0% found this document useful (0 votes)

11 views39 pages

Explicit and Implicit Lexical Knowledge Acquisition of Collocations Under Different Input Condictions (2013)

Uploaded by

Andrés Cerro

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

11 views39 pages

Explicit and Implicit Lexical Knowledge Acquisition of Collocations Under Different Input Condictions (2013)

Uploaded by

Andrés Cerro

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 39

Language Learning ISSN 0023-8333

Explicit and Implicit Lexical Knowledge:

Acquisition of Collocations Under Different
Input Conditions
Suhad Sonbul
Umm Al-Qura University and University of Nottingham

Norbert Schmitt
University of Nottingham

To date, there has been little empirical research exploring the relationship between
implicit and explicit lexical knowledge (of collocations). As a first step in addressing
this gap, two laboratory experiments were conducted that evaluate different conditions
(enriched, enhanced, and decontextualized) under which both adult native speakers
(Experiment 1) and advanced nonnative speakers of English (Experiment 2) acquire
collocations. Three different tests of collocational knowledge were used to assess gains
after treatment: two traditional explicit tests (form recall and form recognition) and an
innovative implicit test (priming). Results from mixed-effects modeling showed that all
conditions led to significant long-term gains in explicit form recall and recognition both
for natives and nonnatives, while no condition facilitated implicit collocational priming
effects either for natives or nonnatives.
Keywords vocabulary; collocations; explicit knowledge; implicit knowledge; priming;
mixed-effects modeling

Introduction
The relationship between explicit and implicit knowledge has long been the
focus of research in psychology (for an overview, see Squire, 2004), and a
similar interest has also emerged in applied linguistics and second language
acquisition (SLA) circles (N. Ellis, 2008). To date, most SLA research into
the explicit/implicit linguistic knowledge distinction has focused on second

Correspondence concerning this article should be addressed to Suhad Sonbul, P.O. Box 10424,
Makkah, Saudi Arabia. Internet: [email protected]

Language Learning 63:1, March 2013, pp. 121–159 121

C 2012 Language Learning Research Club, University of Michigan

DOI: 10.1111/j.1467-9922.2012.00730.x
14679922, 2013, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-9922.2012.00730.x by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Sonbul and Schmitt Explicit and Implicit Lexical Knowledge

language (L2) grammar acquisition (de Graaff, 1997; R. Ellis, 2005), although
the nature of the relationship between explicit and implicit grammatical knowl-
edge remains controversial. In contrast, there has been very little consider-
ation of the distinction’s implications for lexical knowledge. This might be
due to the traditional dictionary metaphor which views the lexicon as noth-
ing more than a list of forms and meanings that can be associated fairly
easily through rote learning. However, the view that vocabulary knowledge
is a much more complex construct is now gaining acceptance, with several
lexical scholars claiming that it entails a number of different dimensions in-
volving both explicit and implicit mental representations (Aitchison, 2003; N.
Ellis, 1994; Nation, 2001; Schmitt, 2010). Still, the exact nature and interre-
lationship of explicit and implicit vocabulary knowledge has only begun to be
investigated.
This study addresses this emerging lexical area by reporting on a two-
experiment study (one with natives and one with nonnatives) exploring how
well various direct and indirect input conditions promote both explicit and
implicit knowledge of one aspect of vocabulary knowledge: collocations.

Explicit Versus Implicit Knowledge: Grammar Research

Although researchers disagree on the precise nature of explicit/implicit knowl-
edge (Bialystok, 1981; Hulstijn, 2007; Paradis, 1994), the central difference
lies in the presence or absence, respectively, of awareness. These two terms are
defined here as follows:
Explicit knowledge consists of the facts that speakers of a language have
learned. These facts are often not clearly understood and may be in
conflict with each other . . . . Explicit knowledge is held consciously, is
learnable and verbalisable, and is typically accessed through controlled
processing when learners experience some kind of linguistic difficulty in
using the L2 . . . In contrast, implicit knowledge is procedural, is held
unconsciously, and can only be verbalized if it is made explicit. It is
accessed rapidly and easily and thus is available for use in rapid, fluent
communication. (R. Ellis, 2006, p. 95, emphasis in original)

In essence, explicit knowledge seems related to declarative knowledge of

language, while implicit knowledge is most closely related to speed and ease of
access in the use of language. R. Ellis demonstrated that, while both dimensions
relate to linguistic knowledge, they are separate constructs. In an influential
validation study (R. Ellis, 2005), he found that measures of explicit knowledge

Language Learning 63:1, March 2013, pp. 121–159 122

14679922, 2013, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-9922.2012.00730.x by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Sonbul and Schmitt Explicit and Implicit Lexical Knowledge

(metalinguistic knowledge test and untimed grammaticality judgment task)

and measures of implicit knowledge (oral narration test, timed grammaticality
judgment task, and oral imitation test) loaded onto two separate factors.
Most SLA researchers agree on the central role played by implicit knowl-
edge in achieving proficiency (N. Ellis, 1993; R. Ellis, 2006). However, the
key question is the relationship between explicit and implicit knowledge. This
issue emerged toward the beginning of the 1980s when the interface debate
was at the heart of L2 grammar instruction research. The interface debate was
based on the conviction that developing implicit knowledge is the ultimate
goal of language learning, and, thus, that direct instruction is only worthwhile
if it can lead to (or at least aid) the growth of implicit knowledge (R. Ellis,
2007). The main question was, thus, whether direct instruction can or cannot
influence the development of implicit knowledge. The noninterface position
held that direct instruction is ineffective (Krashen, 1981; Paradis, 1994). The
strong interface position maintained that explicit knowledge can be converted
into implicit knowledge through practice (DeKeyser, 1994; B. McLaughlin,
Rossman, & McLeod, 1983). The weak interface position is somewhere in-
between, holding that explicit knowledge can feed into implicit knowledge
either directly (if the learner is developmentally ready) or indirectly (through
facilitating noticing or through helping learners notice gaps in their own output)
(R. Ellis, 1994). In other words, although explicit and implicit knowledge are
distinct and dissociated, they do interact substantially (N. Ellis, 2008; N. Ellis
& Laporte, 1997).
Since the 1980s, many studies have been conducted to compare the effec-
tiveness of various direct and indirect instructional conditions on the develop-
ment of explicit and implicit grammatical knowledge (for reviews, see R. Ellis,
2002; Long, 1983). In a meta-analysis conducted at the turn of the millennium,
Norris and Ortega (2000) reviewed 49 studies carried out between 1980 and
1998. The following conclusions were reached: (1) instruction results in large
gains, (2) these gains are durable over time, and (3) direct instruction is more
effective than indirect instruction. In spite of these important findings, Norris
and Ortega point out a number of caveats, the most important being that “test-
ing of learning outcomes usually favors explicit treatments by asking learners
to engage in explicit memory tasks and/or in discrete de-contextualized L2
use” (p. 501) (see also Doughty, 2003, for a similar claim). However, Mackey
and Goo (2007) observe that more studies are now applying measures of im-
plicit grammatical knowledge. This is evident in Spada and Tomita’s (2010)
recent metaanalysis collapsing results of 41 studies that employed explicit (di-
rect grammar instruction) and implicit (indirect manipulation of input with

123 Language Learning 63:1, March 2013, pp. 121–159

no grammar instruction, e.g., enriched exposure) instructional techniques and

utilized measures of either or both types of knowledge (explicit: controlled con-
structed tasks, e.g., grammaticality judgment tasks and multiple-choice items;
implicit: free constructed tasks, e.g., free writing and picture description tasks).
Results of the meta-analysis showed that direct instruction was more effective
than indirect instruction not only in developing explicit knowledge but also in
enhancing implicit knowledge. It should be noted, however, that not all free
constructed tasks are necessarily tapping into implicit knowledge (Spada &
Tomita, 2010; see also White & Ranta, 2002). This is supported by results of a
recent study (Andringa, de Glopper, & Hacquebord, 2011) where a correlation
between an explicit measure (grammaticality judgment task) and an implicit
one (free written response task) was interpreted as evidence that the two are
actually tapping into the same construct. Spada and Tomita (2010) concluded
that there is a need for more validation studies to develop new measures and
evaluate what they are assessing.
In fact, modern psycholinguistic techniques can be useful in exploring the
explicit–implicit distinction. For example, Osterhout, McLaughlin, Pitkänen,
Frenck-Mestre, and Molinaro (2006) reported on a study conducted to gauge
the development of English learners of L2 French in the rule of verbal person
agreement over a period of 8 months. Results showed that, after 80 hours of
instruction, participants performed very poorly in an explicit grammaticality
judgment task, but showed Event-Related Potential (ERP) behavior (P600 com-
ponent elicited for violations) that was not qualitatively different from natives.
Similarly, Tokowicz and MacWhinney (2005) showed that performance of En-
glish learners of L2 Spanish in an explicit measure (grammaticality judgment
task) was distinct from their ERP behavior for gender agreement violations.
Although their explicit performance was near chance, they showed a clear
P600 effect for these violations. Results from these two studies seem to suggest
that, when using a highly sensitive psycholinguistic measure, implicit knowl-
edge appears to develop before explicit knowledge, even after short instruction
periods.
In sum, research has not yet clearly described the interrelationship between
explicit and implicit grammatical knowledge and how it is acquired. Overall, di-
rect instruction seems more effective than indirect instruction in the promotion
of explicit knowledge. It may be more effective in promoting implicit knowl-
edge as well, but this depends on what measures one accepts as tapping into
implicit knowledge. If using psycholinguistic techniques that are undisputedly
implicit in nature (e.g., ERP), then there is some evidence that the development
of implicit knowledge may actually precede explicit knowledge.

Language Learning 63:1, March 2013, pp. 121–159 124

Explicit and Implicit Lexical Knowledge

When considering the explicit–implicit knowledge distinction, there are reasons
to believe that vocabulary is different from grammar. The main difference lies
in what Hulstijn and de Graaff (1994, pp. 104–105) term rule learning of
nonlexical structures versus item learning of lexical structures. In the area
of grammar, this distinction is best exemplified in learning the regular form
of the past tense play–played as opposed to the irregular form become–became,
respectively. Along the same lines, it can be claimed that learning vocabulary
is mostly item learning.
This item view has led to the assumption that vocabulary knowledge is
declarative in nature and can never be implicit. Ullman (2001), for example,
developed his declarative/procedural model of linguistic knowledge assuming,
based on neurological evidence, that vocabulary knowledge is declarative while
grammatical knowledge is procedural. Similarly, Hulstijn (2007) claims that
vocabulary knowledge is symbolic and, thus, explicit in nature. These views
seem to treat the lexicon as a simple dictionary containing only word forms
connected to their meanings, which has led to the distinction between explicit
and implicit lexical knowledge being largely ignored in research.
However, vocabulary knowledge is now viewed as a complex construct
encompassing various aspects. Nation (1990, 2001) categorized a number of
word knowledge aspects including form, meaning, grammatical characteristics,
collocation, and several others. N. Ellis (1994), one of the first scholars to
show an interest in the explicit–implicit contrast for lexis, considered Nation’s
aspects and distinguished two groups of components: (a) components related
to the connection between the form of a word and its meaning and the various
semantic relations between words which can be best learned explicitly and (b)
components related to the form of the word and its receptive (input)/productive
(output) aspects of usage which are best learned implicitly (but can also be
learned explicitly).
N. Ellis’s (1994) distinction does not stop at learning processes, but can also
be applied to knowledge representations. Certain (meaning-related) aspects of
word knowledge are inherently explicit whereas others (formal and usage-
related aspects) might be either explicit or implicit depending on the presence
or absence of awareness. R. Ellis (2004, p. 242) makes a similar claim:

[W]hereas knowledge of the form of a word and of word collocations is

largely implicit . . . Word meanings . . . probably constitute the largest
single area in a learner’s explicit knowledge. In addition, of course,
individuals may develop a conscious awareness of the form of at least

125 Language Learning 63:1, March 2013, pp. 121–159

some words—problematic spellings or the differences between American

and British spellings, for example—and of some collocations.

