The learning burden of collocations: The role of interlexical and intralexical factors

Abstract

This study investigates whether congruency (+/– literal translation equivalent), collocate–node relationship (adjective–noun, verb–noun, phrasal-verb–noun collocations), and word length influence the learning burden of EFL learners’ learning collocations at the initial stage of form–meaning mapping. Eighteen collocations were selected on the basis of a pretest. They were divided into 9 congruent and 9 incongruent collocations and into 6 verb–noun collocations, 6 phrasal-verb–noun collocations and 6 adjective–noun collocations. Forty-one EFL learners (first language: Dutch) were asked to read a word list containing the 18 target collocations, their translation and a sample sentence and to complete four online exercises, in which the 18 collocations were presented twice. Learning gains were measured at three levels of sensitivity: form recall test 1, form recall test 2 (+ clue), form recognition test. Although mixed findings were revealed, all factors seemed to affect the learning difficultly of the collocations. Incongruent collocations appeared to be more difficult to recall than congruent ones. Adjective–noun collocations were better recalled and recognized than (phrasal) verb–noun collocations. Depending on the posttest, participants’ vocabulary size and word length of the individual constituents making up the collocation also affected the learning process.

Keywords

Collocate–node relationship collocations congruency EFL lexicon vocabulary

I Introduction

Foreign language (FL) learners are faced with the enormous task of acquiring a large vocabulary. One of the biggest lexical challenges is the acquisition of collocations (e.g. to throw a party, gross misconduct). However, analyses of learner corpora have revealed that FL learners tend to make collocation errors, of which many are influenced by their first language (L1). Although corpus studies (Laufer & Waldman, 2011; Nesselhauf, 2003; Wolter & Gyllstad, 2013) have revealed that two types of collocations – viz. incongruent (no literal L1 translation equivalent) and verb–noun collocations – present a particular challenge to FL learners, no study has investigated empirically whether congruency (= presence or absence of literal L1 equivalent) and the type of collocation (e.g. verb–noun compared to adjective–noun collocations) also play a role in the initial form–meaning mapping of unfamiliar collocations. In addition, little is known about the role of other intralexical factors such as word length. Therefore, this study sets out to explore whether congruency and the intralexical factors collocation type (= collocate–node relationship) and word length affect the likelihood that learners of English as a foreign language (EFL) will learn a new collocation. In particular, the focus is on EFL learners’ initial form–meaning mapping of new collocations in a vocabulary-focused learning session.

II Background

1 Interlexical factors: L1 influence and congruency

Learner corpus studies have shown the influence of congruency on learners’ use of collocations. Congruency is defined as the presence or absence of a literal L1 translation equivalent (Nesselhauf, 2003). An example of a congruent collocation for Dutch-speaking EFL learners would be to close a deal, an example of an incongruent collocation would be to make an effort because a word-for-word translation from Dutch would be to do an effort.

Nesselhauf (2003) investigated the use of English verb–noun collocations by German-speaking university students. She found that 56% of the collocation errors were L1-based, e.g. make homework (‘do homework’ Hausaufgaben machen) or close lacks (‘close gaps’ Lücken schlieβen). A comparison of congruent and incongruent collocations clearly showed that second language (L2) learners produced far more infelicitous incongruent than congruent collocations. In another study, Laufer and Waldman (2011) analysed the use of English verb–noun collocations by L1 Hebrew speakers at three proficiency levels. Their findings showed that L2 learners at all levels produced errors and that L1-induced errors in particular continued to persist at the most advanced level. Finally, Wolter and Gyllstad (2011, 2013) also found that congruency played a role in L1 Swedish EFL learners’ processing collocations as measured by their reaction times in a primed lexical decision task. In addition, their results revealed that these learners were familiar with more congruent than incongruent collocations and that they produced more errors on the incongruent than on the congruent collocations.

One explanation for FL learners’ difficulty in using L2 collocations correctly might be that FL learners do not notice the formal differences between the L1 and L2 collocation when encountering it in the input because the collocation’s meaning is transparent (Laufer & Girsai, 2008; Laufer & Waldman, 2011; Peters, 2012). As a result, such semantically transparent collocations do not tend to cause comprehension problems but they might lead to problems in the production process (Laufer & Girsai, 2008; Laufer & Waldman, 2011). For instance, Dutch-speaking learners of English will easily understand the semantically transparent but incongruent collocation to make an effort. However, they might not notice that English has the verb to make in this collocation and might use to do an effort because they rely on their L1.

Another explanation suggests that learners tend to operate on the open-choice principle by selecting and combining single words that do not always go together instead of operating on the idiom principle (Laufer & Waldman, 2011; Sinclair, 1991; Wray, 2002). Laufer (2011) argues that FL learners who produce such incorrect combinations have fallen ‘into the trap of … deceptive compatibility’ (p. 44), which means that they are unaware of the combinatorial restrictions of words in the L2. A similar line of thought was developed by Wolter (2006) and Wolter and Gyllstad (2011, 2013), who argue that the acquisition of congruent collocations differs from the acquisition of incongruent collocations because in the latter case FL learners cannot rely on the L1 pattern of single words making the learning process of incongruent collocations more complex. They hypothesize that learners’ L1 may have a facilitative effect on the acquisition of congruent collocations but an inhibitive effect on the acquisition of incongruent collocations. However, this hypothesis has not yet been empirically tested.

