 and 2. The
learners’ notes were coded and analyzed according to four categories: grammar, vocabulary, content, and other. The results
showed that the learners mainly noticed lexical problems when composing their drafts. They were able to find solutions to
these problems from reading the models, and they incorporated appropriate expressions in their rewriting (stage 3).
Since Hanaoka (2007) arguably made the first attempt to explore the possibility of using model texts as a corrective
feedback tool, a series of subsequent studies (e.g., Coyle & de Larios, 2014; Guirao et al., 2015; Hanaoka & Izumi, 2012; Yang &
Zhang, 2010) have used his research methods, which employ a three-stage writing task: (a) writing a picture description, (b)
comparing it with a model, and (c) rewriting the original description without access to the model text. Another common
methodological feature of the studies is the use of note-taking as a noticing measure. In Hanaoka’s (2007) study, participants
were asked to take notes in their native language (i.e., Japanese) on whatever they noticed (i) while they were first attempting
 E.Y. Kang / System 89 (2020) 102196 3

to write a picture description and (ii) while they were comparing it against model texts. Based on these notes, the researcher
was able to trace what learners noticed during the writing and comparison stages.
Using these research methods, recent studies have also assessed the outcomes of models on L2 writing in relation to other
types of WCF. For instance, Hanaoka and Izumi (2012) compared the efficacy of providing reformulated texts (e.g., refor-
mulated versions of an individual learner’s writing without errors) and model texts in adult EFL classes, and Coyle and de
Larios (2014) compared the role played by model texts and error correction in L2 children’s noticing and rewriting. While
these studies involved different groups of L2 learners, they showed similar results related to model texts. First, participants’
written notes showed that they focused mainly on lexis while reading model texts. Second, participants mostly integrated
lexical items they had noticed in the models into their rewriting. However, due to the absence of a control condition, it is not
clear whether the improvements in vocabulary use reported in the participants’ rewriting resulted solely from the model
texts provided or from other variables.
To address the methodological issue, Guirao et al. (2015) re-examined the effectiveness of models in a subsequent study
that included a control group. In their study, children in Grade 5 engaged in a three-stage picture description task in pairs.
Again, similar to the results of the previous studies, the learners primarily focused on lexis. Moreover, findings showed that
learners with higher proficiency were able to notice and incorporate more linguistic features. The study attempted to provide
clearer evidence of the effects of model texts on L2 learners’ noticing and writing performance by including a control group,
but an issue of concern was that the participants jointly engaged in the writing-comparison-rewriting task, which made it
difficult to single out the unique contributions of model texts to gains reflected in rewriting. The pair discussion could have
augmented L2 writers’ noticing during the initial writing and the comparison stages.
The studies mentioned so far employed similar methodologies and exhibited comparable findings, while raising several
common methodological issues that should be addressed in future research. One general finding shown from the studies is
that learners tend to focus on vocabulary problems, both at the moment of their initial writing and at the comparison/
feedback stage. As a result, a significantly large number of words are incorporated from a model text into their rewriting.
However, this finding can be partially accounted for by the picture description writing tasks employed in the studies. Picture
description tasks inherently elicit semantic-focused processing because they push learners to use words embedded in the
pictures. Therefore, they might have primed learners to focus more on words, which might have inflated learners’ noticing of
lexical problems, subsequently leading to the incorporation of more words in their rewriting. Considering the fact that
picture-based writing tasks could induce learners’ semantically oriented noticing processes, there is a need to implement
different genres of writing tasks. In this regard, the current study adopted an argumentative task and examined what learners
notice during the composing and comparing stages. An argumentative writing task was deemed appropriate because it in-
volves coherent organization of opposing ideas at the discourse level. In the previous studies, learners’ noticing of discourse-
level elements could not be examined because they were asked to write in response to only a simple prompt consisting of a
small number of pictures. Consequently, the length of the model text offered was very brief and comprised only a single
paragraph. By adopting an argumentative writing task, the present study could investigate whether L2 writers are in fact able
to notice macro-level features, including the organization of opposing views and coherence, from model texts.
In addition, in the existing studies, to determine the effect of model texts on L2 learners’ subsequent revisions, researchers
only counted how many linguistic features from model texts were incorporated into L2 learners’ writing. However, to pre-
cisely measure improvement in rewriting as a result of models, the current study measured changes made in specific areas of
writing using an analytic rubric. By adopting the rubric, this study attempted to uncover the extent to which a model text
improves L2 learners’ rewriting. Taking these methodological issues into account, the present study investigated the role of a
model text as a feedback tool by addressing the following three research questions: 1) what do L2 learners notice when
composing an argumentative essay? 2) what do L2 learners notice when comparing their essay with model texts? 3) To what
extent does a model text improve L2 learners’ subsequent rewriting?