This more sophisticated perspective highlights the value of the explicit–

implicit distinction in understanding vocabulary knowledge. Unfortunately,
the lack of conventional measures of implicit lexical knowledge has led to most
previous lexical research being explicit in nature, assessing breadth, depth, and
organization of vocabulary knowledge (see Read, 2000, for an overview). Im-
plicit measures used in psychological research are just beginning to be used in
investigating vocabulary acquisition issues (e.g., Elgort, 2011). One example is
J. McLaughlin, Osterhout, and Kim (2004), who studied knowledge of the word
form. They employed two measures of vocabulary form (a lexical decision task
tapping explicit knowledge and an ERP measure gauging implicit knowledge)
to observe the development of a group of English learners of French. Lexi-
cal decision and ERP behavior were obtained longitudinally in three sessions
throughout a semester. Although French learners did not show much improve-
ment in explicit performance, their ERP data reflected sensitivity to lexical
status and semantic relatedness (similar to those found for native speakers in
earlier research) after only 14 hours of instruction. In interpreting these re-
sults, the researchers claim that ERP is a more sensitive measure of linguistic
improvement than explicit tests which might underestimate gains in knowledge.

Collocational Knowledge: Acquisition and Measurement

This section focuses on the other aspect of vocabulary knowledge besides word
form which R. Ellis (2004) considers largely implicit in nature: collocational
knowledge. Most research into collocations has been descriptive in nature, often
analyzing either native speaker corpora or L2 output to see what phraseology
has been produced (see, e.g., Siyanova & Schmitt, 2008). However, there has
been little research into how knowledge about collocations has been acquired.
Durrant and Schmitt (2010) conducted an experiment in a controlled (labo-
ratory) English-as-a-second-language (ESL) setting with advanced nonnatives.
Participants were instructed to read aloud a number of sentences containing
made-up, low-frequency, adjective-noun collocations (to ensure no preknowl-
edge). All participants were tested with a timed cued recall (naming) task with
instructions to remember combinations from the study phase. Results showed
that L2 learners do retain memory of collocations they meet in input even
after one or two exposures and even when these are not enhanced in any way.
However, it should be noted that this study (and others conducted in the class-
room context, e.g., Laufer & Girsai, 2008; Webb & Kagimoto, 2009) did not

Language Learning 63:1, March 2013, pp. 121–159 126

employ any measure of implicit collocational knowledge. Although Durrant and

Schmitt’s test was timed and was thus tapping intuitions, we consider it a test
of explicit knowledge in the sense that it elicited the target form (collocations)
explicitly with a very clear reference to the study phase.
One early study using the priming paradigm did look at implicit knowledge
of collocations. McKoon and Ratcliff (1992, Experiment 3) showed that priming
can occur not only for made-up pairs but also for real-world collocations. In
their study, they extracted noun-noun pairs (e.g., hospital-baby) from a small
corpus. These pairs co-occurred in a six-word window higher than chance (as
measured by high Mutual Information scores) but were not related according to
association norms. Results of a priming/lexical decision test showed that there
was a significant faciliation effect for these collocation pairs over control ones.
This study seems to have opended the door for research into what is now called
collocational priming.
This notion of collocational priming was first hypothesized by Hoey (1991,
2005), who claimed that collocations are acquired incrementally through ex-
posure to text. Language users store concordances of every word encountered
and its context, so each contextual encounter with a word might either create
a new collocation or reinforce and modify an old one. According to Hoey
(2005), there is little difference in the collocational priming process between
adult natives and nonnatives; rather, the key issue is the amount and type of
input. For him, natives generally thrive on the indirect input because they are
likely to be surrounded by a massive amount of linguistic evidence, while non-
natives generally do better with direct exposure because they are not usually
surrounded by such evidence. If there is enough exposure, then collocations
will be learned inductively for both groups of learners, but if there is not enough
exposure, then other more explicit techniques should be used to help native and
nonnative learners. However, different activities raising learners’ attention to
collocational restrictions (e.g., exercises, illustrations, exposure to rich input)
are likely to have differential effects on collocational priming.
Three recent studies examined this notion of collocational priming with
native speakers of English using lexical decision tasks (Durrant & Doherty,
2010, Experiment 1; N. Ellis, Frey, & Jalkanen, 2009; Wolter & Gyllstad,
2011). Results of these studies showed that collocational priming is indeed
possible and that native speakers are sensitive to collocational frequency. More
specifically, a facilitation effect was found for high-frequency collocates over
low-frequency collocates and control pairs. There is some question, however,
of whether these studies were truly tapping into implicit knoweldge, as the
priming techniques employed were not automatic in nature. Automatic priming

127 Language Learning 63:1, March 2013, pp. 121–159

is contrasted with strategic priming in that the former operates rapidly without
participants’ conscious intention (implicit knowledge) while the latter requires
intention and reflects slower processes (explicit knowledge) (Dagenbach, Horst,
& Carr, 1990). After discussing different variables determining the nature of
priming processes (stimulus onset asynchrony [SOA], the nature of task, the
type of semantic relation, relatedness proportion [RP] measured as the number
of related word-prime trials divided by all word-target trials, and nonword ratio
[NR] measured as the ratio of nonword trials to all unrelated trials), McNamara
(2005) concludes:
An investigator interested in the automatic component of semantic
priming would be well advised to use a short SOA (e.g., 200 ms or less), a
low RP (e.g., <0.20?), and an NR of 0.50. Under these conditions, the
task and the type of semantic relation do not seem to matter much. (p. 72)
Unfortunately, none of the three experiments mentioned above met these re-
quirements. Thus, although they did show collocational priming, how this prim-
ing relates to implicit knowledge (vs. strategic explicit knowledge) is not clear.

Present Study: Aim, Operationalization of Terms, and Research

Questions
To our knowledge, no research study to date has evaluated various learning
conditions under which implicit knowledge along with explicit knowledge
of collocations might be developed. The present study employs an estab-
lished psycholinguistic measure in addition to traditional tests to investigate
implicit/explicit lexical knowledge. Two experiments (with natives and non-
natives, respectively) are conducted to investigate the effectiveness of three
typical learning conditions (enriched, enhanced, and decontextualized) under
which explicit (form recall and recognition) and implicit (automatic priming)
collocational knowledge might be acquired. The first two learning conditions
are operationalized (following Reinders & R. Ellis, 2009) from an input per-
spective, as follows:
Enriched input: “input that has been seeded with the target structure so
that learners are exposed to high frequency over a period of time.” (p. 282)
Enhanced input: “input where the target feature has been emphasized in
some way—glossing, bolding or underlining.” (p. 283)
The decontextualized input condition, in turn, is defined as input where the
target feature is presented out of context with direct instruction to commit it to
memory. In accordance with Spada and Tomita’s (2010) definitions (see earlier

Language Learning 63:1, March 2013, pp. 121–159 128

discussion), the first condition is considered an indirect instructional condition

while the second and third are considered direct teaching conditions.
The term collocation is operationalized here as a technical two-word pair
that often co-occurs in medical contexts as evidenced in medical dictionaries.
It should be noted here that the present study investigates initial stages of
collocational learning, whereby a link is built between the two components of
a collocation. Other aspects of acquisition, whereby a form–meaning link is
established, are not assessed. Two testing sessions were included: one immediate
and the other after a 2-week delay. The following research questions were
addressed both for natives (Experiment 1) and nonnatives (Experiment 2):
1. Which type of exposure (enriched, enhanced, or decontextualized), if any,
promotes explicit, form recall knowledge of collocations? Are any more
effective than others?
2. Which type of exposure, if any, promotes explicit, form recognition knowl-
edge of collocations? Are any more effective than others?
3. Which type of exposure, if any, promotes implicit knowledge (automatic
priming effects) of collocations? Are any more effective than others?
Our two experiments included both indirect and direct input and measured
both implicit and explicit lexical knowledge. Thus, it was possible to explore
whether indirect input best facilitates implicit knowledge while direct input best
facilitates explicit knowledge, or whether direct or indirect input is superior
for both types of knowledge. Furthermore, we expected that comparison of
results across the two experiments would enable us to determine if natives and
nonnatives gain similar types of advantage from different input conditions in
terms of learning implicit and explicit knowledge, as Hoey (2005) suggests.

Experiment 1: The Acquisition of Collocations by Adult Natives

Experiment 1 evaluates different conditions under which adult native speakers
of English might acquire explicit and implicit knowledge of novel medical
collocations. It consists of a study phase where participants are exposed to
collocations, followed by immediate and delayed test phases where their learn-
ing gains are measured.

Method
Participants
The 35 native participants (15 males and 20 females) were undergradu-
ates studying at a British university who received a payment (£8) for their

129 Language Learning 63:1, March 2013, pp. 121–159

participation. They had no previous medical training and ranged in age from 18
to 27 (M = 19.54, SD = 1.74). Data for a further 8 participants was excluded
from the experiment because of (1) low (less than 60%) scores in the reading
comprehension test (see Procedure Section below) (N = 1), (2) having too
many (more than 15%) incorrect or extreme responses in the lexical decision
task (N = 2), or (3) not showing up in the delayed testing session (N = 5).

Stimuli
We intended to set up as natural a pedagogical setting as possible, and so wished
to use real collocations as stimuli. However, this required finding collocations
which educated native speakers would not already know. We decided to use
medical collocations, as these are unlikely to be known by people without
medical training. Choosing medical collocations for the present study involved
two stages. First, five medical resources1 were consulted in an attempt to
identify candidates that fit the following criteria:

1. The collocation is transparent and is not technically loaded (i.e., no collo-

cations comprising words of opaque Latin origin or names of drugs).
2. The British National Corpus (BNC) frequency of the collocation, checked
using Davies’s (2007) interface, is very low (0–8 occurrences in the whole
corpus). This step indicated that the average native speaker would be un-
likely to know the collocation.
3. The first word of the collocation has a number of synonyms that might look
equally attractive to a person who has no knowledge of the collocation.

The candidate collocations were piloted with undergraduates similar to the

experiment participants to ensure that they were all essentially unknown.
Second, for each of the 15 collocations which fit the above criteria, a control
pair was devised for the priming task (see Materials section below). This pair
included the second word of the intact collocation along with a synonym of
the first (e.g., decaying lung as a control for vanishing lung), with the pair
not occurring in the BNC. The first word of the control pair was closely
matched with the first word of the intact collocation item-by-item in terms of
the following factors: part of speech, semantic plausibility, BNC frequency (the
frequency per million of the control word is within the range of ±30 of the intact
word frequency), and length (there is no more than one letter difference between
the control and the intact words). Appendix S1 in the Supporting Information
online presents the intact collocations and their controls for Experiment 1.

Language Learning 63:1, March 2013, pp. 121–159 130

Figure 1 Target collocations counterbalanced across treatment conditions

(Experiment 1).

Materials
We then chose a reading passage from The Language of Medicine in English
(Tiersky & Tiersky, 1992) which covered a wide range of medical condi-
tions, thus facilitating the inclusion of the intact collocations. According to
Rott (2009), a single exposure can result in an initial memory trace (see also
Durrant & Schmitt, 2010) which should later be strengthened through subse-
quent exposures. Thus, the passage was adopted and shortened to include three
occurrences of 10 intact collocations, varying according to our two contextu-
alized experimental conditions (i.e., enriched and enhanced) as follows. Five
collocations were embedded in the passage (the enriched condition) while the
other five were embedded and made salient with red, bold font (the enhanced
condition). In the third learning condition, five collocations were taught in
isolation in a PowerPoint presentation (the decontextualized condition), where
each collocation was presented individually on a different slide (no meaning
was provided). Each collocation was presented in red font, flashing on the
screen three times for a total of 10 seconds.
The 15 collocations and the 3 learning conditions were placed into a counter-
balanced design consisting of three experimental blocks, as shown in Figure 1.
Because it was not possible to administer a pretest in order not to attract partic-
ipants’ attention to the critical stimuli before the experiment, a control group
was included as a baseline for comparison. The 35 participants were randomly
assigned to the experimental groups (Block 1, N = 10; Block 2, N = 8; Block
3, N = 8) or to the control group (N = 9).

