Professional Documents
Culture Documents
Contextual Learning of L2 Lexical and Grammatical Collocations With and Without Typographic Enhancement
Contextual Learning of L2 Lexical and Grammatical Collocations With and Without Typographic Enhancement
PII: S0346-251X(24)00017-4
DOI: https://doi.org/10.1016/j.system.2024.103235
Reference: SYS 103235
Please cite this article as: Toomer, M., Elgort, I., Coxhead, A., Contextual learning of L2 lexical and
grammatical collocations with and without typographic enhancement, System (2024), doi: https://
doi.org/10.1016/j.system.2024.103235.
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.
typographic enhancement
Email: toomermark4@gmail.com
f
oo
Telephone: +64 21 246 9707
r
Irina Elgortb
b
-p
School of Linguistics and Applied Language Studies
re
Victoria University of Wellington
PO Box 600
lP
Wellington 6140
New Zealand
na
Email: irina.elgort@vus.ac.nz
Phone: +64 4 463 5970
ur
Averil Coxheadc
Jo
c
School of Linguistics and Applied Language Studies
Victoria University of Wellington
PO Box 600
Wellington 6140
New Zealand
Email: averil.coxhead@vuw.ac.nz
Telephone: +64 63 64 3449
CONTEXTUAL LEARNING OF L2 LEXICAL AND GRAMMATICAL
Abstract
collocational knowledge. However, it is less clear how typographic enhancement affects the
f
oo
(preposition+noun) collocations by Chinese speakers over two days. We manipulated the
r
learning conditions by bolding or not bolding multiple instances of the target collocations in
-p
reading texts. A cued-recall (gapfill) post-test was used to measure collocational knowledge.
re
We observed greater accuracy of responses with grammatical (but not lexical) collocations
lP
when the collocations were bolded than when they were not bolded. We argue that bolding
na
likely made prep+noun collocations more perceptually salient to the learners during
benefit from drawing explicit attention to the whole expressions in written input.
Jo
1. INTRODUCTION
(Schmitt, 2010). The acquisition of second language (L2) MWUs in reading can benefit from
number of studies have found that one type of input enhancement known as input flooding
1
(i.e., repetition of the MWUs in the input) is effective, and its effectiveness increases with an
increased number of repetitions (e.g., Hulstijn et al., 1996; Parkibakht & Wesche, 1997;
Webb, 2007). Another type of input enhancement, typographic enhancement (TE), has been
generally found to be more effective for the learning of MWUs than no-TE (e.g., Boers et al.,
2016; Peters, 2012; Sonbul & Schmitt, 2013; Szudarski & Carter, 2016). In an eye-tracking
study, Choi (2017) observed that participants spent more time reading bolded than unbolded
collocations.
Research into the learning of L2 collocations has, so far, examined mostly the
f
oo
acquisition of lexical collocations (e.g., adjective+noun and verb+noun collocations—see
r
Bahns & Eldaw, 1993; Laufer & Waldman, 2011; Nesselhauf, 2003). Fewer L2 studies have
-p
investigated the acquisition of other types of collocations, such as grammatical collocations
re
(e.g., preposition+noun collocations), even though the latter are very common (Durrant,
lP
2009; Gitsaki, 1999). Also, few studies have compared how different instructional
na
approaches affect the learning of different types of L2 collocations (but see Gitskai, 1999;
Huo, 2014). It is possible that, because grammatical collocations are less salient than lexical
ur
more beneficial for learning grammatical than lexical collocations. The current experiment
investigates the learning of L2 lexical and grammatical collocations from reading, with or
2. LITERATURE REVIEW
words with a moderate level both of transparency and restriction (fixedness) (Howarth,
2
1998a; Nesselhauf, 2003; Paquot & Granger, 2012). In the frequency-based tradition,
multiword units (including collocations) are identified mostly through phrase frequency in
(e.g., t-scores) and mutual information (Hunston, 2002; Sinclair, 1987). Much of the recent
research into MWUs has used a mixed-method approach, combining aspects of both
traditions (e.g., Ackerman & Chen, 2013; Simpson-Vlach and Ellis, 2010).
Collocations in the phraseological tradition are frequently divided into two types: lexical
collocations, comprising two content words and possibly an article (e.g., strong tea, play the
f
oo
guitar); and grammatical collocations, consisting of at least one content word and at least one
r
preposition or a grammatical structure, such as a clause or an infinitive (e.g., under attack;
-p
agreement that + clause) (Benson, Benson, & Ilson, 1997; Gyllstad, 2014; Howarth, 1998a).
re
The majority of phraseological research appears to have been conducted into lexical
lP
collocations (Henriksen, 2012), with some researchers (e.g., Ackerman & Chen, 2013;
na
Howarth, 1998b) claiming that lexical collocations are difficult for second-language learners
to master because they are less fixed and thus less predictable than grammatical collocations.
ur
combinations (e.g., Bahns & Eldaw, 1993; Laufer & Waldman, 2011; Nesselhauf, 2003),
which may be the most common source of lexical-collocation errors among learners (Gitsaki,
1999; Huo, 2014). Verb+noun combinations tend to be abstract, polysemous, and often high-
frequency (Altenberg & Granger, 2001), thereby causing problems for learners.