2 Intralexical factors: Word length and collocate–node relationship

In addition to interlexical factors, the learning burden of a lexical item can also be influenced by intralexical factors, such as pronounceability, orthography, morphology, synformy and semantic features (abstractness, register, idiomaticity, polysemy) (Laufer, 1997). Two factors will be discussed in more detail, viz. word length on the one hand and the collocate–node relationship on the other.

Mixed findings have been reported with regard to the effect of word length on learning single words. Although some studies have revealed that longer words are more difficult to learn (e.g. Campoy, 2008; Hulme, Maughan, & Brown, 1991), Laufer (1997) stresses that it is anything but straightforward to disentangle word length from other factors, such as a word’s morphological transparency or a learner’s exposure to a word. Similarly, Milton (2009) argues that it is difficult to isolate word length from a word’s frequency since the most frequent words tend to be shorter ones as well. Nevertheless, he acknowledges that word length may still operate at the level of individual words. Nation and Webb (2011) argue that longer words are more difficult because there is ‘more to remember in long words than in short words’ (p. 315).

Verb–noun collocations seem to be the most frequently studied collocation type (Paquot & Granger, 2012). Results from studies by Nesselhauf (2003) and Laufer and Waldman (2011) revealed that FL learners seem to struggle most with selecting the correct verb, e.g. make one’s homework. Similarly, Waibel (2007) found that both German and Italian EFL learners produced infelicitous combinations with phrasal verbs, e.g. to carry out power or to carry out a fight. Two hypotheses might explain why EFL learners struggle with verbs in collocations. First, as Boers, Demecheleer, Coxhead, and Webb (2014a) argue, many collocations contain delexicalized verbs (do an exercise, have a shower) which do not carry the semantic weight of the collocation. Consequently, FL learners might not allocate sufficient attentional resources to the verb when they encounter verb–noun collocations. Second, (phrasal) verb–noun collocations show more variation in morphology compared to adverb–adjective or adjective–noun collocations because of inflections such as number, tense, aspect and person. Laufer (2011) also pointed out that one of the difficulties FL learners face when consulting learner dictionaries is that the verb–noun collocation can appear in a different form in the example given. Laufer illustrates this with the collocation to take measures, which appears as They took strong measures against dangerous drivers in one dictionary and as Measures are being taken to reduce crime in the city in another. Both examples show that the collocation appears with the verb in the past (took), in the passive (are being taken) and with the additional adjective strong in the first example.

Phrasal verbs have been shown to be particularly difficult for EFL learners because they consist of a main verb and at least one particle (e.g. bring up, turn down) (Paquot & Granger, 2012). Moreover, phrasal verbs are often not transparent (Schmitt & Redwood, 2011; Siyanova & Schmitt, 2007). Dagut and Laufer (1985) as well as Siyanova and Schmitt (2007) demonstrated that EFL learners tended to avoid phrasal verbs and preferred one-word verbs instead. According to Siyanova and Schmitt, the use of fewer phrasal verbs might reflect the input these L2 learners were receiving since their corpus data revealed that one-word verbs were more frequent compared to phrasal verbs. Other learner variables, such as learners’ L1 (Germanic vs. non-Germanic L1), learners’ proficiency level and out-of-class exposure on the one hand, and the phrasal verb-type (+/– transparent) on the other, can also affect learners’ avoidance of multiword verbs to different degrees (Hulstijn & Marchena, 1989; Liao & Fukuya, 2004; Schmitt & Redwood, 2011). Finally, Liao and Fukuya (2004) showed that participants were more likely to resort to one-word verbs in translation tests than in multiple choice and recall tests.

3 Facilitative factors for learning of collocations

Recently, more studies have also started to explore the acquisition process of collocations. The following factors have been found to boost collocational learning (for an exhaustive review of intervention studies on formulaic sequences, see Boers & Lindstromberg, 2012): visual salience of collocations in the input (Peters, 2012; Sonbul & Schmitt, 2013), explicit contrastive analysis (Laufer & Girsai, 2008), frequency of occurrence (Peters, 2014; Webb, Newton & Chang, 2013), alliteration (i.e. alliterative collocations compared to non-alliterative collocations) (Boers, Lindstromberg, & Webb, 2014b), provision of a dictionary during writing activities (Laufer, 2011), provision of multiple collocations for a limited set of words instead of the same number of unrelated collocations (Webb & Kagimoto, 2011), and avoidance of synonymy when presenting collocations (Webb & Kagimoto, 2011). Unlike studies on single word learning, studies into the acquisition process of collocations have not explicitly addressed the question of whether learners’ vocabulary size plays a mediating role in the learning process. Research on single word learning suggests that there tends to be a positive relationship between learners’ vocabulary size and their learning gains (Montero Perez, Peters, Clarebout, & Desmet, 2014; Zahar, Cobb, & Spada, 2001). However, it is not clear whether this is also the case for collocation learning.

III Aim and research questions

As has become clear in the literature survey, previous studies have shown the influence of congruency on FL learners’ use of collocations and their difficulties in using L2 verb–noun collocations. However, no research has been found that investigated how congruency and intralexical factors such as collocate–node relationship and word length affect the learning difficulty of new collocations at the initial stage of form–meaning mapping. Research on FL learners’ initial form–meaning mapping of new collocations is necessary because it can help us gain a better understanding of the learning difficulty of different types of collocations at the initial stages of the acquisition process.