3. Methodology

3.1. Participants and setting

The study was carried out in a high school in South Korea. The participants were forty Grade 11 female students who had
been studying English as a foreign language for eight years. They were from two intact classes and received 4 hours of in-
struction per week. The main purpose of the English class was to help students to improve written English skills, and the focus
was on reading comprehension, writing, vocabulary, and grammar. To measure their English proficiency, all participants were
asked to complete a practice TOEIC® (Test of English for International Communication) test. Out of a total possible score of
990, the range of scores was from 500 to 850, which is estimated to belong to intermediate (B1) and upper intermediate levels
(Lower B2) of English proficiency according to the Common European Framework of Reference (CEFR; Council of Europe,
2001). The two classes were randomly divided into (a) an experimental group (n ¼ 20), which wrote an argumentative
essay, compared it against a model text, and rewrote it, and (b) a control group (n ¼ 20), which engaged in writing and
rewriting an argumentative essay but did not receive a model text. To make sure that the two groups were comparable with
regard to their English proficiency, participants’ TOEIC scores were analyzed. The groups were found not to be different in
terms of their English proficiency (t ¼ 0.79, p > 0.05).
4 E.Y. Kang / System 89 (2020) 102196

3.2. Tasks and instruments

Similar to the previous studies reviewed (e.g., Coyle & de Larios, 2014; Guirao et al., 2015; Hanaoka, 2007; Yang & Zhang,
2010), the participants completed a three-stage writing task: writing-comparing-rewriting. Participants’ initial writing and
re-writing served as a pretest and a posttest, respectively, and during the comparison stage, only the experimental group
received model texts against which they compared their initial writing. With regard to the writing task, participants were
instructed to write whether they agreed or disagreed with banning smoking in public places. The topic was chosen because a
school-based smoking prevention campaign had been launched right before the study was carried out. Therefore, the par-
ticipants were familiar with the topic at the time of data collection. The writing task was pilot-tested with another group of
learners in the same grade studying at the same school to ensure level appropriateness. Pilot-testing showed that this topic
and writing task were suitable for this study.
Results of the pilot test showed that all students were against the idea of smoking in public places, possibly due to the
influence of the ongoing smoking prevention campaign conducted at school. Based on the results, two model texts arguing
against public smoking were created by a proficient speaker of English. Two models were used to prevent learners from
heedlessly reproducing linguistic items included in a single model and to improve the chance of offering solutions to the
problems they might have noticed during the writing stage (Hanaoka, 2007).
Before participants wrote about their ideas about smoking in public places, the researcher had a 5-min brainstorming session
with the participants in Korean. During the session, the researcher encouraged them to think about the pros and cons of
smoking in public places. However, in line with the results of the pilot-study, when they were asked to write about the topic, all
participants took a stance against smoking in public places, which aligned with the two pre-written models provided.
Right after participants completed the rewriting stage, all participants also completed a questionnaire in Korean. The
questionnaire consisted of two parts, and the control group completed only the first part. The first part included questions
that investigated the participants’ backgrounds with regard to their English language learning. The second part asked par-
ticipants (a) if they felt that a model text was useful in improving their writing and (b) why they felt the way they did.
Participants’ perceived level of usefulness was examined using a 5-point Likert scale ranging from 1 (not useful) to 5 (very
useful), and an additional space was given so that they could put down the reason for their choice on a Likert scale. The
questionnaire was administered to point out the plausible reasons why the learners incorporated certain elements from the
model text into their rewriting.

3.3. Procedure

During the first treatment session, each participant received a piece of paper including a writing prompt. The prompt
follows: “Do you agree or disagree that smoking should be banned in public places?” Participants were asked to individually
write an argumentative essay in English on the piece of paper. They were also asked to write a minimum of 200 words, and all
of their essays were slightly over the limit. They were not allowed to use dictionaries. This instruction was given to investigate
what kinds of linguistic problems learners notice as they write (Research Q1) and to determine if they could find solutions to
the problems only from the models provided (Research Q2). While composing, all participants were also asked to take notes
on the problems that they experienced on a separate piece of paper in their native language. They were provided with a list of
examples of note-taking episodes, as in Hanaoka’s (2007) study: “I don’t know how to say X in English; “I wrote X, but I am
sure if this is correct”; and “What is the past tense of X” (p. 463). Since they were asked to take notes in their native language
(Korean) and were provided with specific instructions on how to take notes on problems they noticed, along with specific
examples, completing both the writing and note-taking tasks during the first treatment session was not likely to cause heavy
cognitive loads. Participants spent approximately 50 min completing the first stage. On average, they spent 30e35 min
writing the essay and used the remaining time to make notes about their problems. Because they were given sufficient time to
complete the writing and note-taking tasks, they might not have been under time pressure. The researcher collected par-
ticipants’ writing and the notes they had taken.
Three days after the first writing stage, only the experimental group received two model texts, and they were given 30 min
to compare their own writing against the model texts (see Appendix A). While comparing, they were instructed to put down,
on a separate paper, anything they found useful in the model text. The following examples of note-taking episodes were again
offered, as in Hanaoka (2007): “I couldn’t say X but A puts Y”; and “I have expressed this idea as X, and A puts it Y” (p. 463). As
for the control group, they were asked to self-correct their initial writing. They were asked to take notes on the errors they
noticed and to jot down self-corrections of their errors on a separate sheet. Upon completion, their notes and original writing
were collected. In the following week, the same writing prompt was distributed to each participant. They did not have access
to their initial drafts or notes while rewriting to allow analysis of their noticing and subsequent incorporations of errors into
the rewritten text. After the rewriting task, they filled out a questionnaire. The participants were given up to 50 min to rewrite
their argumentative essay.