Grammatical collocations can also be problematic (Hemchua & Schmitt, 2006). Schmitt
(2010, p. 55) claims that function words are generally more difficult for language learners to
learn than content words. A common type of grammatical collocation is the prepositional
(e.g., at school, account for) (Biber et al., 1999; Sicherl, 2004). It is possible that prepositions
3
are mostly phraseological in nature in that each preposition is bound to another word (Cosme
& Gilquin, 2008; Mueller, 2011). Errors with prepositions, which include errors with
Schmitt, 2006; Catalán, 1996; Tyler, Mueller, & Ho, 2011). Although such errors do not
generally cause major communication problems, they sometimes change the intended
meaning (e.g., …discussions about laws can go of course vs. …discussions about laws can
go off course…, Catalán, 1996). Thus, knowing how to correctly use prepositions is important
for accuracy and the overall effectiveness of oral and written communication (Howarth,
f
oo
1998b; Catalán, 1996).
r
There are several barriers to learning prepositional collocations. First, prepositions are
-p
generally polysemous, having many different but related senses, many of them figurative
re
(Boers & Demecheleer, 1998). Second, prepositions often do not have congruent translation
lP
equivalents in the first language of the learner (Cosme & Gilquin, 2008). Third, there is
na
variability in the usage of prepositions as shown in the disagreement among native speakers
as to the best choices of prepositions in use (Chodorow, Gamon, & Tetreault, 2010).
ur
have been largely ignored by L2 researchers (e.g., Rundell, 2010). One possible reason for
collocations (Durrant, 2009). One of the few studies to have explored the learning of
grammatical collocations was conducted by Gitsaki (1999). In a gapfill task, she found that
collocations, but more accurately than noun+preposition collocations, by Greek junior high
school students. Gitsaki (1999) suggested that preposition+noun collocations are easier to
learn because they are quite fixed and rule-governed, whereas noun+preposition collocations
4
are “also fixed but less regular, more unpredictable (i.e., no rules can be generated for them)”
(p. 141).
different conditions of learning collocations from reading (bolding vs. no-bolding) affect
cued recall of verb+noun and preposition+noun collocations, by learners of English. The next
section reviews research on learning MWUs from reading, including the use of input
enhancement.
f
oo
2.2. Vocabulary learning from reading and input enhancement
r
-p
Learning vocabulary from reading (or contextual learning) is a major part of second
re
language acquisition. Contextual learning takes place within the context of a meaning-
lP
focused activity. Although researchers refer to this type of learning as incidental (e.g.,
Hulstijn, 2001) to contrast it with deliberate learning, following Elgort et al. (2018) we opt
na
for contextual vocabulary learning as it is a term that does not denote any particular type of
ur
attention that readers pay to unfamiliar vocabulary during reading. This is because, in normal
Jo
L2 reading, L2 learners tend to actively infer meanings of unfamiliar words and expressions
from context and engage vocabulary learning strategies, even under so-called incidental
the learner. Input enhancement is the manipulation of input in a way which makes it more
salient to the learner; Schmidt, the first proponent of the noticing hypothesis, argued that
“more noticing leads to more learning” (Schmidt, 1994, p. 18). Some forms of input
(Boers et al., 2016; Sharwood Smith, 1991; Tomlin & Villa, 1994).
5
Input enhancement techniques include input flooding: increasing the number of
repetitions in the text and typographic enhancement (TE) (e.g., Boers et al., 2006; Ellis, 1999;
Hulstijn, Hollander, & Greidanus, 1996). Repetition is effective for learning both single
words and MWUs, and is often used in conjunction with TE. Studies have found that the
larger the number of exposures to individual words in context the better the learning (e.g.,
Elgort & Warren, 2014; Hulstijn et al., 1996; Paribakht & Wesche, 1997; Webb, 2007).
Repetition in the input can also be effective for the learning of MWUs. Durrant and Schmitt
(2010), for example, report that participants remembered more low-frequency, transparent
f
oo
combinations in a timed naming task after two exposures than one; in addition, verbatim
r
repetition of the combinations had an advantage over varied repetition. Webb, Newton, and
-p
Chang (2013) found that the more often a collocation was encountered in a graded reader (1,
re
5, 10 or 15 times) the more likely it was to be learned, as measured in post-tests of explicit
lP
knowledge. On the other hand, Pellicer-Sanchez (2017), in tests of form and meaning recall
na
and recognition, found no difference between learners’ contextual learning of six collocations
after four repetitions and after eight repetitions in a reading text; however, the lack of
ur
difference may have been due to the small number of items. Szudarski and Carter (2016)
Jo
period overall produced no more collocational learning than six exposures, as measured in
delayed post-tests of recall and recognition. A possible reason for this finding is that the
resulted in a higher burden of learning for the nouns as well as the collocations.
TE, such as bolding, italics, underlining and quotation marks, is commonly used in
key words and phrases (Bramki & Williams, 1984). The small number of studies which have
investigated the effects of TE on the learning of the forms of single words have produced
6
mixed results (e.g., Gass, Svetics, & Lemelin, 2003; Kim, 2006; Peters, 2012). In contrast,
studies investigating the effects of TE on the retention of MWUs (e.g., Boers et al., 2016;
Peters, 2012; Sonbul & Schmitt, 2013; Szudarski & Carter, 2016) have largely found an
advantage for enhanced input over non-enhanced input in offline tests of explicit knowledge.