The first purpose of the present study is to investigate whether congruency affects learners’ initial form–meaning mapping of new collocations. The second purpose is concerned with the type of collocation, viz. the collocate–node relationship, in collocational learning. Specifically, three types of collocations are compared: adjective–noun, verb–noun and phrasal-verb–noun collocations. Third, the study seeks to shed some light on the influence of word length and learners’ vocabulary size in learning collocations.

This study addresses the following research questions.

Does congruency affect the odds that a learner will recall and recognize the form of a new collocation after deliberate study activities?

Does the collocate–node relationship affect the odds that a learner will recall and recognize the form of a new collocation after deliberate study activities?

Does the interaction between congruency and the collocate–node relationship affect the odds that a learner will recall and recognize the form of a new collocation after deliberate study activities?

Does word length of the individual words affect the odds that a learner will recall and recognize the form of a new collocation after deliberate study activities?

Does learners’ vocabulary size affect the odds that a learner will recall and recognize the form of a new collocation after deliberate study activities?

IV Method

To answer the research questions, a classroom-based experiment was set up in which 18 target collocations, taken from students’ course materials, were selected on the basis of a pretest. The collocations were first presented in a word list that contained a translation and a sample sentence for each collocation and were subsequently practiced in online, vocabulary-focused activities. Learning gains were then measured in two form recall tests and one form recognition test.

1 Participants

The participants were 43 Dutch-speaking EFL learners recruited from six parallel first-year business English classes at a Flemish university. All first-year Business Studies students have a compulsory Business English course in their program. Participants had studied English for at least four years prior to the treatment. Most students, however, had had five to six years of formal instruction. Except for one participant,¹ all participants were familiar with the 2,000 most frequent words in English, as measured by the Vocabulary Levels Test (VLT) (Schmitt, Schmitt & Clapham, 2001). The average score was 27.95 out of 30 (standard deviation 1.84). Thirty-four students also mastered the 3,000 word level (average score: 26.05; standard deviation 2.69). A pretest had indicated that these participants were unfamiliar with the target collocations. Since two students did not complete all the exercises, their data were not analysed, bringing the total number of participants to 41 (21 male, 20 female).

2 Target collocations

On the basis of a pretest, 18 target collocations that were unknown to all participants were selected (see Table 1). It was verified that each target collocation was an acceptable collocation by checking it against the Oxford Business English dictionary (Parkinson, 2005), the Oxford collocations dictionary for students of English (Lea, Crowther, & Dignen, 2002) and the BBI (Benson, Benson & Ilson, 1997). Only collocations that were mentioned in at least one of these dictionaries were retained. These three dictionaries define collocations in the phraseological sense, i.e. collocations are lexical combinations characterized by restricted co-occurrence of items. Collocations should be distinguished from free combinations (e.g. buy a book) and idioms (e.g. buy a pig in a poke). Moreover, all 18 collocations were listed in the BYU-BNC (Davies, 2004–). Except for issue shares (187 occurrences), all frequencies ranged from 4 (run up a deficit) to 64 occurrences.

Table 1.

Target collocations divided by congruency and collocate–node relationship.

	Congruent collocations	Incongruent collocations
Adjective–noun collocations	perishable goods	a square deal
	fierce competition	flat sales
	steady work	intangible assets
Verb–noun collocations	exceed the budget	corner the market
	exhaust supply	issue shares
	settle an invoice	slash costs
Phrasal-verb–noun collocations	run up a deficit	tie up capital
	take on staff	take out a mortgage
	write off debts	step up production

The 18 target collocations were divided into nine congruent and nine incongruent collocations (see below for how congruency was determined) and into six adjective–noun collocations, six verb–noun collocations, and six phrasal-verb–noun collocations (see Table 1). The node word in each collocation was a noun. Except for invoice, the nouns in the collocations were all listed in the two highest frequency bands of the Collins Cobuild English Dictionary for Advanced Learners (Sinclair, 2001). On the basis of the VLT results, participants were expected to be familiar with the nouns of the collocations. The adjectives and verbs on the other hand ranged in frequency of occurrence. Collocation length varied from 9 to 17 letters. However, since word length might influence the learning difficulty of an item, it was taken into account as a covariate in the statistical analyses (see also Section IV.6).

It was further verified that all adjective–noun collocations corresponded to Dutch adjective–noun collocations because they can also be compounds in Dutch, e.g. parental leave = ouderschapsverlof (‘parenthood leave’). In addition, all phrasal verbs were separable verbs in Dutch (= verb + separable particle), e.g. write off (‘he writes off’) corresponds to Dutch afschrijven (hij schrijft af). English phrasal verbs can also correspond to Dutch inseparable verbs, e.g. lay off = ontslaan. The phrasal verbs in the phrasal-verb–noun collocations were further checked against the Oxford Dictionary of Phrasal Verbs (Cowie & Mackin, 1993).