3.4. Analysis

In order to investigate the types of problems that participants had noticed at the initial writing stage (Research Q1) and the
features they had noticed from the model texts at the comparison stage (Research Q2), the notes taken by the participants
 E.Y. Kang / System 89 (2020) 102196 5

were analyzed based on the coding scheme slightly adapted from Hanaoka’s (2007) study. Table 1 shows examples of the
coding categories.
To determine whether model texts helped learners improve their subsequent rewriting (Research Q3), participants’ initial
writing and rewriting were assessed based on an analytical scoring rubric to investigate particular areas of improvement.
Similar to the coding scheme for notes, four aspects of participants’ essays were scored: vocabulary, grammar, content and
organization. Vocabulary and grammar each measured learners’ use of accurate forms and meanings of words and gram-
matical elements, whereas content assessed relevant ideas, well supported topic sentences, and convincing supporting
sentences. Finally, organization pertained to the extent to which ideas were developed in a logical fashion and arranged
cohesively and coherently. For each component, the highest and lowest possible scores were 3 and 0, respectively, with
twelve as the highest total score. The scoring criteria were adapted from the Tests in English for Educational Purposes (TEEP)
Attribute Writing Scale (Weir, 1990) (see Appendix B). This scoring scale was chosen mainly because it “was extensively
piloted and revised to make sure that it could be applied reliably by trained raters” (Weigle, 2002, p.115). The analytic rubric
also allowed the researcher to pinpoint which aspects of writing (e.g., lexis, grammar, content, and organization) could be
improved as a result of reading model texts. The four scores from each sub component were then combined to produce a
single global score. The researcher and a native speaking English teacher who was pursuing a Ph.D. in TESOL assessed the
drafts. The raters first scored three papers together to standardize the scoring process, and the remaining papers were
evaluated separately by each rater. The interrater reliability was 97.6%. Any differences were resolved through discussion.
A MANOVA was conducted to investigate whether there is any difference between the model and the control groups in any
of the variables to be studied. Before a MANOVA was run, normal distribution of the dependent variables was checked. The
data fails this assumption, but the test was run regardless as the MANOVA is fairly robust to deviations from normality
(Tabachnick & Fidell, 2001). In addition, the dependent variables were checked for multicollinearity and multivariate outliers,
but no serious violations were found.

4. Results

The first research question was concerned with the types of problems the learners would notice during the initial
argumentative writing task. To investigate this question, participants’ notes were tallied and categorized as pertaining to
lexis, grammar, content, organization, or other issues. Recall that “content” refers to issues related to the generation or
development of ideas. In addition, as shown in Table 1, organizational issues include difficulties in arranging ideas at the
discourse level, whereas “other issues” refers to any issues that did not fit into lexis, grammar, content, or organization. The
results (Table 2) indicated that most learners encountered lexical problems (60.3%) at the moment of writing rather than
grammar (17.2%), content (13.2%), organization (8.5%) or other issues (0.8%). Such vocabulary-focused noticing was observed
in both groups. In addition, learners’ notes indicated that most learners noticed gaps in their knowledge of lexical terms such
as secondhand smoke, ban, harmful, or respiratory disease. With regard to grammar, most participants reported their uncer-
tainty about the use of correct propositions. There were not many instances of notes coded as belonging to the category of
other issues, but the few notes that were categorized as such included queries about the number of supporting ideas
necessary to write a convincing argumentative essay. As for group differences, there were no significant differences in the
total number of problem features noticed by the control and experimental groups (t ¼ 1.01, df ¼ 39, p > 0.05). MANOVA tests
were conducted to find out differences in each type of the problem features noticed (i.e., lexis, grammar, organization, and
content) by the two groups, but no significant differences were found (Wilks’s lambda test statistics ¼ 0.63, F (4, 35) ¼ 1.71,
p < 0.5). This suggests that the participants in both groups noticed similar problems at the moment of initial writing.
The second research question addressed the types of features L2 learners would notice while comparing their essays to
two model texts. In line with Stage 1, lexically-oriented noticing was found in Stage 2. The experimental group mainly noticed
the vocabulary (56%) suggested in the models, as shown in Table 3. In addition, grammar accounted for 16.5% of features of the
noticed from the models, content for 15.7%, discourse for 4.7% and other issues for 7.2%.

Table 1
Coding scheme for notes.

Category Description Examples (originally written in Korean)

Lexis Lexical items - Is ‘indirect smoking’ a correct expression in English?
- I wrote ‘bad stuff’ but the model text includes a word ‘substance.’
Grammar Particular features of grammar - I am not sure if I have to use ‘to’ or ‘for.’
- The model text used present perfect tense in the following sentence: ‘studies have shown
that~”
Content Generation or development of ideas - I cannot come up with any more reasons why smoking should not be allowed in public places.
Organization Structure or organization of ideas. - How can I arrange two opposite ideas? Do I have to include each opposing idea in a separate
paragraph?
- I did not indent the first sentence of each paragraph, but the writer of the model text did.
Others Notes that did not fit into any specific - This model is well-written and shows enough reasons why second-hand smoking should be
category banned.
6 E.Y. Kang / System 89 (2020) 102196

Table 2
Frequencies of problems noticed (stage 1).