In Choi’s (2017) eye-movement study, the enhanced treatment group read a text containing
14 bolded collocations while a comparison group read an unenhanced version of the text. The
TE group spent longer processing bolded collocations in reading than the comparison group
spent processing unbolded collocations, and the TE group performed better on a one-week
f
oo
delayed recall cloze test. Toomer and Elgort (2019) exposed advanced English-as-a-second-
r
language speakers to nine contextual occurrences of low-frequency English lexical (medical)
-p
collocations in three sessions over two days. Collocations that were bolded were more
re
accurately recognised and recalled than collocations that were not bolded.
lP
The difference in the effects of TE on learning outcomes for single words and MWUs
na
may be due to the higher salience of novel single words compared with MWUs, since single
MWUs are known, without enhancement the units may not be recognised as being unknown
Jo
or even as MWUs. Similarly, less salient collocations (e.g., preposition+noun) may benefit
more from TE than more salient ones (e.g., verb+noun). The present study incorporates
using different approaches to input enhancement1. The following research questions were
posed:
7
RQ1. Does repeated exposure to verb+noun and prep+noun collocations in written input
4. METHOD
f
oo
4.1. Participants
Seventy-eight native speakers of Chinese took part in the experiment (56 females and 22
r
-p
males) ranging in age from 17 to 35 (M = 24.19; SD = 4.43). They were ESL students
re
enrolled in the English Proficiency Programme or in degree courses at a university in an
lP
International English Language Testing system, IELTS, scores were 5.0-6.5). Their mean
score on the 2000 level of the Vocabulary Levels Test (VLT, Schmitt, Schmitt, & Clapham,
ur
2001) was 27.62 (SD 2.65), meaning that, on average, they had mastery of the most common
Jo
2000 English word families and would be able to easily understand the treatment materials
(see Materials section). The participants’ mean score for levels 2, 3 and 5 of the VLT was
Each participant who had volunteered to take part in the study was rewarded with a gift
voucher.
4.2 Materials
frequency words were selected for the experiment: 24 pairs of verb+noun collocations (e.g.,
raise the issue) and matching prep+noun collocations (e.g., at issue) with the same terminal
8
noun (see Appendix B for the final set of 38 collocations, and see the Analysis and Results
section for the reasons for exclusion). This was done to enable a direct comparison of the
learning of the two types of collocations. Our mixed-method approach to item selection used
target items and human judgements about their pedagogical value (Ackerman & Chen, 2013).
The following criteria were used when selecting target collocations for the present study:
1. At least one collocation in each verb+noun and prep+noun pair was found in the
Oxford collocations dictionary for students of English (OCD) (2nd Edition) (McIntosh
f
oo
et al., 2009). In fact, nearly all of the target collocations in the study came from the
r
OCD, whose editors have stated that they included collocations that were typical and
-p
that expressed an idea that English learners would want to express (ibid.).
re
2. The collocations comprised two or three words. If a collocation contained three
lP
4. The prepositional combinations did not have synonyms (or near-synonyms) in which
ur
two or more prepositions co-occurred with the same noun (e.g., at a distance and from
Jo
a distance).
6. The constituent words of the collocations were high-frequency words, in the first and
second most frequent thousand words of the BNC-COCA-25 list (Cobb, n.d.; Nation,
2012). The five target nouns at K3 level were trialled with intermediate-level students
9
7. The raw-frequency cut-off for each collocation was 2000 occurrences in COCA
(Davies, 2008-). Note: The original frequencies of the collocations in COCA were
obtained in 2018.
8. The collocations were semi-transparent: their meanings were neither fully transparent
nor completely idiomatic and opaque. In most of the collocations selected, the verbs
and prepositions were used in a figurative sense and the nouns in a literal sense (e.g.,
hold the belief, under protest). Transparency was assessed by two raters, one native
f
oo
We also conducted a norming study to assess the perceived naturalness of the target
r
collocations. The participants (n=28) were Chinese students studying English at two
-p
universities and an English language institute in China. Participants sought were those with
re
valid IELTS scores of 5.0 to 7.0 or the equivalent on China’s English College Test scale, i.e.,
lP
of similar or slightly higher proficiency than the participants in the main study. The Chinese
na
speakers rated the target English collocations and an equal number of matched non-
collocation fillers according to how likely they thought each phrase was to be used in
ur
English, on a scale from 1 (very likely) to 6 (very unlikely) (Appendix C). The rating task
Jo
was administered online, using Qualtrics. A small but significant difference of 0.22 between
the ratings of the target collocations and fillers was found using paired-samples t-test (95%
CI: lower=0.40, upper=0.04, t(37)=-2.50, p = .02). The mean ratings of the target collocations
was 2.70 (SD 0.43). Based on these ratings, which were close to the middle of the 6-point
scale, we concluded that the target collocations had a clear learning potential for our target
participant population. We included the perceived naturalness ratings in the data analysis.