Nesselhauf’s (2003) definition of congruency was adopted to operationalize congruency. A congruent collocation is a collocation that can be translated literally into the L2. L2 words are literal or word-for-word translations if the word in question would – out of context – be translated into the target word in question (Nesselhauf, 2005, p. 221). A non-congruent collocation, on the other hand, cannot be given by a word-for-word translation (Nesselhauf, 2003). Two raters, who were unaware of the aim of the study, were asked to translate the individual constituents making up the collocations out of context to determine the congruency of the collocations. For instance, steady work was considered a congruent collocation as it could be rendered by a word-for-word translation (vast was translated into steady and werk into work). However, to corner the market was considered a non-congruent collocation because neither rater translated veroveren into to corner but into to conquer, to take over. Only items on which the two raters agreed, i.e. both had given or neither had given the English word in question, were retained.

The initial plan was to select more than 18 collocations, but this appeared to be impossible because the pretest had indicated that the participants were familiar with more congruent collocations, especially congruent phrasal-verb–noun collocations, than expected.

3 Treatment

All participants were provided with a word list and four online exercises that constituted the learning materials (see also Figure 1). The word list contained the 18 target collocations, each accompanied by its L1 translation and an English sample sentence, in alphabetical order. Sentences were taken from either a British newspaper or a learner dictionary. The list remained available when students were doing the exercises. The provision of such a word list guaranteed that the participants were first presented with the intact collocations. Here is an example of an item in the word list:

To tie up capital: kapitaal vastzetten

Our capital is all tied up in property.

Figure 1.

Example of two exercise types.

The exercises were designed according to the principle of form-focused (Laufer, 2005) or explicit vocabulary instruction (Schmitt, 2008) since it has been argued that explicit instruction is an optimal approach for initial form–meaning mapping in classroom-based courses (Hulstijn, 2001; Schmitt, 2008), being the focus of this study. In addition, several studies have shown that explicit instruction is beneficial for learning collocations (Laufer & Girsai, 2008; Peters, 2014; Webb & Kagimoto, 2011). Because one encounter does not suffice to create a strong form–meaning link in the mental lexicon (Hulstijn, 1992) and repetition in vocabulary-focused activities is beneficial for word learning (Laufer & Rozovski-Roitblat, 2011; Peters, 2014), the target collocations were offered once in the word list and twice in the exercises.

Specifically, four form-focused vocabulary exercise were designed: two, in which participants had to supply a synonym, an antonym, a superordinate or a hyponym for the collocations (= SASH exercises), and two fill in the gap exercises (= FIG exercises) (see also Figure 1). These exercise types were chosen for reasons of ecological validity as they were in line with students’ course materials. Each exercise consisted of nine items. Participants were always required to provide the collocation as a whole to prevent them from making erroneous combinations (Boers et al., 2014a). Each target collocation occurred twice in the exercises, once in one of the two SASH exercises and once in a fill in the gap exercise.

Online exercises were preferred to paper-and-pencil exercises because the former allow participants to receive immediate and individualized feedback on their responses after completion of each exercise.² The feedback consisted of the correct collocation in case the students provided an incorrect collocation. Each group of students received the four exercises in a different order to control for order effects.

4 Tests

Lexical pre-knowledge was controlled for by means of a pretest. The collocations in the pretest were chosen from students’ Business English course materials for reasons of ecological validity. The pretest, which contained 88 items, also included items that would not be in the treatment in order not to alert participants to possible target items. In addition, some of these distractors were fairly easy to ensure that students would not give up supplying answers. Participants were required to translate the test items into English. The first letter of each word was given to ensure that participants would not supply a correct but alternative answer. In addition, when students had to supply a collocation containing a phrasal verb, it was indicated that the verb consisted of two parts. The following examples illustrate the test format:

een tekort oplopen: to r……………………… (2 parts) a d…………………… (to run up a deficit)

een trouwe klant: a l…………………………c…………………………. (a loyal customer)

Learning of collocations was measured in three immediate written posttests, two form recall tests and one form recognition test, to get a more accurate picture of participants’ learning gains and to measure learning of collocations at different levels of sensitivity (Nation & Webb, 2011, p. 284). In each posttest, participants were explicitly asked to supply a collocation from the word list and the exercises. In the first form recall test, participants were provided with the Dutch translation of the collocations and asked to supply the English collocation. No clue, such as the first letter, was given (see example below).

Form recall test 1

vast werk: …………………….

schulden afschrijven: …………………………….

Unlike the first test, the second form recall test offered a clue, viz. the noun in the collocation, which might trigger learners’ recollection of the complete collocation.

Form recall test 2

vast werk: ………………………. work

schulden afschrijven: ………………………………. debts

Finally, the form recognition test (see Figure 2) was a matching test, in which the participants had to match the individual constituents of the collocations. The 18 collocations were divided into three blocks, each consisting of six collocates (verb, phrasal verb, or adjective) that had to be combined with one of nine nouns. More options were provided to minimize the chances of successful guessing.

Figure 2.

Form recognition test.

Delayed posttests were not administered for two reasons. First, it was not possible to control for the time period between the immediate and delayed posttests because authentic collocations were involved and not collocations made up of pseudowords. So students might look up collocations at home. Second, unlike the pretest, the posttests contained only 18 items and did not contain any distractors, which would have made it considerably easier to remember the target items compared to the many pretest items for instance. As a result, it would not have been possible to determine whether the learning gains measured in the delayed posttests were an effect of the type of collocation or an effect of deliberate, intentional learning at home (see also Hulstijn, 2003; Nation & Webb, 2011).