131 Language Learning 63:1, March 2013, pp. 121–159

Measures
Because the focus of the present experiment is on the distinction between
explicit and implicit collocational knowledge, three measures were developed:
two explicit and one implicit. The two explicit measures were based on Schmitt,
Dornyei, Adolphs, and Durow (2004). The first (form recall) employed a cloze
format to measure participants’ ability to recall the form of the collocation.
Participants were presented with a summary of the reading passage with target
collocations in parentheses. A brief definition of each collocation was included
in the margin as a clue. To avoid any potential effects of this test on the
subsequent form recognition one, the first letter of each missing word was not
included. Alternatively, the number of letters in each space was represented by
dashes to restrict participants’ responses. The following example is the form
recall test item for vanishing lung:

Lung infections might lead to different conditions such as the Decrease in the lung’s
( _ _ _ _ _ _ _ _ _ lung) syndrome. size as shown in an X-ray

The second explicit test measured form recognition. It used the same pas-
sage as in the form recall test, but with multiple-choice items given in the
margin. Each item included five options: the correct first word (the key), three
plausible distracters (one is the control and two other synonyms with the same
part of speech), and one “I don’t know” option to reduce guessing. Below is
the form recognition test item for vanishing lung:

a. fading
b. declining
Lung infections might lead to different conditions such as the ( ________ c. decaying
lung) syndrome. d. vanishing
e. I DON’T KNOW

While the explicit measures are fairly traditional, the measurement of implicit
collocation knowledge required a more innovative approach. We opted to use the
priming paradigm to capture implicit knowledge. In our priming task (designed
R
on E-Prime , Psychology Software Tools, Inc., www.pstnet.com), participants
were presented with the first word of the collocation as the prime and the second
as the target. They had to decide whether the second string of letters is a real
English word or not by pressing the appropriate key (YES or NO) on a response
box. The priming task was designed in accordance with McNamara’s (2005)
guidelines to ensure automatic, nonstrategic priming.
Critical stimuli for this task included the 15 intact (exposed) collocations
and their 15 control pairs. The intact and control primes were combined into

Language Learning 63:1, March 2013, pp. 121–159 132

two counterbalanced lists such that seven or eight intact pairs and seven or eight
control pairs were included in each list. Targets which were matched with their
intact primes in one list were matched with their control primes in the other.
No prime or target word was used more than once in either list. In addition,
53 filler items were included, in which the second item of the pair was either
a word (K = 22) or a nonword (K = 31).2 All nonwords were selected from
the ARC nonword database (Rastle, Harrington, & Coltheart, 2002) and were
pronounceable and orthographically legal. In order to reduce the effect of the
implicit test on the explicit form recognition test, 30 fillers included distracters
from the multiple-choice test as primes. Thus, in each stimuli list, participants
were presented with 68 trials (7 or 8 intact pairs, 7 or 8 control pairs, and 53
fillers). Half of the participants (N = 18) were tested with the first stimuli set,
while the other half (N = 17) were tested with the second set. Twenty practice
trials consisting of nonstudy items were constructed to allow participants to get
used to the lexical decision task.
In each trial, a fixation point was presented (2,000 milliseconds) followed
by the first word of each pair (the prime), which remained on the screen for 150
milliseconds. This was immediately replaced by the second string (the target),
which remained on the screen until the participant had made a lexical decision.
The order of the trials was randomized across subjects to avoid potential order
effects. E-Prime recorded accuracy and response latency for each participant
in all trials.

Procedure
Upon arrival at the lab for the single-person session, the participant signed the
consent form in which he or she was briefed on the purpose of the study. Then,
the study phase began. While participants in the experimental blocks took part
in both study and test phases, those in the control group only took part in the
test phase. The participant was instructed to read the whole passage once within
10 minutes and to try to focus on the message as there would be some compre-
hension questions at the end. Upon finishing, the first author took the reading
passage and participants were instructed to watch a PowerPoint presentation
that included five collocations initiated with the following instruction:

Now, you will be presented with five different phrases that usually occur
in medical contexts. Please study them carefully and try to remember the
combination.

After the presentation, a general comprehension test (eight multiple-choice

items) was administered for a dual purpose. First, it was intended to ensure

133 Language Learning 63:1, March 2013, pp. 121–159

that participants were reading for meaning; the mean score gained by the 26
participants in the experimental group was acceptable (7.26 out of 8.00, 90.87%,
SD = 0.96). Second, the test functioned as an interval between the study phase
and the test phase.
To avoid earlier measures affecting later ones, the tests were carefully
sequenced. The first was the computerized priming task, which took 6 minutes.
The participants then performed a distracter task (a digit-span memory test)
that was intended to minimize any effect of the implicit test on the subsequent
explicit tests. Next the participants completed the paper-and-pencil form recall
test within 10 minutes, followed by the form recognition test (5 minutes).
Finally, participants filled out a short background questionnaire.
At the end of the session, participants were invited to a follow-up experiment
to be conducted two weeks later. This entailed a delayed testing session where
the participants took the explicit and implicit tests without being exposed to
any treatment. To minimize any possible transfer from the immediate to the
delayed testing sessions, the choices were reordered in the delayed multiple-
choice test.

Scoring and Data Analysis

When scoring and coding the explicit form recall test, a response that was
correct in spelling and morphology was coded as CORRECT. Because this
test aimed at assessing knowledge of collocations rather than knowledge of
the word form, a response was also coded as CORRECT if the word was
misspelled but recognizable (vacume for vacuum) or if the correct stem was
provided but a wrong affix was used (e.g., reciprocate for reciprocal) (partially
correct answers constituted only 0.66% of all data points in the recall test). A
response was coded as INCORRECT if no response or a wrong response was
given. Scoring the explicit form recognition test was straightforward; a correct
answer was coded as CORRECT and an incorrect, missing, or “I don’t know”
response was coded as INCORRECT.
We conducted the analysis with R version 2.12.1 (R Development Core
Team, 2010) using mixed-effects models, that is, models that include both fixed
effects (the independent variables) and random effects (participants and items).
Mixed-effects modeling is now becoming a popular method of analysis in psy-
cholinguistic research for a number of reasons. First, it can treat participants and
items as random variables in one model allowing for “the simultaneous gener-
alization of the results on new items and new participants” (Gagné & Spalding,
2009, p. 25). Second, it allows testing main independent variables in addi-
tion to other psycholinguistic (item-related and participant-related) variables.

Language Learning 63:1, March 2013, pp. 121–159 134

Third, it can handle interval-scale (e.g., reaction time) measures through linear
mixed-effects (LME; see Baayen, 2008) models as well as categorical (e.g.,
correct versus incorrect) measures using mixed logit models (the LME alter-
native for categorical data; see Jaeger, 2008). Lastly, it can cope with missing
values and imbalanced designs. Because the present study involved both a cor-
rect/incorrect categorical measure along with an interval-scale reaction time
measure and because we aimed at testing various item-related and participant-
related factors, we opted to use mixed-effects modeling as the analysis method.
The method starts with the simplest (null) model, which includes only the
dependent measure and the random variables. Fixed effects are then added incre-
mentally and χ 2 (likelihood ratio) tests are used to check whether the inclusion
of additional predictors contributes significantly to the model. Once the interim
best-fit model is reached, variables are excluded one by one to check whether
any predictor is redundant (i.e., a variable whose presence/absence in the model
does not lead to any significant difference) and can thus be excluded. Once these
are excluded, we are left with the final best-fit model, which best describes the
overall variance in the research design. As mixed-effects modeling is likely
to be new for many readers, we provide additional detailed description of the
procedure we followed in Appendix S3 of the Supporting Information online.
We developed two mixed-effects models, one for explicit knowledge and
one for implicit knowledge. For explicit knowledge, the analysis (mixed logit
modeling) started with a null model including our binomial dependent variable
(CORRECT versus INCORRECT responses) and participants and items as ran-
dom effects. We then added predictor variables incrementally in the following
order: main effects (TEST TYPE: form recall, form recognition; CONDITION:
enriched, enhanced, decontextualized, control; and SESSION: immediate, de-
layed3 ) and then item-related effects (length of Word 1, log frequency of Word
1, length of Word 2, log frequency of Word 2, frequency of the collocation).
Finally, variables were excluded one by one to check for redundancy to arrive
at the final best-fit model. The procedure for implicit knowledge (using LME
modeling) was the same except that the dependent variable was log response
time4 (reaction times were log-transformed to reduce skewness in the distribu-
tion) instead of CORRECT/INCORRECT responses, and the first main effect
was ITEM TYPE (intact, control) instead of TEST TYPE.

Results
Explicit Form Recall and Form Recognition Tests
An assumption of the experiment is that the medical collocations were unknown
to the participants. This assumption is supported by the control group results

135 Language Learning 63:1, March 2013, pp. 121–159

Table 1 Responses and percentile scores in explicit tests (form recall and recognition)
under all conditions (natives)
Explicit Knowledge Tested

Form Recall Form Recognition

(Cloze-test) (Multiple-choice)

Condition Session Correct % Incorrect % Correct % Incorrect %

Controla Immediate 2 1.5 133 98.5 43 31.8 92 68.2
Delayed 13 9.6 122 90.4 46 34.1 89 65.9
Decontextualizedb Immediate 44 33.8 86 66.2 97 74.6 33 25.4
Delayed 37 28.5 93 71.5 92 70.8 38 29.2
Enhancedb Immediate 36 27.7 94 72.3 93 71.5 37 28.5
Delayed 34 26.2 96 73.8 89 68.5 41 31.5
Enrichedb Immediate 29 22.3 101 77.7 90 69.2 40 30.8
Delayed 34 26.2 96 73.8 84 64.6 46 35.4
a
Max score = 135 (K = 15 × N = 9).
b
Max score = 130 (K = 5 × N = 26).

(Table 1), which showed very low form recall collocation knowledge (1.5%).
However, it should be noted that the recall percentage score for the delayed test
(9.6%) was higher, which suggests that the implicit priming test might have had
a direct effect on explicit test results (see Limitations section). As opposed to
the lack of form recall knowledge by the control group, this group answered one
third of the items in the recognition test correctly (around 32% immediate and
34% delayed). This implies that participants were merely guessing on the four-
option multiple-choice test. Thus, if participants under the treatment conditions
score higher than the 9.6% level (recall) and 34.1% level (recognition), gains
should be attributed to the treatment.
We modeled CORRECT answers likelihood using a mixed logit model.
Main effects were added to the null model incrementally. The addition of TEST
TYPE significantly improved the null model (χ 2 (1) = 429.80, p < .001, N =
2,100). Adding CONDITION further improved the model (χ 2 (3) = 26.68, p <
.001, N = 2,100). Then, a TEST TYPE × CONDITION interaction was added
to the model but this did not lead to any significant improvement (χ 2 (3) = 0.15,
p = 0.99, N = 2,100). Moreover, adding SESSION as a main effect did not
improve the model (χ 2 (1) = 0.08, p = .78, N = 2,100), neither did adding it as
an interacting variable with CONDITION (χ 2 (4) = 5.15, p = .27, N = 2,100) or
TEST TYPE (χ 2 (2) = 1.14, p = .56, N = 2100). SESSION was thus removed
from the model. Item-related variables were then added incrementally, but none

Language Learning 63:1, March 2013, pp. 121–159 136

Table 2 Summary of the best fit mixed logit model for variables predicting natives’
correct explicit responses

Predictor Estimate SE Wald-Z p

(Intercept) −3.15 0.38 −8.38 <.001

TEST TYPE: recognition 2.31 0.12 18.81 <.001
CONDITION: decontextualized 2.11 0.38 5.53 <.001
enhanced 1.90 0.38 4.99 <.001
enriched 1.72 0.38 4.52 <.001
Note. The model has random intercepts for participants and items. N = 2,100, log-
likelihood = -1,034.08.

improved the model. Finally, significant predictors were removed backward but
none was found to be redundant. The final best-fit model is presented in Table 2.
The best-fit mixed logit model indicates that more correct responses were
produced in the recognition test (634/863) than in the recall test (229/863).
Also, more correct responses (237, 252, 270/863) were produced under the
three experimental conditions (enriched, enhanced, decontextualized, respec-
tively) than those produced by the control group (104/863). In order to assess
differences across experimental conditions, a multiple comparison analysis (see
Bretz, Hothorn, & Westfall, 2010) was conducted on the CONDITION variable.
Results showed that more correct responses were produced under the decon-
textualized condition than the enriched condition (Wald Z = 2.55, p = .046)
while no significant difference was found between the enhanced condition and
the other two conditions (enriched: Wald Z = -1.17, p = .63, decontextualized:
Wald Z = 1.38, p = .49). Third, lack of interaction between TEST TYPE and
CONDITION implies that (1) the advantage of the recognition gains over re-
call gains was present under all conditions and (2) significant differences across
conditions were present both for recall and recognition gains. Finally, the fact
that SESSION was not a significant predictor implies that explicit gains did not
drop significantly after the 2-week gap and that the effects reported above are
true for immediate and delayed results.
Overall, immediate and delayed results (for both recall and recognition
tests) can be summarized as follows:

Enhanced = Enriched > Control

Enhanced = Decontextualized > Control
Decontextualized > Enriched

137 Language Learning 63:1, March 2013, pp. 121–159

Figure 2 Percentages and comparisons of explicit (form recall and recognition) scores
under all conditions in both sessions (natives).