Twelve 170-word texts containing the target collocations were created on various, mostly
business- and employment-related, topics (Appendix D). The lexical load in the texts was
controlled to ensure that each text had a lexical coverage score of 98%+ (Hu & Nation, 2000;
10
Schmitt, Jiang, & Grabe, 2011). The words assumed to be known in the texts were well-
known place names, words on the BNC/COCA list of the 2000 most frequent English words
(Nation, 2012), and a number of 3000-level words on the BNC/COCA list which a trial had
shown to be known by intermediate-level students of English. Each text contained six target
verb+noun and six prep+noun collocations, used once. Three of each collocation type were
Each text was read twice by the participants, who were exposed to a total of six
occurrences of each collocation: three occurrences in varied contexts on the first day of
f
oo
treatment, and three more occurrences on the second day of treatment (in the same texts as
r
they read on day one). The repetition of the texts on the second day of treatment was intended
-p
to strengthen the learning of the collocations by reducing the cognitive burden on participants
re
and promoting fluency of form recognition (Durrant & Schmitt, 2010). Thus, we used a
lP
combination of varied repetition and verbatim repetition (ibid.). The participants answered an
na
easy comprehension question after each text without looking back at the text. The questions
following the repeated versions of the texts on day two were different from those on day one
ur
balanced in two versions of the texts (collocations bolded in one version were not bolded in
the other version, and vice versa). Each collocation was presented in the same learning
condition to the same participant throughout the experiment (across all texts); in other words,
each participant saw a particular collocation either bolded or unbolded for all six textual
occurrences. Each prep+noun collocation was placed in the same learning condition as its
11
4.3. Procedure
n=27; subgroup 2: n=26) or the No-treatment group (n=25) in the order in which they
volunteered for the study.2 The Treatment group participants were randomly assigned to one
of the two subgroups to receive the alternate versions of the experiment, in which the learning
conditions were counterbalanced. Two variables, each with two levels, were used: learning
condition: bolding and no-bolding; and phrase type: V+N and prep+N. Participants read 12
f
texts, each containing six occurrences of V+N collocations and six occurrences of prep+N
oo
collocations. The same texts were read twice, first on day one and again on day two; on the
r
second day, the texts were presented in a randomly pre–determined order different from that
-p
on day one. Each participant read 12 bolded V+N combinations, 12 unbolded V+N
re
combinations, 12 bolded prep+N combinations and 12 unbolded prep+N combinations across
lP
all texts (see Appendix D). Overall, all texts included collocations in both TE and no-TE
na
conditions; the combined time for both text-reading sessions was about 50 minutes.
ur
The post-test was conducted on day 2, after the learning treatment. The order of the
elements of the session on the second day was: learning treatment → intervening task →
Jo
sentence reading task3 → gapfill test → Vocabulary Levels Test (VLT, Schmitt, Schmitt, &
Clapham, 2001) → yes/no self-report of prior familiarity with the target collocations. To
minimise the effect of episodic memory on the outcome measures, participants performed an
intervening task after the learning treatment: they counted backwards, subtracting by 7s from
The No-treatment group participants attended only one session. They completed only the
12
4.4. Measures
A gapfill post-test was used to measure outcomes of the contextual learning of the
participants to the purpose of the study and avoid unintentional learning of the target
collocations before the treatment phase. This was critical because one of the main-interest
predictors in our study was whether or not readers’ attention was deliberately drawn to the
target collocations through TE. To measure the degree of prior knowledge in the study, we
compared the Treatment group participants’ knowledge of the target collocations after the
f
oo
treatment with the No-treatment group participants’ knowledge, using the same tests. In
r
addition, the participants self-reported their prior familiarity with each collocation after the
-p
main post-tests. No delayed post-tests were administered since the purpose of the experiment
re
was to measure initial learning, not long-term retention (Hulstijn, 2001).
lP
The pen-and-paper gapfill post-test measured productive cued recall of the collocations (see
ur
Appendix E). The sentences (one for each target collocation) were adapted from the treatment
Jo
texts. The first word of each collocation was removed (e.g., If you are a business owner, it is
important that your business does not _______ a loss on sales.). Treatment group participants
were instructed to write the first word of the collocations from the reading texts in the gap.
An answer incorrectly spelled or with a morphological error was marked correct if the word
was easily identifiable and the provided collocation matched the target collocation.
Alternative answers (e.g., make a comparison instead of draw a comparison) were marked as
incorrect, since the purpose of the test was to evaluate how accurately the learners were able
to recall the collocations encountered in the reading-texts (and not just provide any plausible
alternative). Thus, the gapfill test measured how likely the participants were able to retrieve
13
the first word of the target collocations, having been exposed to these collocations in the
All participants completed a yes/no self-report of prior familiarity with the target collocations
(a modified version of Meara & Buxton, 1987) at the end of the study after all other post-
tests. The meaned self-report scores for each collocation (subjective measure) were added to
the meaned control group participants’ gapfill scores for the same collocation (objective
f
measures) to create prior knowledge scores (see Appendix F). These scores were included in
oo
the mixed-effects regression models of the gapfill accuracy scores in order to partial out the
r
-p
effect of prior knowledge of the collocations. The reported familiarity with the collocations
re
was not significantly different for the Treatment and No-treatment groups (t=-0.96 p=.34),
lP
meaning that participants in both groups overall had similar levels of prior familiarity with
The post-test data were analysed in two steps: in step 1, the preliminary comparison was
Jo
made between the outcome variables for the Treatment and No-treatment groups (i.e.,
response accuracy in the gapfill task); in step 2, the analyses of the same outcome variables
were performed for the Treatment group only, adding Learning Condition as the primary-
interest predictor. The goal of the preliminary analyses was to establish whether the results
for the Treatment and No-treatment groups were significantly different, i.e., to show that
some contextual learning took place even if the collocations had been partially known before
the learning treatment. In other words, the preliminary analysis was needed to confirm the
learning potential for the target collocations. On confirming the difference between the two
groups, the main analysis was conducted on the Treatment group data to establish whether
14
one of the learning conditions resulted in a better learning outcome. Our primary-interest
predictor in both analyses was Phrase Type. In addition, Group (Treatment/Control) was used
bold) was used as a primary-interest predictor in the main analysis. We also included self-
reported Prior Knowledge, Vocabulary Levels test score, and the mean naturalness ratings in
the analyses as covariates. Initial models were fitted with a three-way interaction (preliminary
model: Group x Phrase Type x Prior Knowledge; main model: Learning Condition x Phrase
f
oo
The gapfill accuracy data (correct=1; incorrect=0) were analysed using a generalized
r
linear mixed-effects model (Jaeger, 2008), using the R package, lme4 (Bates et al., 2015).