5 Procedure

The experiment consisted of two sessions. Each session was organized in six parallel groups during students’ regular English classes within a one-week period.

Session 1. Two weeks prior to the treatment, the pretest was administered. At that stage, participants were not informed that they were participating in an experiment. The pretest was presented as an intake test at the beginning of the year.

Session 2. The second session was organized during one class period (2 hours) and consisted of three consecutive stages: (1) learning session, (2) Vocabulary Levels Test, and (3) posttests.

(1) At the start of this session, participants were told that they were to take part in a study on EFL learners’ perception of online exercises. They were not informed that the aim of the experiment was vocabulary learning. Next, all participants received the word list containing all the target collocations, their translation and a sample sentence for each collocation (see also Section IV.3). Having read this word list, participants were told that they had to carry out four online exercises on these 18 collocations. It was shown how the online exercises and the feedback worked. The order of the four exercises was different for each of the six parallel groups of students to control for any order effects. Thus, some students started with a fill-in-the-gap exercise, whereas other students began with a SASH exercise (for a description of the exercise types, see Section IV.3). Participants could do the exercises at their own pace and ask questions in case anything was unclear. They received feedback on incorrect answers (= provision of the correct response) immediately after completion of each of the four exercises. (2) Having finished the four exercises, participants handed in the word list and started taking the Vocabulary Levels Test, which allowed for a short time gap between the treatment and the posttests (= stage 2). (3) Finally, all participants were unexpectedly tested on their knowledge of the target collocations (= stage 3). They first took the form recall test without clues, then the second form recall test (with clues), and third the form recognition test. The posttests were paper-and-pencil tests. Students needed approximately 90 to 120 minutes to complete all tasks and tests. At the end, they were debriefed about the aim of the study.

6 Scoring and analyses

All posttests were scored dichotomously. A correct answer received one point, an incorrect answer zero points. Answers in the two form recall posttests that contained minor spelling mistakes but that did not (or hardly) affect the pronounceability, such as a squere deal (instead of square deal), were scored as correct; instangible assets or tangible assets (instead of intangible assets) were not. The two form recall tests were scored by two raters. The interrater reliability was very high (r = .999; p < .0001 for the first form recall test; r =.999; p < .0001 for the second form recall test). The items on which the two raters did not agree were discussed and a consensus was reached.

To answer the research questions, a repeated measures logistic regression (= a Generalized Estimating Equation or GEE in SPSS) was deemed appropriate (Agresti, 2002; Peng & So, 2002) because it is suited for the analysis of dichotomous data (e.g. correct or incorrect response in a posttest). By analysing the relationship between the independent variables, which can be categorical (= factors) or continuous variables (= covariates), and the dichotomous dependent variable, the odds in favor of a correct answer in one of the three posttests can be predicted. A GEE has the advantage over an AN(C)OVA that it allows for the inclusion of main independent variables (congruency and collocate–node relationship) as well as item-related variables, such as word length, and learner-related variables, such as vocabulary size in one model. Unlike linear regression analyses or AN(C)OVAs, a repeated measures logistic regression does not require the assumptions of normality, homogeneity of variance or linearity to be met. However, the case–parameter ratio should be > 10 to 1 and there should not be multicollinearity. A GEE is based on the number of observations (number of participants multiplied by the number of items) and not on test scores as in an AN(C)OVA. This means that the combination ‘Participant, Item and Score’ defines, for each observation or case, a particular score (correct or incorrect) on a particular item for a particular participant. The odds ratio (= exp^β or exponential parameter estimate) can be considered a measure of effect size and predicts the odds in favor of a correct response with a specific parameter.³

First, the assumptions were tested. The dependent variable was a dichotomous variable (correct or incorrect response in the posttests), the case–parameter ratio was acceptable ( > 10 to 1) and no multicollinearity was detected. Next, one analysis for each posttest was computed in SPSS using Generalized Estimating Equations (GEE). To predict the odds of a correct response in one of the three posttests, the following parameters were entered into the model as factors: congruency, collocate–node relationship, and finally the interaction between congruency and collocate–node relationship. The following parameters were entered as covariates into the model: participants’ vocabulary size, word length of the first word (= number of syllables) (= word length 1) and word length of the second word (= number of syllables) (= word length 2). After fitting this first model, parameters that did not contribute significantly to the regression model were removed from the model (backward elimination) and the model was refit again until it included only significant main effects and interactions. Pairwise comparisons were computed for each test to determine the interaction contrasts (congruency*collocate–node relationship) and the contrasts for the main factors (congruency and collocate–node relationship). Main factors will only be interpreted if there is no significant interaction between congruency and collocate–node relationship.