These results are also graphically illustrated in Figure 2. The three experimental
conditions led to higher explicit gains than the control group (around 16–19
percentage points for recall and 30–37 percentage points for recognition both
on the delayed test5 ). Also, the decontextualized condition led to more correct
answers than the enriched condition on the delayed test (two percentage point
difference in recall and six in recognition).

Implicit Priming Test

If implicit memory for the intact collocations was formed during the study
phase, the experimental conditions should produce faster (shorter) reaction
times to these collocations than to their control counterparts, but this should
not be the case for the control group. Table 3 presents mean reaction times
under all conditions.
First we checked the overall accuracy in the lexical decision task, and
it was high (97.43%).6 We then fit a LME model with log reaction time as
the dependent variable and main effects (ITEM TYPE, CONDITION, and
SESSION) added incrementally as predictors to the null model. ITEM TYPE
did not improve the null model (χ 2 (1) = 0.00, p = 1.00, N = 1,001) and
was thus removed. Then, CONDITION was added as a predictor but did not

Language Learning 63:1, March 2013, pp. 121–159 138

Table 3 Mean reaction time (RT) in milliseconds for intact and control items under all
conditions (natives)

Intact Items Control Items

Difference in
Condition Session RT Mean SEa RT Mean SEa RT Mean

Control Immediate 695.81 26.31 702.02 27.59 6.21

Delayed 651.89 17.93 639.00 21.67 −12.89
Decontextualized Immediate 644.59 26.16 650.11 23.48 5.52
Delayed 605.10 19.81 637.27 23.47 32.17
Enhanced Immediate 656.00 23.77 679.15 28.61 23.15
Delayed 611.84 19.04 661.43 23.38 49.59
Enriched Immediate 674.00 25.95 648.18 25.91 −25.82
Delayed 644.32 24.44 648.66 24.37 4.34
a
Standard error of mean.

lead to any significant improvement in the null model (χ 2 (3) = 4.00, p = .26,
N = 1,001), and so was also removed. SESSION was added as a predictor and
was found to significantly improve the model (χ 2 (1) = 10.11, p = .001, N =
1,001) with shorter (faster) reaction time in the delayed session. After testing
all main effects, item-related variables were added incrementally to assess their
effects on log reaction time. Only Word 1 length (χ 2 (1) = 5.27, p = .02, N =
1,001), Word 2 length (χ 2 (1) = 18.13, p < .001, N = 1,001), and Word 2 log
frequency (χ 2 (1) = 15.04, p < .001, N = 1,001) led to significant improvement
in the model. Thus, the shorter the first word and the shorter/more frequent the
second word, the shorter the log reaction time is. We then tested interactions
between significant predictors (SESSION, W1 length, W2 length, and W2 log
Frequency). Only the interaction between W2 length and W2 log Frequency led
to a significant improvement in the model (χ 2 (1) = 5.71, p = .02, N = 1,001).
In this model, W2 log frequency ceased to be a significant predictor. Thus, the
effect of the second word frequency on reaction times (RT) does not seem to be
direct but is rather modulated by length (i.e., more frequent second words lead
to faster RT only if they are short). Finally, removing the significant predictors
backward did not show any redundant effects, and so all were retained. Table 4
presents the final best-fit model.
The nonsignificant interaction between CONDITION and ITEM TYPE in-
dicates that natives’ RT for intact versus control primes was not predicted by
the type of input (enriched, enhanced, decontextualized, or no input [control])
either in the immediate or the delayed sessions. This result provides a clear
answer to the third research question; namely, natives did not develop implicit

139 Language Learning 63:1, March 2013, pp. 121–159

Table 4 Summary of the best fit LME model for variables predicting natives’ log
reaction time
Predictor Estimate MCMC SE t pMCMC Pr(>|t|)

(Intercept) 6.15 6.15 0.05 114.05 0.001 <.001

SESSION: immediate 0.04 0.04 0.01 3.16 0.006 0.002
Word 1 Length 0.02 0.02 0.00 3.62 0.001 0.004
Word 2 Length 0.03 0.03 0.00 6.03 0.001 <.001
Word 2 Log Frequency 0.05 0.05 0.04 1.30 0.15 0.20
W2 Length × W2 Log −0.01 −0.01 0.01 −2.36 0.03 0.02
Frequency
Note. The model has random intercepts for participants and items. N = 1,001, R2 =
0.54. MCMC = Monte Carlo Markov chain; pMCMC = p values estimated by the
MCMC chain method using 10,000 simulations; Pr(>|t|) = p values obtained with the
t test using the difference between the number of observations and the number of fixed
effects as the upper bound for the degrees of freedom.

Figure 3 Mean reaction times for intact and control items under all conditions in both
sessions (natives).

knowledge of collocations under any condition, at least as measured by our

priming measure. This can be summarized as follows (and is graphically illus-
trated in Figure 3):

Immediate: Enriched condition = no effect

Enhanced condition = no effect

Language Learning 63:1, March 2013, pp. 121–159 140

Decontextualized condition = no effect

Control group = no effect
Delayed: Enriched condition = no learning
Enhanced condition = no learning
Decontextualized condition = no learning
Control group = no learning

Discussion
Experiment 1 compared three treatment conditions for their ability to promote
collocation learning. On the explicit side of knowledge (form recall and form
recognition), all treatment conditions led to learning. Moreover, the decontex-
tualized condition led to higher gains than the enriched condition for both types
of knowledge and in both testing sessions. However, the enhanced condition
seems to stand in the middle between the other two treatment conditions, in that
it did not lead to any significant difference. Furthermore, just as with single
words, form recognition knowledge is stronger than form recall knowledge.
In terms of implicit knowledge, one may have speculated that explicit
learning conditions would lead to explicit knowledge, while implicit knowl-
edge would accrue best from an indirect learning condition without an explicit
component. However, we found that natives did not develop either short-term
or long-term implicit memory traces for intact collocations under any treat-
ment condition. To conclude, it seems that promoting explicit knowledge in
native speakers is possible with a wide range of treatment conditions. Implicit
knowledge, on the other hand, seems to be difficult to develop from such a short
treatment period.

Experiment 2: The Acquisition of Collocations by Adult

NonNatives
Experiment 2 builds upon Experiment 1 and aims at investigating the condi-
tions under which advanced adult nonnatives might acquire English medical
collocations. The design mirrors that of Experiment 1 except for one modifica-
tion: collocations were chosen carefully to ensure a suitable level of difficulty
for nonnatives.
Method
Participants
The 43 participants (13 males and 30 females) were postgraduate students
(master’s = 25, doctoral = 18) majoring in nonmedical fields at a British

141 Language Learning 63:1, March 2013, pp. 121–159

university, who had all met the university entry requirement (minimum IELTS
score of 6.00 or TOEFL score of 550). They ranged in age from 21 to 41 (M =
28.65, SD = 4.91) and came from a variety of first language (L1) backgrounds
(Arabic = 6, Bengali = 1, Bulgarian = 1, Chinese = 9, Dutch = 2, Farsi = 1,
French = 2, Hindi = 1, Italian = 1, Kurdish = 1, Luxemburgish = 1, Malay =
2, Polish = 2, Portuguese = 2, Romanian = 2, Russian = 1, Spanish = 3,
Swedish = 1, Swiss = 1, Thai = 1, Turkish = 1, and Yoruba = 1). The
participants had spent a mean of 25.59 months in the United Kingdom (SD =
26.91, Min = 6, Max = 132), but were first exposed to English at an average age
of 10.78 years (SD = 4.41, Min = 3, Max = 22). Their self-rated proficiency
scores (on a scale from 1 = very poor to 5 = excellent) were: reading M =
4.14, SD = 0.74; writing M = 3.97, SD = 0.89; speaking M = 3.86, SD = 0.64,
and listening M = 3.95, SD = 0.69. Their overall self-rated proficiency score
(averaged across skills) was 3.98 (SD = 0.62). The participants were offered
a £10 payment for their participation. Data for a further 5 participants were
excluded from the experiment for having too many (more than 15%) incorrect
or extreme responses in the lexical decision task (N = 3) or for not showing up
for the delayed test (N = 2).

Stimuli
The procedures from the previous experiment were employed in choosing med-
ical intact collocations and their controls, except for the following additional
criterion:

Component words of each collocation should belong to the most frequent

3,000 lemmas in the BNC (Leech, Rayson, & Wilson, 2001) or to the
General Service List (West, 1953).

This step was intended to ensure that the nonnatives knew the individual words
comprising each collocation. We ended up with 15 viable items (see Appendix
S2 in the online Supporting Information). These were divided into three bal-
anced sets of five collocations, which were then counterbalanced across the
three treatment conditions, as shown in Figure 4.

Materials
The passage from Experiment 1 was adapted to include three occurrences of
each collocation and simplified somewhat to be of a suitable level of difficulty
for the nonnative participants.7 Other than this, the treatment conditions were
the same. The 43 participants were randomly assigned to the experimental
groups (Block 1, N = 11; Block 2, N = 11; Block 3, N = 12) or to the

Language Learning 63:1, March 2013, pp. 121–159 142

Figure 4 Target collocations counterbalanced across treatment conditions

(Experiment 2).

control group (N = 9). Participants in these blocks did not differ in terms of
the time they spent in the United Kingdom, age they were exposed to English,
or self-rated proficiency (all ps > .20).

Measures, Procedures, Scoring, and Data Analysis

The designs of the two explicit tests (cloze and multiple-choice) and the implicit,
priming test followed guidelines from Experiment 1. Likewise, the procedures
and scoring8 of this experiment were the same as those followed in Experiment
1, with two exceptions. First, because nonnatives are obviously slower than
natives in lexical decision, we decided to extend the long cutoff time in the im-
plicit measure to 1,800 milliseconds. Responses shorter than 150 milliseconds
or longer than 1,800 milliseconds (constituting 1.97% of all data points) were
excluded from further analysis. Second, one participant-related factor (overall
self-rated proficiency score averaged across skills) was included to check its
effect on explicit and implicit gains.

Results
Explicit Form Recall and Form Recognition Tests
The experiment assumed that medical collocations were unknown to the non-
native participants, and the results of the control group on the immediate tests
seem to confirm this (Table 5), with only a 2.2% score on the form recall test,
and a 29.6% on the form recognition test, which is close to chance. However, it
should be noted that, similar to results of Experiment 1, the percentage scores
for the delayed tests (10.4% recall, 45.9% recognition) are far higher.