-p
The odds ratios (OR) and standardised effect sizes (Chinn, 2000) are reported for this
re
analysis. The odds are defined as the probability of an event occurring divided by the
lP
probability of it not occurring (Field, 2013, p. 767). When the odds are more than 1.0, a
na
“success” (in this case, a correct answer) is more likely than a “failure” (an incorrect answer)
The most complex models (with three-way interactions and all predictors of interest)
Jo
were fitted to the data first; the final model was identified using backwards stepwise model
simplification and the likelihood ratio test for model comparisons (Baayen et al., 2008).
Participants and items were included in all models as crossed random effects. The final
models contained random slopes supported by the data (i.e. parsimonious mixed models
based on Matuschek et al., 2017). (For the final model formulas, see Appendix G.) The
emmeans package (Lenth, 2019) was used to conduct the post-hoc analyses, and Bonferroni p
value correction was used for multiple comparisons. The categorical predictors with two
levels were contrast coded as -.05 and 0.5, in order to interpret their main effect in the
models.
15
5. ANALYSIS AND RESULTS
5.1. Analysis
Ten collocations in total were excluded from the analysis. Five collocations were initially
excluded for the following reasons. One collocation (make arrangements) was reported as
known by 90% of all participants in the prior knowledge test and 56% of the No-treatment
group participants provided correct responses to this collocation in the gapfill post-test. We
removed four items (shed tears, push the limits, by profession and under protest) due to
f
experimenter errors in the treatment materials. To ensure comparability of lexical and
oo
grammatical collocations, we had to also exclude the five corresponding collocations of the
r
other type (e.g., by arrangement, through tears), leaving a total of 38 of the original 48
-p
collocations for analysis.
re
lP
5.2. Results
Descriptive statistics for the gapfill response accuracy are presented in Table 1. Section a) of
na
the table shows response accuracy by Group and Phrase type (Preliminary analysis).
ur
Treatment participants had a mean accuracy rate of 0.21 of the 38 collocations (8.0
Jo
collocations) compared with 0.05 (1.9 collocations) for the No-treatment group. Section b)
shows response accuracy for the Treatment group by Learning Condition and Phrase Type
(Main analysis). For the prep+noun collocations, the mean accuracy rate in the TE condition
was noticeably higher than that in the no-TE condition (i.e., 6.1 vs. 3.2 out of 19
collocations). For the verb+noun collocations, however, the mean accuracy rate in the TE
condition was only slightly higher than in the no-TE condition (3.6 vs. 3.0 out of 19
collocations). The reliability of the gap-fill task was acceptable: Cronbach's Alpha = 0.88.
16
Table 1. Descriptives: Accuracy of responses in the gapfill task
f
oo
(+)bold 0.32 0.47 0.19 0.39 0.25 0.43
In the preliminary analysis (see Table 2), a significant difference in the accuracy of
r
-p
gapfill responses between the Treatment and No-treatment groups was observed. The odds of
re
obtaining an answer that matched the target collocation for the Treatment group were 8 times
lP
the odds of obtaining a matched answer in the No-treatment group, with a large standardised
na
mean difference (SMD) effect size of d=1.15. The absence of an interaction between Group
and Phrase Type suggests that there was a large effect of the contextual learning treatment on
ur
the accuracy of gapfill responses irrespective of the collocation type. There were also
Jo
significant positive main effects of the self-reported prior knowledge, vocabulary knowledge
(VLT) and phrase naturalness ratings on the accuracy of gapfill responses in this analysis.
Table 2. Preliminary analysis. Gapfill response accuracy analysis by group: Fixed effects
Parameter β SE z p Odds 95% CI 95% CI d
Ratios Lower Upper
(Intercept) -4.12 0.37 -11.27 < 2.0E-16
Group=Treatment 2.09 0.36 5.75 8.8E-09 8.05 3.96 16.39 1.15
PhrType=verb+noun -0.56 0.12 -4.65 3.4E-06 0.57 0.45 0.72 0.31
PrevKnow=yes 0.95 0.13 7.20 5.9E-13 2.58 1.99 3.34 0.52
VLT.c 0.05 0.01 5.92 3.2E-09 1.05 1.04 1.07 0.03
lg.m.rating.c -0.84 0.21 -3.90 9.6E-05 0.43 0.29 0.66 0.46
rating.sd.c 2.52 0.60 4.21 2.6E-05 12.47 3.85 40.39 1.39
17
Note. VLT.c = Vocabulary Levels Test score (cumulative), centred; lg.m.rating.c = mean
phrase naturalness rating, log-transforms, centred; rating.c = phrase naturalness rating, log-
transforms, centred. Intercept levels: Group=No-treatment group, Phrase Type = prep+noun,
Prior Knowledge = no.