V Results

1 Online exercises

Students completed four online exercises, in which the target collocations had to be supplied twice. The descriptive statistics show that participants performed better in the SASH exercises than in the fill-in-the-gap exercises (7.54 and 7.66 out of 9 vs. 6.20 and 5.10 out of 9). In the former exercises, the scores ranged from 5 to 9, in the latter from 2 to 9.⁴

2 Posttests

In this study, the descriptive statistics provide information on the number or percentage of correct/incorrect responses per collocation type (= number of observations). They are presented in Tables 3 and 9; the inferential statistics are provided in Tables 4 –8. For each posttest, a GEE analysis was computed with the 18 target collocations as repeated measures within participants (n = 41) and the posttest score as dependent variable. The GEE analysis was based on 738 observations or cases (41 participants multiplied by 18 target items). The main factors were coded as follows: congruency = 0 (= incongruent) or 1 (= congruent); collocate–node relationship = AN (= adjective–noun collocation), VN (= verb–noun collocation) or PVN (= phrasal-verb–noun collocation). Table 2 lists the parameters which contributed significantly to the model and were retained in each GEE analysis.

Table 2.

List of parameters in each Generalized Estimating Equations analysis.

Test	Parameters
Form recall test 1	congruency, collocate–node relationship, congruency*collocate–node relationship, word length 1, word length 2, vocabulary size
Form recall test 2	congruency, collocate–node relationship, word length 1, word length 2
Form recognition test	collocate–node relationship, word length 1

Note. * stands for “statistically significant”, but I have removed the * because it is clear from the table.

a Congruency

The first research question focuses on whether congruency affects the odds that a learner will recall or recognize the form of a new collocation. The analysis of congruency was based on 369 observations for the congruent and 369 observations for the incongruent items (41 participants multiplied by 9 items).

In the first form recall test, there were more correct responses for the congruent (167 correct responses or 45.3%) than for the incongruent collocations (130 correct responses or 35.2%) (see Tables 4 and 5). Although the test of model effects in the GEE revealed that congruency predicted the odds in favor of a correct response, its effect was conditioned by the collocate–node relationship given the significant interaction between the two parameters (p < .0001) (see Tables 4 and 5).⁵

A similar pattern was revealed in the second form recall test: more correct responses for the congruent (186 correct responses or 50.4%) than for the incongruent items (131 correct responses or 35.5%). The GEE revealed that congruency was a significant predictor in the second form recall test. The odds in favor of a correct response were only 45% (exp^β = exp^−.80 = .45) with an incongruent item compared to a congruent collocation (= reference value) or, to put it differently, they were two times higher with a congruent than with an incongruent collocation (1/exp^β or 1 / exp^.45 = 2.23) (see Tables 6 and 7).

Given the small difference in the number of correct responses (292 for the congruent and 266 for the incongruent items) in the form recognition test, it is hardly a surprise that congruency did not contribute significantly to the regression model, neither as a main factor nor in interaction with the variable collocate–node relationship.

b Collocate–node relationship

The second research question concerns the variable collocate–node relationship, for which 246 observations per collocation subtype (41 participants multiplied by 6 items) were collected. In the first form recall test, there were more correct responses for the adjective–noun collocations (115 correct responses or 46.7%) than for the verb–noun (93 correct responses or 37.8%) and phrasal-verb–noun (89 correct responses or 36.2%) collocations, as can be gleaned from Table 3. The Test of Model effects in the GEE analysis indicated that the collocate–node relationship was indeed a significant predictor of a correct response (see Tables 4 and 5). However, its effect was conditioned by congruency given the significant interaction with congruency (p < .0001) (see below).

Table 3.

Number of correct responses (percentages in brackets) per collocation type and per posttest.

	Congruent(Max = 369)*		Incongruent(max = 369)		AN (max = 246)		VN(max = 246)		PVN(max = 246)
	Correct	Incorrect	Correct	Incorrect	Correct	Incorrect	Correct	Incorrect	Correct	Incorrect
Test 1	167	202	130	239	115	131	93	153	89	157
%type	(45.3%)	(54.7%)	(35.2%)	(64.8%)	(46.7%)	(53.3%)	(37.8%)	(62.2%)	(36.2%)	(63.8%)
%tot	(22.6%)	(27.4%)	(17.6%)	(32.4%)	(15.6%)	(17.8%)	(12.6%)	(20.7%)	(12.1%)	(21.3%)
Test 2	186	183	131	238	128	118	95	151	94	152
%type	(50.4%)	(49.6%)	(35.5%)	(64.5%)	(52%)	(48%)	(38.6%)	(61.4%)	(38.2%)	(61.8%)
%tot	(25.2%)	(24.8%)	(17.8%)	(32.2%)	(17.3%)	(16%)	(12.9%)	(20.5%)	(12.7%)	(20.6%)
Test 3	292	77	266	103	223	23	170	76	165	81
%type	(79.1%)	(20.9%)	(72.1%)	(27.9%)	(90.7%)	(9.3%)	(69.1%)	(30.9%)	(67.1%)	(32.9%)
%tot	(39.6%)	(10.4%)	(36%)	(14%)	(30.2%)	(3.1%)	(23%)	(10.3%)	(22.4%)	(11%)

Note. * Max = 41 participants multiplied by 9 target items = 369; ** Max = 41 participants multiplied by 6 items = 246; AN = adjective–noun collocation; VN = verb–noun collocation; PVN = phrasal-verb–noun collocation; %type = % within type of collocation; %tot = % within Total; Test 1 = Form recall test 1; Test 2 = Form recall test 2; Test 3 = Form recognition test.

Table 4.

Generalized Estimating Equations analysis of Form recall test 1: Test of model effects: Type III.