143 Language Learning 63:1, March 2013, pp. 121–159

Table 5 Responses and percentile scores in explicit tests (form recall and recognition)
under all conditions (nonnatives)
Explicit Knowledge Tested

Form Recall Form Recognition

(Cloze-test) (Multiple-choice)

Condition Session Correct % Incorrect % Correct % Incorrect %

a
Control Immediate 3 2.2 132 97.8 40 29.6 95 70.4
Delayed 14 10.4 121 89.6 62 45.9 73 54.1
Decontextualizedb Immediate 72 42.4 98 57.6 142 83.5 28 16.5
Delayed 63 37.1 107 62.9 125 73.5 45 26.5
Enhancedb Immediate 88 51.8 82 48.2 148 87.1 22 12.9
Delayed 80 47.1 90 52.9 133 78.2 37 21.8
Enrichedb Immediate 60 35.3 110 64.7 127 74.7 43 25.3
Delayed 63 37.1 107 62.9 114 67.1 56 32.9
a
Max score = 135 (K = 15 × N = 9).
b
Max score = 170 (K = 5 × N = 34).

We modeled CORRECT answers likelihood using a mixed logit model. The

addition of TEST TYPE significantly improved the null model (χ 2 (1) = 432.52,
p < .001, N = 2,580). Adding CONDITION further improved the model
(χ 2 (3) = 65.56, p < .001, N = 2,580). Then, a TEST TYPE × CONDITION
interaction was added to the model not leading to any significant improvement
(χ 2 (3) = 4.92, p = .18, N = 2,580). Moreover, adding SESSION as a main
effect did not improve the model (χ 2 (1) = 1.70, p = .19, N = 2,580). However,
adding it as an interacting variable with CONDITION improved the model
(χ 2 (4) = 28.14, p < .001, N = 2,580) and its interaction with TEST TYPE
further improved the model (χ 2 (1) = 5.30, p = .02, N = 2,580). Item-related
variables were then added incrementally but none improved the model. Finally,
adding the single participant-related variable (overall self-rated proficiency
score) improved the model (χ 2 (1) = 12.65, p < .001, N = 2,580). Furthermore,
this variable interacted with TEST TYPE (χ 2 (1) = 4.88 p = .03, N = 2,580)
and SESSION (χ 2 (1) = 10.76, p = .001, N = 2,580), significantly improving
the model. As a final step in the analysis, significant predictors were removed
backward. Only the TEST TYPE × SESSION interaction was found to be
redundant (χ 2 (1) = 3.28, p = .07, N = 2,580) and was thus removed from the
model. The final best-fit model is presented in Table 6.
Significant main effects in Table 6 can be summarized as follows: (1) more
correct responses were produced in the recognition test (891/1,334) than in
the recall test (443/1,334); (2) more correct responses were produced (364,

Language Learning 63:1, March 2013, pp. 121–159 144

Table 6 Summary of the best fit mixed logit model for variables predicting nonnatives’
correct explicit responses
Predictor Estimate SE Wald-Z p
(Intercept) −7.63 1.05 −7.25 <.001
TEST TYPE: recognition 3.87 0.72 5.36 <.001
CONDITION: decontextualized 1.60 0.35 4.54 <.001
enhanced 2.12 0.36 5.96 <.001
enriched 1.42 0.35 4.04 <.001
SESSION: immediate 1.42 0.72 1.98 0.047
Overall Proficiency 1.32 0.25 5.37 <.001
CONDITION: (decontextualized) X SESSION: (immediate) 1.48 0.31 4.70 <.001
(enhanced) 1.42 0.32 4.48 <.001
(enriched) 1.17 0.31 3.76 <.001
TEST TYPE: (recognition) X Overall Proficiency −0.44 0.18 −2.45 0.01
SESSION: (immediate) X Overall Proficiency −0.61 0.17 −3.55 <.001
Note. The model has random intercepts for participants and items. N = 2,580, log-
likelihood = −1,249.

449, 402/1,334) under the three experimental conditions (enriched, enhanced,

decontextualized, respectively) than those produced by the control group
(119/1,334); (3) SESSION led to a significant drop in explicit gains over the
2-week delay with more correct responses in the immediate than the delayed
session (680, 654/1,334, respectively); and (4) as the proficiency of L2 learners
increased, they produced more correct answers in explicit measures. It should
be noted, however, that these main effects were modulated by a number of
interactions. First, the TEST TYPE × overall proficiency interaction seems
to suggest that the difference between gains achieved in the two explicit tests
(recall versus recognition) is modulated by overall proficiency with a larger
difference between recall and recognition knowledge as proficiency increases
(see Figure 5, left panel). Second, the normal direction of the SESSION effect
(showing attrition in knowledge) changed as it interacted with overall profi-
ciency. More proficient L2 learners showed a clear improvement in explicit
knowledge in the delayed session (Figure 5, right panel).
Third, because the CONDITION × SESSION interaction was significant,
we defined a contrast matrix to explore all possible comparisons for this specific
interaction through multiple comparison analysis (see Table 7)9 , and three
points are worth noticing: (1) the difference between the control group and the
three experimental conditions was present in both sessions; (2) among the three
experimental conditions, only one difference was significant in both sessions
(i.e., the difference between the enhanced and the enriched conditions); and

145 Language Learning 63:1, March 2013, pp. 121–159

Figure 5 Interaction between proficiency and TEST TYPE (left panel) and between
proficiency and SESSION (right panel) in the logit model for the nonnative explicit data
(Note. The vertical axis depicts the probability of CORRECT answers transformed from
log odds).

(3) a significant difference between the immediate and the delayed sessions
was only present for the control group (with fewer correct responses in the
immediate session). Finally, CONDITION did not interact with TEST TYPE
suggesting similar results for both recall and recognition tests.
To conclude, the three treatment conditions led to delayed recall gains of
27–37 percentage points and to delayed recognition gains of 21–32 percentage
points over the control group. Also, the enhanced condition led to more correct
answers than the enriched condition in the delayed test (10 and 11 percentage
point difference in recall and recognition, respectively). These results can be
summarized as follows (and are illustrated in Figure 6):

Decontextualized = Enriched > Control

Decontextualized = Enhanced > Control
Enhanced > Enriched

Implicit Priming Test

Overall accuracy in the lexical decision task was high (98.53%). Mean reaction
time under all conditions is presented in Table 8.
We fit a LME model with log reaction time as the dependent variable and
main effects (ITEM TYPE, CONDITION, and SESSION) added incrementally

Language Learning 63:1, March 2013, pp. 121–159 146

Table 7 Multiple comparisons for the CONDITION × SESSION interaction (nonna-

tives’ explicit results)

Predictor Estimate SE Wald-Z p

Immediate Session Contrasts

Decontextualized–Control 2.57 0.35 7.35 <.001
Enhanced–Control 2.95 0.35 8.37 <.001
Enriched–Control 2.18 0.35 6.26 <.001
Enhanced–Decontextualized 0.38 0.18 2.13 0.37
Enriched–Decontextualized −0.39 0.17 −2.30 0.27
Enriched–Enhanced −0.77 0.17 −4.39 <.001
Delayed Session Contrasts
Decontextualized–Control 1.36 0.33 4.10 <.001
Enhanced–Control 1.76 0.33 5.26 <.001
Enriched–Control 1.21 0.33 3.64 <.001
Enhanced–Decontextualized 0.40 0.17 2.32 0.26
Enriched–Decontextualized −0.15 0.17 −0.90 0.98
Enriched–Enhanced −0.55 0.17 −3.21 0.02
Immediate Versus Delayeda
Control −0.83 0.23 −3.64 0.01
Decontextualized 0.38 0.17 2.25 0.30
Enhanced 0.36 0.18 2.05 0.41
Enriched 1.42 0.17 0.85 0.99
a
With delayed as the reference level.

Table 8 Mean reaction rime (RT) in milliseconds for intact and control items under all
conditions (nonnatives)

Intact Items Control Items

Difference in
Condition Session RT Mean SEa RT Mean SEa RT Mean

Control Immediate 711.93 25.81 712.78 28.03 0.85

Delayed 644.23 24.20 627.32 27.47 −16.91
Decontextualized Immediate 715.29 22.22 673.01 24.74 −42.28
Delayed 705.85 28.96 664.99 22.21 −40.86
Enhanced Immediate 657.58 20.08 761.04 29.20 103.46
Delayed 677.56 22.59 726.80 29.03 49.24
Enriched Immediate 783.68 30.50 712.72 25.64 −70.96
Delayed 688.96 24.86 693.73 24.58 4.77
a
Standard error of mean.

147 Language Learning 63:1, March 2013, pp. 121–159

Figure 6 Percentages and comparisons of explicit (form recall and recognition) scores
under all conditions in both sessions (nonnatives).

to the null model. Similar to Experiment 1, ITEM TYPE did not improve the
null model (χ 2 (1) = 0.00, p = 1.00, N = 1,246), neither did CONDITION
(χ 2 (3) = 6.77, p = .08, N = 1,246). However, SESSION did improve the model
(χ 2 (1) = 17.63, p < .001, N = 1,246) with shorter (faster) reaction time in the
delayed session. Only two-item related variables further improved the model:
Word 2 length (χ 2 (1) = 27.92, p < .001, N = 1,246) and Word 2 log frequency
(χ 2 (1) = 13.42, p < .001, N = 1,246). The shorter and more frequent the
second word, the shorter the log reaction time is. Finally, overall proficiency
was added to the model and was found to significantly improve it (χ 2 (1) = 7.05,
p = .008, N = 1,246). The more proficient the participant, the shorter his/her
log reaction time is. We then tested interactions between significant predictors
(SESSION, W2 length, W2 Frequency, and overall proficiency), but none led to
any significant improvement in the model. Finally, when removing significant
predictors backward, none was found to be redundant. Table 9 presents the final
best-fit model.
Similar to Experiment 1, the best fit LME model seems to suggest that
nonnatives’ RT performance for intact as opposed to control items was
not predicted by condition (i.e., no interaction between ITEM TYPE and
CONDITION). It can be concluded, thus, that implicit knowledge was not

Language Learning 63:1, March 2013, pp. 121–159 148

Table 9 Summary of the best fit LME model for variables predicting nonnatives’ log
reaction time
Predictor Estimate MCMCmean SE t pMCMC Pr(>|t|)

(Intercept) 6.62 6.61 0.16 40.60 0.001 <.001

SESSION: immediate 0.05 0.05 0.01 4.19 0.001 <.001
Word 2 Length 0.05 0.05 0.00 8.46 0.001 <.001
Word 2 Log Frequency −0.04 −0.04 0.01 −4.01 0.001 <.001
Overall Proficiency −0.11 −0.11 0.04 −2.71 0.007 0.007
Note. The model has random intercepts for participants and items. N = 1,246, R2 =
0.48. MCMC = Monte Carlo Markov chain; pMCMC = p values estimated by the
MCMC chain method using 10,000 simulations; Pr(>|t|) = p values obtained with the
t test using the difference between the number of observations and the number of fixed
effects as the upper bound for the degrees of freedom.

Figure 7 Mean reaction times for intact and control items under all conditions in both
sessions (nonnatives).

developed under any condition. The results can be summarized as follows (see
Figure 7):

Immediate: Enhanced condition = no effect

Decontextualized condition = no effect
Enriched condition = no effect
Control group = no effect
Delayed: Enhanced condition = no learning
Decontextualized condition = no learning

149 Language Learning 63:1, March 2013, pp. 121–159

Enriched condition = no learning

Control group = no learning

Discussion
As opposed to native speakers, explicit learning plays a much more prominent
role in L2 vocabulary acquisition (e.g., Nation, 2001; Schmitt, 2008), although
little is known about the learning of collocations. Experiment 2 explored three
learning conditions and found that they all led to significant gains in collocation
knowledge at both the form recall and form recognition levels of mastery. This
was true for the immediate test, but more importantly, it was also the case of the
delayed test. Therefore, all three of the treatments were effective in facilitating
durable learning. Moreover, the enhanced condition was more effective than
the enriched condition in enhancing both types of explicit knowledge (though
not more effective than the decontextualized condition). Finally, recognition
gains were always higher than recall gains and this difference increased as the
proficiency of the L2 learners increased.
What about implicit knowledge? The immediate and delayed priming tests
showed that there was no improvement of this knowledge. It seems that the types
and amounts of treatment exposure in this experiment were not sufficient to
establish long-term implicit knowledge, at least not that the priming paradigm
could capture. Overall, although the treatments all facilitated the learning of
explicit collocational knowledge, they did not facilitate measurable learning of
implicit knowledge.