In the main analysis of the Treatment group response accuracy (see Table 3), there was
a significant interaction (z=3.25 p<.01 d=.45) between the Learning Condition and Phrase
Type, such that bolding significantly improved the accuracy of responses for the prep+noun
collocations (by about 13%) but not for the verb+noun collocations (Figure 1). These results
f
suggest that TE was particularly beneficial in the contextual learning of prep+noun
oo
collocations. The main effect of Learning Condition was medium (d=0.50), with the odds of
r
-p
correct recall being about 2.5 times greater in the (+)bold than (-)bold condition. We also
re
observed significant main effects of self-reported prior knowledge (z=5.58 p<.001),
vocabulary knowledge (VLT) (z=5.57 p<.001), the mean phrase naturalness rating (z=-3.27
lP
p<.01) and the standard deviation of the phrase naturalness rating (z=3.74 p<.001). The
na
presence of self-reported prior knowledge was associated with more accurate gapfill
ur
responses, as were higher VLT scores, higher ratings of naturalness and greater variability in
Jo
the ratings. These results suggest that that L2 collocational learning is dependent on a number
Table 3. Main analysis. Gapfill response accuracy analysis for the Treatment group: Fixed
effects
18
Note. Intercept levels: Learning Condition = (-)bold, Phrase Type = prep+noun, Prior
Knowledge = no.
Figure 1. Gapfill response accuracy for the Treatment group: Results by Learning Condition
f
r oo
-p
re
lP
5. DISCUSSION
na
5.1. Discussion
ur
Six contextual exposures to L2 collocations over two days resulted in significant learning.
Jo
The knowledge of the target collocations measured by the gapfill test was significantly
greater for the Treatment group (21% accuracy), compared with the No-treatment group (5%
accuracy), across both collocation types. The gapfill accuracy scores for the bolding and no-
bolding conditions were 25% and 17% respectively in the Treatment group. These findings
suggest that repeated exposure to L2 collocations in written input can lead to their learning.
verb+noun and prep+noun collocations, the participants developed explicit knowledge of the
collocations, which they could retrieve when the collocations were presented in context.
These results corroborate the findings of the previous studies which report gains through
19
repetition of collocations in written texts (e.g., Durrant & Schmitt, 2010; Sonbul & Schmitt,
Let us now consider the effect of the learning conditions (i.e., TE = (+)bold and no-TE =
accuracy in recalling the first word of the prep+noun collocations than no-TE (by about
15%). The odds of recalling prep+noun collocations were 2.5 times higher in the TE than the
no-TE learning condition, but bolding did not affect the recall of verb+noun collocations.
Bolding likely made prep+noun collocations more perceptually salient to the learners,
f
oo
drawing attention to them and possibly helping the learners detect a formal and semantic link
r
between the prepositions and the nouns. Prepositions may be often ignored by readers for
-p
several reasons. First, they tend to contain fewer letters than verbs. Second, prepositions tend
re
to be more frequent than verbs in natural language, thus being seen as very well known
lP
(Boers et al., 2014). Third, prepositions may seem very familiar because they are highly
na
polysemous. Since readers often do not notice prepositions, typographic enhancement may
have a stronger effect in increasing the salience of prepositions than it does the salience of
ur
verbs. It appears that, once the prep+noun collocations were noticed, it was more likely they
Jo
would be recalled.
Unexpectedly, no difference was found between the accuracy of the cued recall of
verb+noun collocations encountered in the TE and no-TE conditions. The lack of TE effect in
to the present study. One possible reason is that bolding did not make the verb+noun
combinations more salient to the learners because verbs are already sufficiently salient in the
input compared with prepositions. Another possible reason is the type of collocations in the
present study compared with the types in other studies in which TE produced greater learning
of lexical collocations than no-TE (e.g, Sonbul & Schmitt, 2013; Toomer & Elgort, 2019).
20
Our study participants reported relatively high familiarity with the target collocations, and the
control group had a 5.0% accuracy rate for the collocations in the recall (gapfill test). This
suggests greater prior knowledge than the minimal knowledge of technical lexical
collocations reported by Toomer and Elgort (2019), with their control group accuracy rate of
1.62% and low self-reported familiarity ratings across all participants. The assumption that
collocations in the present study were known could have resulted in less effort by the learners
It is commonly argued that more attention to and noticing of particular linguistic items
f
oo
leads to better learning of those items (e.g., Schmidt, 1994). L2 learners may fail to notice
r
common MWUs because the constituent words are frequent and because many of the MWUs
-p
appear familiar and lack novelty (Bishop, 2004; Boers, Lindstromberg, & Eyckmans, 2014).
re
As a result, techniques which promote noticing of MWUs, such as TE, have been found to
lP
lead to gains in their knowledge (Boers et al., 2016; Peters, 2012; Szudarski & Carter, 2016).
na
Our findings support this: repetition of the L2 collocations in the input, with and without TE,
facilitated the development of offline knowledge in our experiment. Also, compared with no-
ur
TE, TE led to a larger amount of learning for the less salient type of collocations.