Source	Wald Chi-square	df	Sig
(Intercept)	.98	1	.321
Congruency	13.31	1	.0001
collocate–node relationship	11.11	2	.004
collocate–node relationship *Congruency	17.73	2	.0001
Word length 1	34.49	1	.0001
Word length 2	18.38	1	.0001
Vocabulary size	4.61	1	.03

Note. * stands for “statistically significant”, but I have removed the * because it is clear from the table.

Table 5.

Generalized Estimating Equations analysis of Form recall test 1: Parameter estimates.

Parameter	β	SE	95% Wald confidence interval		Wald Chi-square	p	exp^β
			Lower	Upper
(Intercept)	−1.61	1.48	−4.51	1.30	1.18	.28	.20
Congruency = 0	−.12	.32	−.75	.51	.14	.71	.89
Congruency = 1	0^a	.	.	.	.	.
Collocation type = AN	1.21	.26	.70	1.71	22.12	.000	3.34
Collocation type = PVN	.36	.24	−.12	.84	2,20	.14	1.44
Collocation type = VN]	0^a	.	.	.	.	.
Congruency = 0* Collocation type = AN	−1.42	.35	−2.10	−.75	16.98	.000	.24
Congruency = 0* Collocation type = PVN	−.27	.38	−1.01	.48	.49	.48	.77
Congruency = 0* Collocation type = VN	0^a
Word length 1	−.59	.10	−.78	−.39	34.49	.000	.56
Word length 2	−.45	.11	−.66	−.25	18.38	.000	.64
Vocabulary size	.03	.01	.00	.05	4.61	.032	1.03

Note. ^a Set to zero because this parameter is redundant. * statistically significant, β: regression coefficient.

Table 3 shows that participants’ scores on the second form recall test were in line with the first one. The test of model effects of the GEE analysis (see Tables 6 and 7) indicated that the collocate–node relationship significantly affected the odds in favor of a correct answer, whereby it was two times more likely (exp^.72 = 2.04) to have a correct response with an adjective–noun collocation than with a verb–noun collocation. Pairwise comparisons revealed that the number of correct responses was significantly higher for the adjective–noun collocations compared to the verb–noun (p < .0001) and phrasal-verb–noun (p = .002) collocations. No difference was found between the verb–noun and phrasal-verb–noun collocations (p = .41).

Table 6.

Generalized Estimating Equations analysis of form recall test 2: Test of model effects: Type III.

Source	Wald Chi-square	df	Sig
(Intercept)	18.69	1	.0001
Congruency	24.26	1	.0001
Collocate–node relationship	23.64	2	.0001
Word length 1	40.60	1	.0001
Word length 2	6.57	1	.01

Note. * statistically significant.

Table 7.

Generalized Estimating Equations analysis of form recall test 2: Parameter estimates for form recall test 2.

Parameter	β	SE	95% Wald confidence interval		Wald Chi-square	p	exp^β
			Lower	Upper
(Intercept)	1.09	.28	.53	1.65	14.67	.0001	2.98
Congruency = 0	−.80	.16	−1.12	.48	24.26	.0001	.45
[Congruency = 1]	0^a	.	.	.	.	.
Collocation type = AN	.72	.16	.39	1.04	18.52	.0001	2.04
Collocation type = PVN	.16	.20	−.23	.55	.63	.43	1.17
Collocation type = VN	0^a	.	.	.	.	.
Word length 1	−.54	.08	−.70	−.37	40.60	.0001	.59
Word length 2	−.21	.08	−.37	−.05	6.57	.01	.81

Note. ^aSet to zero because this parameter is redundant. * statistically significant, β: regression coefficient.

Finally, the difference in the number of correct responses between the adjective–noun and (phrasal) verb–noun collocations was larger in the form recognition test, as can be seen in Table 3. The odds in favor of a correct response (see Tables 8 and 9) were four times (exp^1.45 = 4.25) higher with an adjective–noun collocation than with a verb–noun collocation. The pairwise comparisons revealed that the difference between the adjective–noun collocations on the one hand and the verb–noun and phrasal-verb–noun collocations on the other was significant (p < .0001). There was no difference between the verb–noun and phrasal-verb–noun collocations (p = .32).

Table 8.

Generalized Estimating Equations analysis of form recognition test: Tests of model effects: Type III.

Source	Wald Chi-square	df	Sig
(Intercept)	1.90	1	.169
Collocate–node relationship	37.65	2	.0001
Word length 1	9.22	1	.002

Note. * statistically significant.

Table 9.

Generalized Estimating Equations analysis of form recognition test: Parameter estimates for recognition test.

Parameter	β	SE	95% Wald confidence interval		Wald Chi-square	p	exp^β
			Lower	Upper
(Intercept)	−.01	.33	−.66	.65	.00	.98	.99
Collocation type = AN	1.45	.25	.96	1.93	34.27	.000	4.25
Collocation type = PVN	−.17	.18	−.52	.17	.96	.33	.84
Collocation type = VN	0^a	.	.	.	.	.	.
Word length 1	.45	.15	.16	.73	9.22	.002	1.56

Note. ^a Set to zero because this parameter is redundant. * statistically significant, β: regression coefficient.

c Interaction between congruency and collocate–node relationship

Research question 3 is related to the interaction between congruency and the collocate–node relationship. The descriptive statistics are provided in Table 10. The GEE analysis was based on 123 observations per collocation type (41 participants multiplied by 3 items), e.g. 123 observations for the congruent adjective–noun collocations.