General Discussion
The Relationship Between Explicit and Implicit Lexical Knowledge
Previous research has shown that direct conditions are generally more effec-
tive than indirect conditions in promoting lexical learning, but it must be said
that this research has a number of important limitations: (1) the lexical items
researched have almost always been individual words, (2) the aspect of word
knowledge addressed has almost always been the form-meaning link, and (3)
the research measurements have almost always been explicit tests. It is there-
fore perhaps not surprising that direct approaches have usually been shown
to have an advantage, because the form-meaning link of individual words is
relatively amenable to explicit introspection, and the explicit measurements
have privileged explicit knowledge. Thus, it is unclear whether nonexplicit in-
direct learning conditions can be as, or more, effective in promoting implicit
knowledge, simply because implicit knowledge has seldom been tested before.

Language Learning 63:1, March 2013, pp. 121–159 150

Our study design allows such a comparison between direct/indirect input

and explicit/implicit knowledge. The high-proficiency educated natives bene-
fited from both indirect and direct (enhanced and decontextualized) exposure
to collocations, in terms of enhancing explicit (recall and recognition) knowl-
edge. Moreover, the decontextualized condition showed a clear advantage over
the enriched condition in developing both recall and recognition knowledge
although it did not hold this advantage over the other direct (enhanced) condi-
tion. These results are in line with previous research which shows that explicit
input best leads to explicit knowledge. But does explicit input still hold this
advantage when it comes to enhancing implicit knowledge? The native results
suggest that it does not. None of the three treatment conditions led to significant
implicit knowledge gains. Thus, with very proficient speakers, it seems that the
rather small amounts of input in our various treatment conditions may lead to
explicit lexical knowledge, but not implicit priming effects.
What about advanced nonnatives? First, similar to natives, all of the three
treatments led to durable form recognition and form recall collocation learning.
Second, unlike natives, the enhanced condition seems to lead to higher explicit
gains than the enriched condition although it did not hold this advantage over
the decontextualized condition. Also, the decontextualized condition did not
lead to any higher gains than the enriched condition. This is incongruent with
research with individual words, where more direct treatments almost always
hold an advantage. However, it fits with Bishop’s (2004) finding that formulaic
sequences that were enhanced (underlined and in red font) were clicked upon
more often for glosses than the same sequences when appearing in plain font.
This certainly suggests the benefits of making formulaic language more salient
in texts for L2 learners through typographical enhancement. Third, nonnatives
did not develop any durable implicit memory traces of collocations.
In both experiments, there was a clear dissociation between explicit and
implicit lexical knowledge, and it looks as if one does not imply the other.
Implicit knowledge (as measured using the priming paradigm) seems to be
harder to enhance than explicit knowledge not only for nonnative speakers, but
also for native users of the language. The fact that implicit knowledge was
harder to enhance than explicit knowledge seems to contradict recent ERP evi-
dence from grammar (Osterhout et al., 2006; Tokowicz & MacWhinney, 2005)
and vocabulary (J. McLaughlin et al., 2004) research where implicit knowl-
edge was found to precede explicit knowledge. It should be noted, however,
that the automatic priming paradigm might not be sensitive enough in capturing
the very initial implicit memory traces built after such a short exposure period.
This short treatment period (with no built-in recycling) may have disadvantaged

151 Language Learning 63:1, March 2013, pp. 121–159

the learning of implicit knowledge more than that of explicit knowledge. That
is, it may be that recycling is more important for the learning of implicit knowl-
edge than it is for explicit knowledge. R. Ellis (2007) observes that studies that
did show growth in implicit grammatical knowledge involved prolonged and
substantial instruction.

Natives Versus Nonnatives: Absolute Learning of Explicit Collocation

Knowledge
Hoey (2005) suggests that native speakers and nonnative learners of a language
do not differ in how they acquire collocations, and that it mainly depends on
the type and amount of input they receive. It is thus interesting to compare
the native results with the advanced-proficiency nonnatives to explore this
claim. As neither group developed implicit collocational knowledge, we will
only discuss results of the two explicit measures. Both natives and nonnatives
showed clear evidence of durable learning of explicit knowledge under all
treatment conditions with one direct condition (decontextualized for natives
and enhanced for nonnatives) showing an advantage over the indirect, enriched,
condition. Thus, our results seem to support Hoey’s (2005) claim. Adult natives
and nonnatives do not seem to gain greatly different advantages from direct
versus indirect input when acquiring collocations: both natives and nonnatives
benefited more from direct types of input. This result is also in line with
Spada and Tomita’s (2010) conclusion that direct instruction leads to higher
grammatical explicit knowledge gains than indirect instruction. It seems that
explicit knowledge gains after treatment is one area where vocabulary and
grammar findings overlap.
It is also useful to consider the amount of absolute long-term (delayed)
learning that accrued from the input conditions. After the focused instruction
(decontextualized) treatment, natives were able to recognize the collocations
form about 37 percentage points better than the nonstudy control condition;
after the enhanced treatment, about 34 percentage points better; and after the en-
riched treatment, about 30 percentage points better. Typically form recall mas-
tery is more difficult to achieve, but even here, the decontextualized treatment
led to about 19 percentage point gains and the enhanced/enriched conditions
led to about 16 percentage point gains. One might expect that the nonnatives
would have less uptake from the treatments. This was the case for the easier
form recognition test (about 21–32 percentage points better than the controls),
but not for the higher-level form recall knowledge (about 27–37 percentage
points better than the controls). This result may seem surprising as both natives
and nonnatives were exposed to the same amount of input. One might speculate

Language Learning 63:1, March 2013, pp. 121–159 152

that while natives tend to focus mainly on the communicative meaning, non-
natives inevitably focus relatively more on the language form, and this may be
a factor in enhancing their learning of collocational word combinations from
input (we did not test the learning of the collocation’s meaning) at least on the
explicit side. This conclusion is sustained by the finding that, in comparison
with the enriched condition, natives benefited more from decontextualized ex-
posure where their attention was explicitly focused on the collocation’s form,
while advanced nonnatives benefited more from having their attention focused
on the collocations through contextualized, salient exposure.

Limitations and Suggestions for Future Research

Inevitably, there are a number of limitations that need to be noted in this study.
First, it is possible that some participants had some previous knowledge of the
collocations targeted in this study. However, although we did not administer
a pretest in order not to direct participants’ attention to these collocations,
our piloting indicated that learners similar to our participants had no previous
knowledge of the specialist medical collocations. Second, all participants took
all tests in the same order and in a short period of time. In spite of the procedures
taken to minimize the effect of earlier tests on later ones, the more proficient L2
learners in the control group showed a significant improvement in the delayed
explicit tests. This might be explained as a direct effect from the priming
test to the explicit tests. It should be noted, however, that despite this potential
undesirable effect of the priming test, the statistical analysis dealt carefully with
that limitation as only gains that were significantly higher than those achieved
by the control group were considered to be a reliable evidence of learning.
Third, the study used a relatively small number of stimuli, but the use of the
priming test made this unavoidable, as it was necessary to find collocations
for which suitable control counterparts (controlled for frequency, length, and
semantic plausibility) could be found. Finally, many factors known to affect L2
learning were not considered in the present study (e.g., time on task, repetition,
and recycling). The short, massed treatment, for example, might have put the
indirect approach (enriched condition) at a disadvantage because unenhanced
exposure is said to aid development in knowledge gradually over time (N. Ellis,
2002; R. Ellis, 2009; Spada & Tomita, 2010). Thus, results of the present study
should be interpreted carefully and may not yet have clear implications for
language teaching practice.
While the area of vocabulary studies is now establishing an understand-
ing of the acquisition of explicit knowledge (e.g., Schmitt, 2008), it is only

153 Language Learning 63:1, March 2013, pp. 121–159

just beginning to consider how to facilitate fluent and automatic (i.e., implicit)
knowledge. Future research should attempt to develop and validate other tests
of implicit lexical knowledge (eye tracking, ERP, among other behavioral, free
constructed, tasks) to inform the explicit/implicit debate. Another interesting
line for future research concerns how beginning and intermediate English-as-
a-foreign-language learners (as opposed to the advanced ESL learners inves-
tigated here) develop implicit, in addition to explicit, collocational knowledge
in language classrooms. Our results, along with Webb and Kagimoto (2009),
have shown proficiency as a major predictor of performance on explicit tests
of collocational knowledge. It is interesting to see what would happen when
implicit collocational knowledge is assessed.
The nature of explicit and implicit lexical knowledge and the interrelation-
ship between them is an emerging lexical issue that is likely to attract increased
attention in the future. This study is a start in exploring this topic, but can
only be considered a first step. It will surely raise more questions than it has
answered, but hopefully it has at least begun the discussion.
Revised version accepted 20 December 2011

Notes
1 (i) Blakiston’s Gould medical dictionary (Gould & Gennaro, 1979), (ii) Collins
dictionary [of] medicine (Youngson, 2004), (iii) Mosby’s medical, nursing, and
allied health dictionary (Anderson, Keith, & Novak, 2002), (iv) Stedman’s medical
terms and phrases (Stedman, 2004), and (v) Stedman’s medical dictionary for the
health professions and nursing (Stedman, 2005).
2 We tried our best to achieve McNamara’s (2005) recommended ratios for automatic
priming. Relatedness proportion (RP) was close to the 0.20 value (the number of
related trials was 7 or 8 out of all 37 word-word trials (7 or 8/37 ≈ 0.20). As for
nonword ratio (NR), there were 31 nonword trials out of all the 60 or 61 unrelated
trials resulting in the acceptable NR of ≈ 0.50 (31/60 or 61).
3 When categorical variables are entered as predictors in a logit mixed model or a
LME model, the levels are ordered alphabetically and the first is treated as the
reference level (e.g., control is the reference level for the CONDITION variable
here). Thus, when any level is found to be significant, this would mean that it is
significant in comparison to the reference level.
4 For the implicit, priming test, responses shorter than 150 ms or longer than 1300 ms
(constituting only 2.15% of all data points) were excluded from further analysis. The
long cut-off time in the present priming task is far higher than that used in previous
research. However, it should be noted that words used in the present lexical decision

Language Learning 63:1, March 2013, pp. 121–159 154

task are rather low-frequency (see Appendix S1 in the Supporting Information) and
would be processed slower than those included in previous research.
5 Because immediate tests can only inform about short-term effects, we prefer to
interpret immediate gains as short-term enhancement effects. We reserve the notion
of “learning” for gains on delayed tests, as only they can demonstrate acquisition
that is durable (see Schmitt, 2010, for more on this).
6 For both experiments, only correct responses were included in the final reaction time
analysis.
7 The mean reading-comprehension score gained by the 34 participants in the
experimental group was acceptable. All participants scored higher than 60% in this
test (Mean = 6.88 out of 8.00, 86.03%, SD = 0.84).
8 Only 1.09% of the data points coded as CORRECT in the explicit recall test
represent partial, rather than complete, knowledge in Experiment 2.
9 This analysis was run in a model including the CONDITION × SESSION
interaction only (excluding all other significant predictors) in order to catch all
possible contrasts for this interaction.