Jo
The findings of this experiment with respect to the learnability of lexical and
grammatical collocations (RQ3) overall support the claims of Howarth (1998b) and
Ackerman and Chen (2013) that (1) lexical collocations are more difficult to master than
grammatical collocations and (2) grammatical collocations are more easily internalised
because they are more predictable (Ackerman & Chen, 2013). Our findings also partially
corroborate Gitsaki’s (1999) finding of higher scores in a gap-fill task for previously
encountered prep+N than V+N collocations. Finally, prep+N collocations were more
amenable than V+N collocations to gains resulting from attentional manipulation (when
21
summary, grammatical collocations encountered multiple times in reading were more likely
5.2. Limitations
There are several limitations of this study that we need to acknowledge. The transparency of
the collocations was not formally normed (although we aimed to choose semi-transparent
collocations using selection criteria). Nevertheless, learning may have been affected by
f
with some (but not all) of the target collocations, the effect of one or both learning conditions
oo
may have been reduced or amplified by this prior knowledge; to mitigate this, we attempted
r
-p
to partial out the effect of Prior Knowledge by including it as a predictor in the analyses.
re
Finally, the measurement of Prior Knowledge may not have always been accurate because it
lP
was self-reported.
knowledge. Bolding appears to be particularly beneficial for less salient collocations that are
partially known, such as grammatical collocations containing short function words and
collocations comprising high-frequency familiar words; it may not be needed for more salient
repetition without TE. Toomer and Elgort (2019) found that not bolding lexical collocations
produced greater processing fluency of the collocations than bolding but less collocational
22
knowledge. Future experiments could also test a mixed approach to the use of TE in the
contextual learning of collocations, with TE used only on the first occurrence in the text but
not on subsequent occurrences. This is consistent with Ellis’s (2007) claim that focused
attention may only need to be applied once to a non-salient linguistic item (since explicit
knowledge can be gained very fast). After the item has been noticed the first time, “mere use
in processing for meaning is enough for implicit tallying” (Ellis, 2007, p. 30). Such an
approach might facilitate a more evenly balanced development of explicit knowledge of the
target MWUs and their online processing. The initial-only TE would also reflect the approach
f
oo
often used in university textbooks when introducing technical terms.
r
-p
re
1
The results reported in the article are part of a larger multi-experiment project which
lP
2
The VLT scores of the Treatment and No-treatment groups were not statistically different
ur
(t=-0.04 p=.97). Because participants assigned to the No-treatment group were drawn from
Jo
the same student population as those in the Treatment groups and did not differ in terms of
their L2 vocabulary knowledge, we were satisfied that we could use the No-treatment
participants’ knowledge and processing of the target collocation as a baseline, with which the
which included the studied verb+noun and prep+noun collocations (e.g., The government will
create work for many jobless people) and the matching control phrases (The government will
develop work for many jobless people). Participants were instructed to read the sentences for
meaning and answer yes/no comprehension questions (no words from the target collocations
were used in these questions). This task was administered to estimate online processing of the
23
collocations. The results of the self-paced sentence-reading task are not presented in the
f
r oo
-p
re
lP
na
ur
Jo
24
References
Ackermann, K., & Chen, Y-H. (2013). Developing the Academic Collocation List (ACL)—A
corpus-driven and expert-judged approach. Journal for Academic Purposes, 12, 235-
247.
Agresti, A. (2007). An introduction to categorical data analysis (2nd ed.). Hoboken: New
Altenberg, B., & Granger, S. (2001). The grammatical and lexical patterning of MAKE in
f
native and non-native student writing. Applied Linguistics, 22, 173-195.
oo
Baayen, R. H. (2008). Analyzing linguistic data. A practical introduction to statistics using R,
r
Cambridge: Cambridge University Press.-p
re
Bahns, J., & Eldaw, M. (1993). Should we teach EFL students collocations? System, 21, 101-
114.
lP
Bates, D., Maechler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects
na
Benson, M., Benson, E., & Ilson, R. (1997). The BBI dictionary of English word
Jo
Biber, D., Johansson, S., Leech, G., Conrad, S., & Leech, G. (1999). Longman grammar of
spoken and written English. Edinburgh Gate, Harlow: Pearson Education Limited.
25
Boers, F., Demecheleer, M., He, L., Deconinck, J., Stengers, H., & Eyckmans, J. (2016).
Boers, F., Lindstromberg, S., & Eyckmans, J. (2014). Some explanations for the slow
41-62.
Boers, F., Eyckmans, J., Kappel, J., Stengers, H., & Demecheleer, M. (2006). Formulaic
sequences and perceived oral proficiency: putting a Lexical Approach to the test.
f
oo
Language Teaching Research, 10, 1-17.
r
Bramki, D., & Williams, R. (1984). Lexical familiarization in economics text, and its
-p
pedagogic implications in reading comprehension. Reading in a Foreign Language, 2,
re
169-181.
lP
Catalán, R. (1996). Frequency and variability in errors in the use of English prepositions.
na
Chinn, S. (2000). A simple method for converting an odds ratio to effect size for use in meta-
ur
Chodorow, M., Gamon, M., & Tetreault, J. (2010). The utility of article and preposition error
correction systems for English language learners: Feedback and assessment. Language
Choi, S. (2017). Processing and learning of enhanced English collocations: An eye movement
Cobb, T. (n.d.). Compleat Web VP v.2.5, Compleat Lexical Tutor. Retrieved from
https://www.lextutor.ca/vp/comp/
Cosme, C., & Gilquin, G. (2008). Free and bound prepositions in a contrastive perspective:
The case of with and avec. In S. Granger & F. Meunier (Eds.), Phraseology: an
26
interdisciplinary perspective (pp. 259-272). Amsterdam/ Philadelphia: John Benjamins
Publishing Company.