All (N ¼ 40) Model Group (N ¼ 20) Control Group (N ¼ 20)

n % Mean SD n % Mean SD n % Mean SD

Lexis 284 60.3 7.0 2.1 140 63.3 6.8 1.8 144 57.7 7.2 0.9
Grammar 81 17.2 1.6 1.8 38 17.2 1.7 1.1 43 17.2 1.5 0.7
Contentrowhead 62 13.2 1.3 2.4 25 11.3 1.3 1.8 37 14.8 1.2 0.4
Organization 40 8.5 0.8 1.2 15 6.8 0.8 0.9 25 10 0.5 0.5
Other 4 0.8 0.3 0.3 3 1.4 0.2 0.4 1 0.4 0.4 0.2
Total 471 100 10.6 7.8 221 100 10.8 1.2 250 100 10.3 0.5

To answer the third research question, which was concerned with the effects of model texts on L2 writing development,
learners’ initial (pretest) and rewriting (posttest) were compared. Recall that learners’ initial writing and rewriting tasks were
scored using an analytic scoring rubric (Appendix B). Table 4 presents the writing test results. With respect to the total score
on the pretest, there was no significant difference between the two groups [t ¼ 0.98, df ¼ 39, p > 0.05]. As shown in Table 4,
both the control and experimental groups showed posttest improvements, but the experimental group performed signifi-
cantly better on the posttest [t ¼ 4.1 df ¼ 39, p < 0.05]. For the scores for the four aspects, a MANOVA test revealed that the
treatment had a significant effect on the different categories (i.e., lexis, grammar, content, and organization). The multivariate
(Wilks’s lambda test statistics ¼ 0.60, F (4, 35) ¼ 5.86, p < 0.05), and follow-up univariate ANOVA tests showed that there were
significant differences in the vocabulary [F (1, 38) ¼ 6.32, p < 0.05] and content [F (1, 38) ¼ 5.60, p < 0.05] categories. These
results suggest that the experimental group’s improvement was significantly greater for lexis and content.
In order to check to what extent such improvement can be traced to the model texts provided, the experimental group’s
notes made in Stage 1 and 2 as well as their rewriting produced in Stage 3 were analyzed. First, the problems noticed by the
experimental group (notes taken in Stage1) were categorized as either solvable or unsolvable depending on whether the
problems could be resolved based on the model texts (Hanaoka, 2007). For instance, if a note addressed a query regarding a
certain expression in English, and a relevant expression was offered in the models, the note was classified as ‘solvable’.
Specifically, in the current study, many participants’ notes indicated that they did not know a specific English word for the
involuntary inhaling of smoke from other people’s cigarettes. Model A provided a solution, “secondhand smoking” and Model
B also included another possible solution, “passive smoking”. As possible solutions to learners’ noticed problems were
presented in the model texts, the lexical problem noticed by the participants was regarded as “solvable”. An “unsolvable”
problem, however, indicated problems for which solutions did not exist within the model texts. Consistent with the findings
of previous studies, among the features identified as solvable, a majority was related to vocabulary. Out of 140 lexical
problems noticed by the learners, 104 (74.3%) were solvable and 36 (25.7%) unsolvable. The learners were able to notice most
of the solutions (n ¼ 95, 91%) while the reading model texts, as suggested by the notes produced in Stage 2, and they were able
to incorporate 62% of them in their rewritten output. This suggests that a large amount of words from the models were
incorporated into the subsequent rewriting produced by the treatment group, resulting in improvement in vocabulary use.
However, the learners were also able to notice words that they did not initially report as problems in Stage1. That is, from
reading the model texts, learners not only picked up new expressions that they had felt the need for in writing their initial
drafts, but they also picked up additional expressions to better convey their meaning in the second drafts.
As for the other categories, all of the features the learners noticed as problems in Stage 1 were much less solvable (content:
42%, organization: 28%, grammar: 12%) compared to vocabulary. However, the learners incorporated content and organi-
zational features that they did not initially address as problems in Stage 1. They seemed to notice propositions while reading
model texts and incorporated them into their writing in Stage 3, which resulted in their improvement in content. As for
organization, many students indented the first sentence of each paragraph in their rewriting after reading model texts.
As for participants’ perceived usefulness of model texts, it was rated on a 5-point scale, where 1 ¼ not useful at all and
5 ¼ very useful. It was found that most students perceived models as somewhat helpful in improving their writing (M ¼ 3.7,
SD ¼ 0.97). Their responses to open-ended questions indicated that they were able to learn new words and phrases as well as
become familiar with the organizational structure of argumentative essays and the use of supporting details. Fifteen

Table 3
Frequencies of features noticed from a model (stage 2).