References
Aitchison, J. (2003). Words in the mind: An introduction to the mental lexicon (3rd
ed.). Oxford, UK: Blackwell.
Anderson, D. M., Keith, J., & Novak, P. D. (2002). Mosby’s medical, nursing, and
allied health dictionary (6th ed.). London: Mosby.
Andringa, S., de Glopper, K., & Hacquebord, H. (2011). Effect of explicit and implicit
instruction on free written response task performance. Language Learning, 61,
868–903.
Baayen, R. H. (2008). Analyzing linguistic data: A practical introduction to statistics
using R. Cambridge, UK: Cambridge University Press.
Bialystok, E. (1981). Some evidence for the integrity and interaction of two knowledge
sources. In R. Anderson (Ed.), New dimensions in research on the acquisition and
use of a second language (pp. 62–74). Rowley, MA: Newbury House.
Bishop, H. (2004). Noticing formulaic sequences—A problem of measuring the
subjective. LSO Working Papers in Linguistics, 4, 15–19.
Bretz, F., Hothorn, T., & Westfall, P. (2010). Multiple comparisons using R. Boca
Raton, FL: Chapman & Hall/CRC.
Dagenbach, D., Horst, S., & Carr, T. H. (1990). Adding new information to semantic
memory: How much learning is enough to produce automatic priming? Journal of
Experimental Psychology: Learning, Memory, and Cognition, 16, 581–591.
Davies, M. (2007). VIEW: Variation in English words and phrases. Retrieved July 22,
2011, from http://corpus.byu.edu/bnc/
de Graaff, R. (1997). The eXperanto experiments: Effects of explicit instruction on
second language acquisition. Studies in Second Language Acquisition, 19, 249–276.

155 Language Learning 63:1, March 2013, pp. 121–159

DeKeyser, R. (1994). Implicit and explicit learning of L2 grammar: A pilot study.

TESOL Quarterly, 28, 188–194.
Doughty, C. J. (2003). Instructed SLA: Constraints, compensation, and enhancement.
In C. J. Doughty & M. H. Long (Eds.), The handbook of second language
acquisition (pp. 256–310). Malden, MA: Blackwell.
Durrant, P., & Doherty, A. (2010). Are high-frequency collocations psychologically
real? Investigating the thesis of collocational priming. Corpus Linguistics and
Linguistic Theory, 6, 125–155.
Durrant, P., & Schmitt, N. (2010). Adult learners’ retention of collocations from
exposure. Second Language Research, 26, 163–188.
Elgort, I. (2011). Deliberate learning and vocabulary acquisition in a second language.
Language Learning, 61, 367–413.
Ellis, N. C. (1993). Rules and instances in foreign language learning: Interactions of
explicit and implicit. European Journal of Cognitive Psychology, 5, 289–318.
Ellis, N. C. (1994). Consciousness in second language learning: Psychological
perspectives on the role of conscious processes in vocabulary acquisition. AILA
Review, 11, 37–56.
Ellis, N. C. (2002). Frequency effects in language processing. Studies in Second
Language Acquisition, 24, 143–188.
Ellis, N. C. (2008). Implicit and explicit knowledge about language. In J. Cenoz & N.
H. Hornberger (Eds.), Encyclopedia of language and education (2nd ed.). Volume 6:
Knowledge about language (pp. 119–132). New York: Springer.
Ellis, N. C., Frey, E., & Jalkanen, I. (2009). Psycholinguistic reality of collocation and
semantic prosody (1): Lexical access. In U. Römer & R. Schulze (Eds.), Exploring
the lexis-grammar interface (pp. 89–114). Amsterdam: John Benjamins.
Ellis, N. C., & Laporte, N. (1997). Contexts of acquisition: Effects of formal
instruction and naturalistic exposure on second language acquisition. In A. M. B. de
Groot & J. F. Kroll (Eds.), Tutorials in bilingualism: Psycholinguistic perspectives
(pp. 89–113). Mahwah, NJ: Erlbaum.
Ellis, R. (1994). Implicit/explicit knowledge and language pedagogy. TESOL
Quarterly, 28, 166–172.
Ellis, R. (2002). Does form-focused instruction affect the acquisition of implicit
knowledge? Studies in Second Language Acquisition, 24, 223–236.
Ellis, R. (2004). The definition and measurement of L2 explicit knowledge. Language
Learning, 54, 227–275.
Ellis, R. (2005). Measuring implicit and explicit knowledge of a second language:
A psychometric study. Studies in Second Language Acquisition, 27, 141–172.
Ellis, R. (2006). Current issues in the teaching of grammar: An SLA perspective.
TESOL Quarterly, 40, 83–107.
Ellis, R. (2007). Explicit form-focused instruction and second language acquisition. In
B. Spolsky & F. M. Hult (Eds.), The handbook of educational linguistics (pp.
437–455). Malden, MA: Wiley-Blackwell.

Language Learning 63:1, March 2013, pp. 121–159 156

Ellis, R. (2009). Implicit and explicit learning, knowledge, and instruction. In R. Ellis
et al. (Eds.), Explicit and implicit knowledge in second language learning, testing
and teaching (pp. 3–25). Bristol, UK: Multilingual Matters.
Gagné, C. L., & Spalding, T. L. (2009). Constituent integration during the processing
of compound words: Does it involve the use of relational structures? Journal of
Memory and Language, 60, 20–35.
Gould, G. M., & Gennaro, A. R. (1979). Blakiston’s Gould medical dictionary: A
modern comprehensive dictionary of the terms used in all branches of medicine and
allied sciences, with illustrations and tables (4th ed.). New York/London:
McGraw-Hill.
Hoey, M. (1991). Patterns of lexis in text. Oxford, UK: Oxford University Press.
Hoey, M. (2005). Lexical priming: A new theory of words and language. London:
Routledge.
Hulstijn, J. H. (2007). Psycholinguistic perspectives on language and its acquisition. In
J. Cummins & C. Davison (Eds.), The international handbook of English language
teaching (pp. 783–796). Norwell, MA: Springer.
Hulstijn, J. H., & de Graaff, R. (1994). Under what conditions does explicit knowledge
of a second language facilitate the acquisition of implicit knowledge? AILA Review,
11, 97–112.
Jaeger, T. F. (2008). Categorical data analysis: Away from ANOVAs (transformation or
not) and towards logit mixed models. Journal of Memory and Language, 59,
434–446.
Krashen, S. D. (1981). Second language acquisition and second language learning.
London: Pergamon Press.
Laufer, B., & Girsai, N. (2008). Form-focused instruction in second language
vocabulary learning: A case for contrastive analysis and translation. Applied
Linguistics, 29, 694–716.
Leech, G., Rayson, P., & Wilson, A. (2001). Word frequencies in written and spoken
English: Based on the British National Corpus. London: Longman.
Long, M. H. (1983). Does second language instruction make a difference? A review of
the research. TESOL Quarterly, 17, 359–382.
Mackey, A., & Goo, J. (2007). Interaction research in SLA: A meta-analysis and
research synthesis. In A. Mackey (Ed.), Conversational interaction in second
language acquisition: A series of empirical studies (pp. 407–452). Oxford, UK:
Oxford University Press.
McKoon, G., & Ratcliff, R. (1992). Spreading activation versus compound cue
accounts of priming: Mediated priming revisited. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 18, 1155–1172.
McLaughlin, B., Rossman, T., & McLeod, B. (1983). Second language learning: An
information-processing perspective. Language Learning, 33, 135–158.
McLaughlin, J., Osterhout, L., & Kim, A. (2004). Neural correlates of second
language word learning: Minimal instruction produces rapid change. Nature
Neuroscience, 7, 703–704.
157 Language Learning 63:1, March 2013, pp. 121–159
14679922, 2013, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1467-9922.2012.00730.x by Pontificia Universidad Catolica De Chile, Wiley Online Library on [20/08/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Sonbul and Schmitt Explicit and Implicit Lexical Knowledge

McNamara, T. P. (2005). Semantic priming: Perspectives from memory and word

recognition. New York: Psychology Press.
Nation, I. S. P. (1990). Teaching and learning vocabulary. New York: Newbury House.
Nation, I. S. P. (2001). Learning vocabulary in another language. Cambridge, UK:
Cambridge University Press.
Norris, J. M., & Ortega, L. (2000). Effectiveness of L2 instruction: A research
synthesis and quantitative meta-analysis. Language Learning, 50, 417–528.
Osterhout, L., McLaughlin, J., Pitkänen, I., Frenck-Mestre, C., & Molinaro, N. (2006).
Novice learners, longitudinal designs, and event-related potentials: A means for
exploring the neurocognition of second language processing. Language Learning,
56, Supplement 1, 199–230.
Paradis, M. (1994). Neurolinguistic aspects of implicit and explicit memory:
Implications for bilingualism. In N. C. Ellis (Ed.), Implicit and explicit learning of
languages (pp. 393–419). London: Academic Press.
R Development Core Team. (2010). R: A language and environment for statistical
computing. Vienna, Austria: R Foundation for Statistical Computing.
Rastle, K., Harrington, J., & Coltheart, M. (2002). 358,534 Nonwords: The ARC
nonword database. The Quarterly Journal of Experimental Psychology, 55A,
1339–1362.
Read, J. (2000). Assessing vocabulary. Cambridge, UK: Cambridge University Press.
Reinders, H., & Ellis, R. (2009). The effects of two types of input on intake and the
acquisition of implicit and explicit knowledge. In R. Ellis et al. (Eds.), Explicit and
implicit knowledge in second language learning, testing and teaching (pp.
281–302). Bristol, UK: Multilingual Matters.
Rott, S. (2009). The effect of awareness-raising on the use of formulaic constructions.
In R. Corrigan, E. Moravcsik, H. Ouali, & K. Wheatley (Eds.), Formulaic language
(pp. 405–422). Amsterdam: John Benjamins.
Schmitt, N. (2008). Instructed second language vocabulary learning. Language
Teaching Research, 12, 329–363.
Schmitt, N. (2010). Researching vocabulary: A vocabulary research manual.
Basingstoke, UK: Palgrave Press.
Schmitt, N., Dornyei, Z., Adolphs, S., & Durow, V. (2004). Knowledge and acquisition
of formulaic sequences: A longitudinal study. In N. Schmitt (Ed.), Formulaic
sequences: Acquisition, processing, and use (pp. 55–86). Amsterdam: John
Benjamins.
Siyanova, A., & Schmitt, N. (2008). L2 learner production and processing of
collocation: A multi-study perspective. Canadian Modern Language Review, 64,
429–458.
Spada, N., & Tomita, Y. (2010). Interactions between type of instruction and type of
language feature: A meta analysis. Language Learning, 60, 263–308.
Squire, L. R. (2004). Memory systems of the brain: A brief history and current
perspective. Neurobiology of Learning & Memory, 82, 171–177.

Language Learning 63:1, March 2013, pp. 121–159 158

Stedman, T. L. (2004). Stedman’s medical terms and phrases. London: Lippincott

Williams & Wilkins.
Stedman, T. L. (2005). Stedman’s medical dictionary for the health professions and
nursing (Illustrated 5th ed.). London: Lippincott Williams & Wilkins.
Tiersky, E., & Tiersky, M. (1992). The language of medicine in English (2nd ed.).
Englewood Cliffs, NJ: Prentice Hall Regents.
Tokowicz, N., & MacWhinney, B. (2005). Implicit and explicit measures of sensitivity
to violations in second language grammar: An event-related potential investigation.
Studies in Second Language Acquisition, 27, 173–204.
Ullman, M. T. (2001). The neural basis of lexicon and grammar in first and second
language: The declarative/procedural model. Bilingualism: Language and
Cognition, 4, 105–122.
Webb, S., & Kagimoto, E. (2009). The effects of vocabulary learning on collocation
and meaning. TESOL Quarterly, 43, 55–77.
West, M. (1953). A general service list of English words. London: Longman.
White, J., & Ranta, L. (2002). Examining the interface between metalinguistic task
performance and oral production in a second language. Language Awareness, 11,
259–290.
Wolter, B., & Gyllstad, H. (2011). Collocational links in the L2 mental lexicon and the
influence of L1 intralexical knowledge. Applied Linguistics, 32, 430–449.
Youngson, R. M. (2004). Collins dictionary [of] medicine (3rd ed.). Glasgow:
HarperCollins.

Supporting Information
Additional Supporting Information may be found in the online version of this
article at the publisher’s website:

Appendix S1. Intact Medical Collocations and their Controls (Experiment 1).
Appendix S2. Intact Medical Collocations and their Controls (Experiment 2).
Appendix S3. Mixed-Effects Modeling Procedures in the Study.