Davies, M. (2008-) The Corpus of Contemporary American English (COCA): 560 million
Durrant, P. (2009). Investigating the viability of a collocation list for students of English for
Durrant, P., & Schmitt, N. (2010). Adult learners’ retention of collocations from exposure.
f
oo
Ellis, N. (2002). Frequency effects in language processing: A review with implications for
r
theories of implicit and explicit language acquisition, Studies in Second Language
education: Studies in honour of Rod Ellis (pp. 17-34). Oxford: Oxford University Press.
Elgort, I., Brysbaert, M., Stevens, M., & Van Assche, E. (2018). Contextual word learning
ur
Elgort, I., & Warren, P. (2014). L2 vocabulary learning from reading: Explicit and tacit
lexical knowledge and the role of learner and item variables. Language Learning, 64,
365–414.
Field, A. (2013). Discovering statistics using IBM SPSS statistics (4th Edition). London: Sage.
Gass, S., Svetics, I., & Lemelin, S. (2003). Differential effects of attention. Language
27
Gyllstad, H. (2014). Grammatical collocation. In C. Chappelle (Ed.). The encyclopaedia of
Hemchua, S., & Schmitt, N. (2006). An analysis of lexical errors in the English compositions
a progress report. In Bardel, C., Lindqvist, C., & Laufer, B. EUROSLA Monographs
Howarth, P. (1998a). Phraseology and second language proficiency. Applied Linguistics, 19,
f
oo
24-44.
r
Howarth, P. (1998b). The phraseology of learners’ academic writing. In A. Cowie (Ed.),
-p
Phraseology: theory, analysis and applications (pp. 161-186). Oxford: Oxford
re
University Press.
lP
Hu, M., & Nation, P. (2000). Unknown vocabulary density and reading comprehension.
na
Press.
Hulstijn, J., Hollander, M., & Greidanus, T. (1996). Incidental vocabulary learning by
advanced foreign language students: the influence of marginal glosses, dictionary use,
and reoccurrence of unknown words. The Modern Language Journal, 80, 327-339.
Huo, Y. (2014). Analyzing collocation errors in EFL Chinese learners’ writings based on
28
Jaeger, T.F. (2008). Categorical data analysis: Away from ANOVAs (transformation or not)
and towards logit mixed models. Journal of Memory and Language, 59, 434–46.
Kroll, N., & Kellicutt, M. (1972). Short-term recall as a function of covert rehearsal and of
intervening task. Journal of Verbal Learning and Verbal Behaviour, 11, 196-204.
Laufer, B., & Waldman, T. (2011). Verb-noun collocations in second language writing: a
f
oo
Lenth, R. (2019). emmeans: Estimated Marginal Means, aka Least-Squares Means. R
r
package version 1.4.1.
-p
Matuschek, H., Kliegl, R., Vasishth, S., Baayen, H., & Bates, D. (2017). Balancing Type I
re
error and power in linear mixed models. Journal of Memory and Language, 94, 305–
lP
315.
na
Meara, P., & Buxton, B. (1987). An alternative to multiple choice vocabulary tests. Language
Testing, 4, 142-154.
ur
https://www.victoria.ac.nz/lals/about/staff/publications/paul-nation/Information-on-the-
BNC_COCA-word-family-lists.pdf
Nesselhauf, N. (2003). The use of collocations by advanced learners of English and some
Obermeier, A., & Elgort, I. (2021). Deliberate and contextual learning of L2 idioms: The
29
Paquot, M., & Granger, S. (2012). Formulaic language in learner corpora. Annual Review of
Paribakht, T. & Wesche, M. (1997). Vocabulary enhancement activities and reading for
University Press.
f
oo
Peters, E. (2012). Learning German formulaic sequences: The effect of two attention-drawing
r
techniques. Language Learning Journal, 40, 65-79.
-p
Rieder, A. 2003. Implicit and explicit learning incidental vocabulary acquisition. Vienna
re
English Working Papers.
lP
Schmitt, N., Jiang, X., & Grabe, W. (2011). The percentage of words known in a text and
Schmitt, N., Schmitt, D., & Clapham, C. (2001). Developing and exploring the behaviour of
two new versions of the Vocabulary Levels Test. Language Testing, 18, 55-88.
Sharwood Smith, M. (1991). Speaking to many minds: On the relevance of different types of
30
Sicherl, E. (2004). On the content of prepositions in prepositional collocations. Studies in the
Simpson-Vlach, R., & Ellis. N. (2010). An academic formulas list: New methods in
Sonbul, S., & Schmitt, N. (2013). Explicit and implicit lexical knowledge acquisition of
f
oo
collocations under different input conditions. Language Learning, 63, 121-159.
r
Szudarski, P. & Carter, R. (2016). The role of input flood and input enhancement in EFL
-p
learners’ acquisition of collocations. International Journal of Applied Linguistics,
re
26(2).
lP
Tomlin, R., & Villa, V. (1994). Attention in cognitive science and second language
na
Toomer, M., & Elgort, I. (2019). The development of implicit and explicit knowledge of
ur
Tyler, A., Mueller, C., & Ho, V. (2011). Applying cognitive linguistics to learning the
semantics of English to, for and at: An experimental investigation. Vigo International
46-65.
Webb, S., Newton, J., & Chang, A. (2013). Incidental learning of collocation. Language
31
We, the authors, declare that we have no potential competing interests.
This manuscript has not been previously published and is not under consideration for
publication elsewhere. Its publication is approved by all authors. If published, it will not
be published elsewhere in the same form without the consent of the copyright holder.
Mark Toomer
Irina Elgort
Averil Coxhead
f
r oo
-p
re
lP
na
ur
Jo