Model Group (N ¼ 20)

n % Mean SD
Lexis 132 56 7.2 1.1
Grammar 37 15.7 2.1 1.1
Content 39 16.5 2.3 0.9
Organization 11 4.7 0.5 0.2
Other 17 7.2 0.8 0.3
Total 236 100 12.9 0.7
 E.Y. Kang / System 89 (2020) 102196 7

Table 4
Descriptive statistics for the pre- and posttest scores.

Control Model Group

Pretest Posttest Pretest Posttest

M SD M SD M SD M SD
Total 6.25 2.80 7.34 3.09 6.31 2.44 9.97 2.36
Content 1.15 0.76 1.54 0.69 1.19 0.91 2.57 0.71
Organization 1.27 0.72 2.13 0.79 1.30 0.47 2.37 0.84
Grammar 2.18 0.79 2.22 0.77 2.13 0.61 2.22 0.75
Vocabulary 2.15 0.82 2.28 0.89 2.32 0.78 2.97 0.35

Note. Maximum score: content ¼ 3, organization ¼ 3, grammar ¼ 3, vocabulary ¼ 3, total ¼ 12.

participants (75%) stated that their understanding of text coherence and organization of argumentative essays improved as a
result of the model texts. Ten participants (50%) mentioned that they were able to identify problems in their own writing by
reading the model texts. Some of them also said that such awareness helped them to improve their rewriting. Two partic-
ipants (10%) commented that they had hoped to read more than two model texts so that they could improve their writing
more. However, there were also students who expressed the model texts were not enough for them to correct errors. These
students stated that although they learned new words, and expressions, and became aware of more ideas on content from the
models, they would have preferred to receive explicit corrections. In addition, three students stated that there were unfa-
miliar with some expressions in the model texts, and avoided using these expressions in their rewriting.

5. Discussion

The main purpose of this study was to determine the usefulness of model texts as corrective feedback tools. Although the
existing studies on the role of model texts in L2 writing used a picture-description writing task and one-paragraph model
texts, this study adopted a different writing task, namely argumentative writing, and models consisting of multiple para-
graphs. Furthermore, by employing note-taking, this study was able to investigate what kinds of problems EFL learners
noticed at the moment of writing (Research Q1) and what they noticed from model texts (Research Q2). It further examined
whether model texts contributed to improvement in students’ rewriting using an analytic scoring rubric (Research Q3).
As for the first research question, learners’ notes indicated that they mainly noticed their lexical problems at the moment
of writing. This finding is in line with previous research on model texts (e.g., Coyle & de Larios, 2014; Coyle et al., 2018; Guirao
et al., 2015). Similarly, VanPatten (2004) argues that L2 learners prioritize meaning over grammatical forms, implying
meaning-based processing as the default processing approach. In the current study, however, it is important to note that
participants noticed much fewer vocabulary problems (60.3%) than those reported in previous studies. For instance, according
to the data reported by Hanaoka (2007) and other studies, the lexical problems learners noticed accounted for more than 80%
of the total problems (e.g., Coyle & de Larios, 2014; Guirao et al., 2015). The possible reason for the difference might be related
to the type of the writing tasks used. Previous studies employed a picture-based descriptive task, which inherently required
learners to describe items and actions reflected in the picture prompts. While completing the task, the learners might have
made effort to come up with suitable nouns, verbs, adjectives, and adverbs to describe the picture prompts, which might have
directed most of their attention to their lexical deficiencies. However, an argumentative task used in this study required
learners to not merely retrieve relevant vocabulary but to develop their argument and clearly arrange their ideas to convince
readers. Therefore, the participants in this study diversified their linguistic concerns with content and organization in
addition to vocabulary. This finding illustrates that genres of writing tasks can affect how learners distribute their attention at
the moment of writing.
Moreover, in contrast to the previous studies (e.g., Coyle & de Larios, 2014; Guirao et al., 2015), which reported less than 5%
of grammar-related problems, participants in this study paid more attention to their grammar-related problems (17.2%),
which increase the reported percentage. The superior awareness of grammatical issues at the composing stage helped them
more readily direct their attention to grammatical features embedded in the models during the comparison stage. This higher
attention to grammatical features reported in the study seems to be related to the age of the participants. Although other
studies mainly focused on young learners such as primary school children (e.g., Guirao et al., 2015; Coyle & de Larios, 2014;
García Mayo & Loidi Labandibar, 2017), this study investigated high school students. It is plausible that with a higher level of
cognitive abilities and L2 proficiency, the high school participants had more metalinguistic awareness of grammar rules. Due
to such advantages, the participants in this study might have been able to notice more grammatical features during the initial
writing and comparison stages. Furthermore, this study was conducted in a setting where well-formedness and accurate use
of language were emphasized. L2 learners in such contexts tend to be better at noticing grammatical forms than learners in a
context where instruction is meaning-oriented (Sheen, 2004). The instructional setting, therefore, might have played a role in
raising learners’ grammar awareness.
The second research question asked what L2 writers noticed from model texts when they were asked to compare their
drafts against the models. L2 writers mainly noticed the vocabulary suggested in the models, but one notable finding of this
study is that the participants were also highly concerned about content compared to those in other studies (e.g., Guirao et al.,
8 E.Y. Kang / System 89 (2020) 102196