159 Language Learning 63:1, March 2013, pp. 121–159

Mostafa - Kim 2020 Input-Output Instruction Automatized Explicit Knowledge
No ratings yet
Mostafa - Kim 2020 Input-Output Instruction Automatized Explicit Knowledge
26 pages
Critical Heart in Children and Infants 2019 PDF
100% (3)
Critical Heart in Children and Infants 2019 PDF
1,184 pages
Imp Exp Chap 07
No ratings yet
Imp Exp Chap 07
88 pages
Implicit and Explicit Learning
No ratings yet
Implicit and Explicit Learning
46 pages
Explicit Explanation
No ratings yet
Explicit Explanation
31 pages
Han and Finneran - The Interface Debate
No ratings yet
Han and Finneran - The Interface Debate
20 pages
Meth Summaries
No ratings yet
Meth Summaries
26 pages
Implicit&Explicit Learning
No ratings yet
Implicit&Explicit Learning
36 pages
Akakura 2012
No ratings yet
Akakura 2012
29 pages
Applied Linguistics-2014-Fordyce-6-28
No ratings yet
Applied Linguistics-2014-Fordyce-6-28
23 pages
Language Learning - 2023 - Schurz - Measuring Teenage Learners Automatized Explicit and or Implicit Knowledge A
No ratings yet
Language Learning - 2023 - Schurz - Measuring Teenage Learners Automatized Explicit and or Implicit Knowledge A
36 pages
Ellis 1994
No ratings yet
Ellis 1994
8 pages
The Measurement of Implicit and Explicit Knowledge: The Language Learning Journal
No ratings yet
The Measurement of Implicit and Explicit Knowledge: The Language Learning Journal
17 pages
Spada Et Al-2010-Language Learning
No ratings yet
Spada Et Al-2010-Language Learning
46 pages
Sample Template For Comparison
No ratings yet
Sample Template For Comparison
4 pages
LHE3221 - Weeks 2 and 3
No ratings yet
LHE3221 - Weeks 2 and 3
35 pages
1 s2.0 S0346251X13001437 Main
No ratings yet
1 s2.0 S0346251X13001437 Main
20 pages
At The Interface Dynamic Interactions of Explicit and Implicit Language Knowledge - Ellis
No ratings yet
At The Interface Dynamic Interactions of Explicit and Implicit Language Knowledge - Ellis
37 pages
Ellis 2005
No ratings yet
Ellis 2005
48 pages
Explicit Versus Implicit Grammar
No ratings yet
Explicit Versus Implicit Grammar
6 pages
Explicit Implicit Interface Summary
No ratings yet
Explicit Implicit Interface Summary
9 pages
Analyzed Knowledge, Metalanguage, and Second Language Proficiency
No ratings yet
Analyzed Knowledge, Metalanguage, and Second Language Proficiency
13 pages
Andringa2015 3.1
No ratings yet
Andringa2015 3.1
12 pages
Lichtman (2020)
No ratings yet
Lichtman (2020)
24 pages
Automacity in Second Language Vocabuary Learning
No ratings yet
Automacity in Second Language Vocabuary Learning
10 pages
2 Metalinguistic Chinese Researcher
No ratings yet
2 Metalinguistic Chinese Researcher
15 pages
The Effectiveness of Implicit and Explicit Cognitive Processing in Incidental Vocabulary Acquisition
No ratings yet
The Effectiveness of Implicit and Explicit Cognitive Processing in Incidental Vocabulary Acquisition
15 pages
TEL - Volume 5 - Issue 1 - Pages 45-71
No ratings yet
TEL - Volume 5 - Issue 1 - Pages 45-71
27 pages
Explicit, Implicit Knowledge
No ratings yet
Explicit, Implicit Knowledge
28 pages
1st Guide - Methodology
No ratings yet
1st Guide - Methodology
2 pages
Evaluating The Effectiveness of Explicit Instruction On Implicit and Explicit L2 Knowledge
No ratings yet
Evaluating The Effectiveness of Explicit Instruction On Implicit and Explicit L2 Knowledge
29 pages
Implicit and Explicit Learning of Languages Nick Ellis Ed London Academic Press 1994 PP Vii 599 5000 Cloth
No ratings yet
Implicit and Explicit Learning of Languages Nick Ellis Ed London Academic Press 1994 PP Vii 599 5000 Cloth
3 pages
Does Intensive Explicit Grammarinstruction Make All The Difference
No ratings yet
Does Intensive Explicit Grammarinstruction Make All The Difference
31 pages
Learning Second Language Grammar Rules: An Experiment With A Miniature Linguistic System
No ratings yet
Learning Second Language Grammar Rules: An Experiment With A Miniature Linguistic System
32 pages
Task-Based Language Teaching For Beginner-Level-WT - Summaries
No ratings yet
Task-Based Language Teaching For Beginner-Level-WT - Summaries
6 pages
Measuring Implicit and
No ratings yet
Measuring Implicit and
33 pages
Instructed Grammar Byrd
No ratings yet
Instructed Grammar Byrd
18 pages
Incidental Development of Implicit and Explicit Grammar Knowledge
No ratings yet
Incidental Development of Implicit and Explicit Grammar Knowledge
10 pages
The Role of Explicit Knowledge, Noticing, and Attention in L2 Writing and Error Correction: Quantitative and Qualitative Analyses
No ratings yet
The Role of Explicit Knowledge, Noticing, and Attention in L2 Writing and Error Correction: Quantitative and Qualitative Analyses
22 pages
Explicit and Implicit Grammar Teaching: Prepared By: Josephine Gesim & Jennifer Marcos
100% (1)
Explicit and Implicit Grammar Teaching: Prepared By: Josephine Gesim & Jennifer Marcos
17 pages
An Inductive Approach To Young Learner Grammar: David E. Shaffer (Chosun University) Disin@chosun - Ac.kr
No ratings yet
An Inductive Approach To Young Learner Grammar: David E. Shaffer (Chosun University) Disin@chosun - Ac.kr
8 pages
Implicit and Explicit Learning
No ratings yet
Implicit and Explicit Learning
46 pages
Explicit Grammar Rules and
No ratings yet
Explicit Grammar Rules and
11 pages
Macaro Masterman Does Explicit Instruction Make All The Difference
No ratings yet
Macaro Masterman Does Explicit Instruction Make All The Difference
32 pages
Rod Ellis - The Structural Syllabus and Language Acquisition PDF
50% (2)
Rod Ellis - The Structural Syllabus and Language Acquisition PDF
24 pages
Topic 4 Second Language Learning 2
No ratings yet
Topic 4 Second Language Learning 2
50 pages
A Case For Explicit Grammar Instruction
No ratings yet
A Case For Explicit Grammar Instruction
10 pages
Instruction-English-Language-Essay - PHP: Wang, D. (2000) - Vocabulary Acquisition: Implicit Learning and Explicit Teaching
No ratings yet
Instruction-English-Language-Essay - PHP: Wang, D. (2000) - Vocabulary Acquisition: Implicit Learning and Explicit Teaching
6 pages
Punished by Rewards
100% (1)
Punished by Rewards
48 pages
Explicit Grammar and Implicit Grammar Teaching
100% (1)
Explicit Grammar and Implicit Grammar Teaching
5 pages
Green Hecht 1992 Explicit Instruction
No ratings yet
Green Hecht 1992 Explicit Instruction
17 pages
Mergers and Acquisitions
No ratings yet
Mergers and Acquisitions
123 pages
Fotos and Ellis - Communicating ABout Grammar A Task Based Approach PDF
0% (1)
Fotos and Ellis - Communicating ABout Grammar A Task Based Approach PDF
25 pages
Intensive Grammar Teaching
No ratings yet
Intensive Grammar Teaching
32 pages
Waiver of Rights
100% (10)
Waiver of Rights
2 pages
Mitosis
No ratings yet
Mitosis
15 pages
Secondary Market DR S Sreenivasa Murthy
No ratings yet
Secondary Market DR S Sreenivasa Murthy
33 pages
Noticing Noonan
No ratings yet
Noticing Noonan
6 pages
Oracle HCM Cloud Training Outline
No ratings yet
Oracle HCM Cloud Training Outline
6 pages
Early On Kenpo History
No ratings yet
Early On Kenpo History
4 pages
2019 Winter Model Answer Paper
No ratings yet
2019 Winter Model Answer Paper
21 pages
Noam Chomsky - The 10 Media Manipulation Strategies
100% (2)
Noam Chomsky - The 10 Media Manipulation Strategies
4 pages
Project Insight - Cipla LTD
No ratings yet
Project Insight - Cipla LTD
14 pages
MCA 301 Data Mining Notes
No ratings yet
MCA 301 Data Mining Notes
6 pages
Success Against The Odds
No ratings yet
Success Against The Odds
194 pages
Ex - Mayor Sanchez
No ratings yet
Ex - Mayor Sanchez
3 pages
The Beginnings of British Literature Old English Anglo-Saxon and Medieval Literature
No ratings yet
The Beginnings of British Literature Old English Anglo-Saxon and Medieval Literature
108 pages
793 Utility Software
No ratings yet
793 Utility Software
126 pages
STERN White Paper 2017-03 Withcover (1) - 0
No ratings yet
STERN White Paper 2017-03 Withcover (1) - 0
270 pages
Towards A Third Food Regime: Behind The Transformation
No ratings yet
Towards A Third Food Regime: Behind The Transformation
13 pages
( Ybm ( ) ) - 1 1
No ratings yet
( Ybm ( ) ) - 1 1
13 pages
Inbound 7418254903065815207
No ratings yet
Inbound 7418254903065815207
78 pages
L4C1 Examiner Report March 2022
No ratings yet
L4C1 Examiner Report March 2022
7 pages
DSCP & Vlan Priority
No ratings yet
DSCP & Vlan Priority
13 pages
Rational of The Body Shop - Project 2 (Branded Interactions)
No ratings yet
Rational of The Body Shop - Project 2 (Branded Interactions)
5 pages
Final Test: With Answer Key
No ratings yet
Final Test: With Answer Key
8 pages
317-322, Published June 08, 2024
No ratings yet
317-322, Published June 08, 2024
6 pages
Behaviour Management of An Anxious Child
No ratings yet
Behaviour Management of An Anxious Child
5 pages
Villarba Vs Court of Appeals
No ratings yet
Villarba Vs Court of Appeals
16 pages
Gcse Ict: by The End of This Session, You Will Be Able To
No ratings yet
Gcse Ict: by The End of This Session, You Will Be Able To
10 pages
OECD Work Education Skills Policy Products Services For Countries
No ratings yet
OECD Work Education Skills Policy Products Services For Countries
9 pages
SCFR1 JHS Currhead Consolidation
No ratings yet
SCFR1 JHS Currhead Consolidation
2 pages
Vocabulary Learning Strategies and Foreign Language Acquisition
From Everand
Vocabulary Learning Strategies and Foreign Language Acquisition
Višnja Pavičić Takač
2/5 (8)
Social research applied to english language teaching in Colombian contexts: Theory and Methods
From Everand
Social research applied to english language teaching in Colombian contexts: Theory and Methods
Wilder Yesid Escobar Alméciga
5/5 (1)
Insights into Autonomy and Technology in Language Teaching
From Everand
Insights into Autonomy and Technology in Language Teaching
Chun Lai
No ratings yet
Children talking about talking: The reflexive emergence of language
From Everand
Children talking about talking: The reflexive emergence of language
Talbot J. Taylor
No ratings yet
Investigating Tasks in Formal Language Learning
From Everand
Investigating Tasks in Formal Language Learning
María del Pilar García Mayo
No ratings yet
Studies of Fossilization in Second Language Acquisition
From Everand
Studies of Fossilization in Second Language Acquisition
Nancy M. P. King
No ratings yet
Third or Additional Language Acquisition
From Everand
Third or Additional Language Acquisition
Gessica De Angelis
No ratings yet
Fossilization in Adult Second Language Acquisition
From Everand
Fossilization in Adult Second Language Acquisition
Nancy M. P. King
No ratings yet