2015; Coyle & de Larios, 2014; García Mayo & Loidi Labandibar, 2017; Hanaoka, 2007) while reading model texts. Participants
of the model group were able to solve their content-related issues at the comparison stage. In this study, learners were
required to argue either for or against smoking in public places. Interestingly, all participants were against the issue. To
successfully complete the writing task, learners had to develop ideas to support their opinions. Most learners were not able to
come up with suitable ideas at the initial stage of writing, but they seemed to pick up some persuasive statements and
relevant facts to support their arguments while reading model texts. Based on such supporting details in the model texts,
participants were able to make stronger arguments when they were asked about the topic at the rewriting stage. To provide
more concrete examples, parts of one student’s initial and rewritten drafts are shown below.

5.1. (Student A in the model group, initial draft)

For non-smoking people, the smoke is really threatening. Indirect smoking is more harmful for health. By the law, some
public places prohibit smoking in there. But smoking people often reject this. I think it should be punished. Smoking in public
places is an act of not considering other people. Many people irritate from the smoke and smell of that. Moreover, children or
pregnant woman can be susceptible.

5.2. (Student A in the model group, rewritten draft)

Smoking can cause a lot of health problems for both smokers and non-smokers. According to research, second hand smoke
is very harmful. It can cause many diseases such as lung cancer and asthma. It is more harmful to children because their
bodies are developing.
Tobacco has many harmful chemicals in it. There are cancer-causing substances. When smokers and non-smokers are in
the same room, they inhale almost same toxins. It could lead to premature death for both smokers and non-smokers.
As suggested in the examples, Student A revised her ideas to be more specific. In the first draft, she did not specifically
explain why smoking is threatening and harmful. However, she elaborated her ideas by including supporting sentences and
more details, which are underlined above. She also added concrete examples, such as lung cancer and asthma, and provided
reasons to explain why smoking harms non-smokers, too. After reading the model texts, she might have realized clearer
examples and more of them could be included, and as a result, she addressed the issue by incorporating new content into her
updated draft. Interestingly, the new ideas are from the model texts provided during Stage 2.
The third research question asked whether model texts could contribute to improvement in students’ rewriting. Partic-
ipants in the model group generally improved their rewritten essays by incorporating content or other elements from the
model texts. This was also demonstrated by their superior overall writing scores compared to those of the control group. In
addition to the higher overall score, participants in the model group specifically received higher scores on vocabulary and
content compared to those in the control group. The result does not come as a surprise, considering that learners’ higher level
of noticing vocabulary was found at the initial writing and comparison stages. Learners’ use of more specific and context-
appropriate words also led to the higher scores on content. This was because learners’ ideas were well-developed and
supported with content-appropriate words. The overall higher score on the model group’s rewriting supports the positive
effect of models found in other studies (e.g., Ca novas Guirao et al., 2015; Coyle & de Larios, 2014; Hanaoka, 2007), but by
specifically measuring several aspects of language skills during the posttest, this study singled out that model texts are helpful
particularly improving the content and vocabulary of learners’ writing.
As for questionnaires regarding the participants’ perceived usefulness of model texts, the participants generally perceived
models as being helpful when rewriting their initial draft. However, several participants expressed their negative views on
the model texts in the questionnaire. As mentioned in the result section, the reported negative views were related to (i)
participants’ lack of understanding of some of the expressions or vocabulary included in the models and (ii) their preference
for more explicit corrections. For instance, three students reported that they were not able to fully use the models because
there were several expressions and vocabulary words that they did not understand. In this study, model texts had been
written based on the results of the pilot-test, but the proficiency levels of all participants were not completely homogeneous.
Specifically, the TOEIC scores of the three students that expressed negative views were at the lower end of the proficiency
distribution among the participants, and therefore, they might not have fully benefitted from the model texts. This finding
also adds support to Hanaoka (2007) and Guirao et al. (2015), who showed that more proficient learners tend to notice more
linguistic features from model texts than lower-level learners. This suggests that lower-level learners need more guidance
and support when using models as feedback tools.
Another reported factor influencing participants’ attitudes toward model texts was their personal preference for explicit
correction. Research investigating L2 learners’ preference for corrective feedback has shown that English language learners
tend to prefer explicit correction (Lee, 2013). Considering that using model texts is an implicit way of providing corrections,
not all learners might have recognized it as a form of feedback, even though they learned new words and expressions from the
models. To maximize benefits from model texts, teachers could demonstrate how to use models to improve their writing and
give additional instruction on the linguistic features embedded in the models or writing skills.
 E.Y. Kang / System 89 (2020) 102196 9

6. Conclusion

This study investigated the usefulness of models as a feedback tool using a three-stage writing task. The results show that
the use of model texts can be helpful to draw learners’ attention to various aspects of writing, such as vocabulary, grammar,
and content. Learners noticed the linguistic features of model texts while reading them and subsequently incorporated the
forms into their rewriting. An analysis of learners’ performance during a posttest revealed that model texts were more
effective in enhancing learners’ vocabulary and content than other aspects of writing. The importance of this finding relates to
the fact that previous studies only employed picture-description writing tasks (Guirao et al., 2015; Coyle & de Larios, 2014;
Hanaoka, 2007; Yang & Zhang, 2010; García Mayo & Loidi Labandibar, 2017; Coyle et al., 2018; Hanaoka, 2007) and single-
paragraph texts as models. In this study, by using a different writing task, such as an argumentative essay, and longer
model texts, it further strengthened the positive effects of model texts on students’ writing and found that model texts can
help learners improve their writing by incorporating vocabulary, content, and grammar from models.
The study, however, has limitations that need to be considered for future research. First, this study only used note-taking
to measure what participants noticed at the moment of writing and comparison. However, because the act of taking notes is
slow and painstaking, participants did not take explicit notes on everything they noticed. Future studies may employ different
measures, such as think-alouds and stimulated recall, to compare the results. Second, this study did not employ a delayed
posttest, so it was difficult to determine whether the positive effects found in this study persist over time. In addition, the
participants wrote an argumentative essay on a single topic. Therefore, future studies should use a longitudinal research
design with an extended number of writing tasks to better understand the effect of model texts on L2 learners’ writing.
Finally, as for the genre of writing task, this study used an argumentative essay. Specifically, all participants wrote about why
they disagree with public smoking and then received two model texts arguing against public smoking. Alignment of content
between participants’ writing and model texts made it possible to investigate what elements learners incorporated into their
rewriting using the models. However, based on the study, it is difficult to find out what learners do in their rewriting when
they are given model texts entirely different from what they wrote initially in terms of content. To shed more light on the role
of model texts in L2 writing, future studies using argumentative tasks might also consider employing model texts that reflect
more than one view. Last but not least, although this study confirmed the previous finding that more proficient learners tend
to notice more linguistic features from model texts than lower-level learners, future research exploring how individual
differences affect the way learners use model texts may provide more information about what type of learner benefits more
from models.

CRediT authorship contribution statement

Eun Young Kang: Conceptualization, Methodology, Writing - original draft, Writing - review & editing.

Appendix C. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.system.2019.102196.

Appendix A. Model texts used in the study

[A model text 1]

Smoking negatively affects health. It can even cause deadly diseases like cancer. Many countries have already banned
smoking in public places. In my opinion, other countries should follow suit and ban it.
There are several benefits to banning smoking in public places. A smoking ban will definitely improve the quality of the air
we breathe. Cigarette smoke contains cancer-causing chemicals. In addition to cancer, smoking causes several other health
problems. In fact, studies have shown a tremendous rise in the occurrence of lung cancer among people who smoke regularly.
Some studies have also shown that people who smoke more than three cigarettes a day have increased chances of developing
cancer. Smoking may also cause other problems like heart attack, asthma, and respiratory illnesses.
The biggest problem with smoking is that exposure to secondhand smoke, which is the smoke exhaled by smokers, also
negatively affects health. Non-smokers who are exposed to secondhand smoke can increase their risk of developing heart
disease, lung cancer, or respiratory diseases. It is particularly harmful for young children and pregnant women. If people are
allowed to smoke in crowded public places like railway stations or bus stands, its adverse effects will have to be borne by all
people standing next to the smoker. Banning smoking in public venues is the only way to protect non-smokers from the
dangerous chemicals in secondhand smoke.
Another serious problem that smoking causes is pollution. Smoking not only aggravates pollution but it also causes the
depletion of the ozone layer which protects us from the sun’s harmful ultraviolet rays.
In conclusion, I personally believe that all governments should ban smoking in public places because smoking plays a
significant role in damaging our health and our environment. This might cause some inconvenience to chain smokers, but
ultimately, this ban will benefit them as well.
10 E.Y. Kang / System 89 (2020) 102196

[A model text 2]

Medical studies have shown that smoking not only leads to health problems for the smoker, but also for people close by. As
a result of this, many believe that smoking should not be allowed in public places, while others oppose such a ban. Although
both sides have put forth rationales to support their viewpoints, I strongly believe that a ban is the most appropriate course of
action.
Opponents of such a ban argue against it for several reasons. Firstly, they say that “passive” or “secondhand” smokers,
people who are exposed to the smoke exhaled by the people smoking, have the choice to avoid the harmful effects of smoking.
They can simply choose not to visit places where smoking is permitted. Moreover, opponents of a smoking ban believe a ban
would possibly drive many bars and pubs out of business, as smokers would not go there anymore. They also argue that it is a
matter of free will; since smoking is not against the law, individuals should have the freedom to smoke where they wish.
Despite the various reasons why some people are against prohibiting smoking in public places, there are more convincing
arguments in favour of a ban. First and foremost, it has been proven that tobacco consists of carcinogenic compounds which
cause serious harm to a person’s health. Given that non-smokers who are exposed to secondhand smoke inhale many of the
same toxins as smokers, smoking can cause serious illnesses such as heart disease, lung cancer, and asthma for both smokers
and nonsmokers alike. It is simply not fair for nonsmokers to have to suffer these consequences. Even though some people
argue that nonsmokers can choose not to visit places where smoking is permitted, it is unfair to impose this obligation on
nonsmokers and restrict the places that they can visit. Moreover, although some people worry that smoking will drive pubs
and restaurants out of business, I think there are many ways that pubs and restaurants could adapt to a smoking ban without
going out of business. For example, they can allow restricted smoking areas. In any case, people’s health is far more important
than business.
In conclusion, it is clear that it should be made illegal to smoke in public places. This would improve the health of
thousands of people, and that is most definitely a positive development.

Appendix B. Scoring rubrics

Scoring Rubric (adapted from Weir, 1990)

Content Organization Vocab Grammar

0 - No relation to the writing prompt - No organization of ideas -Almost no adequacies in vocabulary -Almost all grammatical pat-
or content terns are inaccurate
1 -Limited relevance to the writing prompt -Very little organization of -Frequent inadequacies in vocabulary or -Frequent grammatical
ideas or content repeated use of vocabulary inaccuracies
2 -Generally relevant but there are some gaps -Some organizational skills -Some inadequacies in vocabulary or -Some grammatical
or redundant information. are evident. circumlocution inaccuracies
3 - Relevant and adequate content - Clear organization in - Adequate use of vocabulary - Almost no grammatical
general -Almost no inadequacies in vocabulary inaccuracies

References

Canovas Guirao, J., Roca de Larios, J., & Coyle, Y. (2015). The use of models as a written feedback technique with young EFL learners. System, 52(1), 63e77..
Council of Europe. (2001). Common European Framework of reference for languages: Learning, teaching, assessment. Cambridge: Cambridge University Press.
Coyle, Y., & de Larios, J. R. (2014). Exploring the role played by error correction and models on children’s reported noticing and output production in a L2
writing task. Studies in Second Language Acquisition, 36(3), 451e485.
Coyle, Y., & de Larios, J. R. (2014). Exploring the role played by error correction and models on children’s reported noticing and output production in a L2
writing task. Studies in Second. Language Acquisition, 36(3), 451e485.
Coyle, Y., Guirao, J. C., & de Larios, J. R. (2018). Identifying the trajectories of young EFL learners across multi-stage writing and feedback processing tasks
with model texts. Journal of Second Language Writing, 42, 25e43.
García Mayo, M. P., & Loidi Labandibar, U. (2017). The use of models as written corrective feedback in EFL writing. Annual Review of Applied Linguistics, 37,
110e127.
Gass, S. M., & Mackey, A. (2007). Input, interaction, and output in second language acquisition. Theories in second language acquisition: An introduction.
Guirao, J. C., de Larios, J. R., & Coyle, Y. (2015). The use of models as a written feedback technique with young EFL learners. System, 52, 63e77.
Hanaoka, O. (2007). Output, noticing, and learning: An investigation into the role of spontaneous attention to form in a four-stage writing task. Language
Teaching Research, 11(4), 459e479.
Hanaoka, O., & Izumi, S. (2012). Noticing and uptake: Addressing pre-articulated covert problems in L2 writing. Journal of Second Language Writing, 21(4),
332e347.
Izumi, S. (2003). Comprehension and production processes in second language learning: In search of the psycholinguistic rationale of the output hy-
pothesis. Applied Linguistics, 24(2), 168e196.
Kang, E., & Han, Z. (2015). The efficacy of written corrective feedback in improving L2 written accuracy: A meta-analysis. The Modern Language Journal,
99(1), 1e18.
 E.Y. Kang / System 89 (2020) 102196 11

Kang, E. Y. (2010). Effects of output and note-taking on noticing and interlanguage development. Working Papers in TESOL and Applied Linguistics, 10(2),
19e36.
Lee, E. J. (2013). Corrective feedback preferences and learner repair among advanced ESL students. System, 41(2), 217e230.
Qin, J., & Karabacak, E. (2010). The analysis of Toulmin elements in Chinese EFL university argumentative writing. System, 38(3), 444e456.
Schmidt, R. (1990). The role of consciousness in second language learning. Applied Linguistics, 11, 129e158.
Schmidt, R. (2001). Attention. In P. Robinson (Ed.), Cognition and second language instruction (pp. 3e32). Cambridge: Cambridge University Press.
Sheen, Y. (2004). Corrective feedback and learner uptake in communicative classrooms across instructional settings. Language teaching research, 8(3),
263e300.
Swain, M. (1995). Three functions of output in second language learning. In G. Cook, & B. Seidhofer (Eds.), Principles and practice in applied linguistics (pp.
125e144). Oxford: Oxford University Press.
Tabachnick, B. G., & Fidell, L. S. (2001). Using multivariate statistics (4th ed.). Boston: Allyn and Bacon.
VanPatten, B. (2004). Processing instruction. Mahwah, NJ: Lawrence Erlbaum.
Weigle, S. (2002). Assessing writing. Cambridge, UK: Cambridge University Press.
Weir, C. J. (1990). Communicative language testing. New York, NY: Prentice Hall.
Yang, L., & Zhang, L. (2010). Exploring the role of reformulations and a model text in EFL students’ writing performance. Language Teaching Research, 14(4),
464e484.