Table 10.

Number of correct/incorrect responses per collocation type (percentages in brackets).

	CongAN		IncongAN		CongVN		IncongVN		CongPVN		IncongPVN
	(Max = 123)		(Max = 123)		(Max = 123)		(Max = 123)		(Max = 123)		(Max = 123)
	Correct	Incorrect	Correct	Incorrect	Correct	Incorrect	Correct	Incorrect	Correct	Incorrect	Correct	Incorrect
Test 1	70 (9.49%)	53 (7.18%)	45 (6.1%)	78 (10.57%)	41 (5.56%)	82 (11.11%)	52 (7.05%)	71 (9.62%)	56 (7.59%)	67 (9.08%)	33 (4.47%)	90 (12.20%)
Test 2	75 (10.16%)	48 (6.50%)	53 (7.18%)	70 (9.49%)	49 (6.64%)	74 (10.03%)	46 (6.23%)	77 (10.43%)	62 (8.40%)	61 (8.27%)	32 (4.34%)	91 (12.33%)
Test 3	114 (15.45%)	9 (1.21%)	109 (14.77%)	14 (1.90%)	87 (11.79%)	36 (4.88%)	83 (11.25%)	40 (5.42%)	91 (12.33%)	32 (4.34%)	74 (10.03%)	49 (6.64%)

Notes. Max = 123 = 41 participants multiplied by 3 items; Cong = congruent; Incong = incongruent; AN = adjective–noun collocation; VN = verb–noun collocation; PVN = phrasal-verb–noun collocation; Test 1 = Form recall test 1; Test 2 = Form recall test 2; Test 3 = Form recognition test.

As can be seen, the highest number of correct responses in the first form recall test was found for the congruent adjective–noun collocations, whereas the lowest number for the incongruent phrasal-verb–noun collocations. An unexpected result was the low number of correct responses for the congruent verb–noun collocations (41), which can largely be attributed to the low recall score of the item settle an invoice (only 7 correct responses). The GEE analysis revealed a significant interaction between congruency and collocate–node relationship. The odds in favor of a correct response were four times higher (1/e^−1.42 = 4.17) with a congruent adjective–noun collocation than with a congruent verb–noun collocation (see Tables 4 and 5). The pairwise comparisons revealed that the congruent adjective–noun collocations differed significantly from the congruent verb–noun collocations (p = .001) and the incongruent phrasal-verb–noun collocations (p = .002). In the second form recall test as well as in the form recognition test, the interaction between congruency and collocate–node relationship did not contribute significantly to the model.

d Word length

The GEE analyses, computed for 738 observations (41 participants multiplied by 18 items), revealed that word length of both the first and the second word of the collocation was negatively correlated with form recall. This means that the longer the word the lower the odds of a correct response. Specifically, for each syllable more in the word, the odds in favor of a correct response would be only 56% in the first and 59% in the second posttest (exp^−.59 = .56; exp^−.54 = .59); the odds in favor of a correct response with a noun with one additional syllable (= word length 2) would be only 64% (exp^−.45 = .64) in the first and 81% (exp^−.21 = .82) in the second posttest. However, a different picture emerged in the form recognition test. The GEE analysis showed that only the first word’s length affected the odds in favor of a correct response. Unlike the correlation in the form recall tests, the relationship between word length and learning gains was now positive (see Tables 8 and 9). Specifically, for each syllable more in the adjective/verb, the odds in favor of a correct response would be 1.5 times higher (exp^.45 = 1.56). A possible explanation for this mixed finding will be provided in Section VI.

e Vocabulary size

The GEE analysis indicated that vocabulary size only contributed significantly to the model of the first form recall posttest (see Tables 4 and 5). Unsurprisingly, the relationship was positive (b = .03). Thus, the larger a learner’s vocabulary size the higher the odds that she will recall a new collocation. For every 100 words more in a learner’s vocabulary size, the odds that she would recall a new collocation in the first form recall test were 13 times higher (exp^100*.03 = 13.46).

VI Discussion

This study looked at the initial form–meaning mapping of new collocations by EFL learners in a short vocabulary-focused learning treatment. In particular, it aimed to investigate whether congruency, collocate–node relationship, word length, and finally learners’ vocabulary size predict the odds that EFL learners correctly recall or recognize specific types of collocations (congruent vs. incongruent collocations, adjective–noun vs. (phrasal) verb–noun collocations).

1 Congruency

The findings suggest that congruency only played a role in the two form recall tests but not in the form recognition test, whereby incongruent collocations seem to be more difficult to learn than congruent collocations. In addition, congruency was conditioned by the collocate–node relationship in the first form recall test.

Before turning to the discussion of the difference between the congruent and incongruent collocations, as revealed in the second form recall test, the unexpected finding regarding the verb–noun collocations in the first form recall test should be addressed. An explanation might be found in the low number of different items per subgroup and in the low score of the congruent verb–noun collocation to settle an invoice in particular. Although the two raters translated vereffenen into settle and a bilingual dictionary mentions it as its first translation, participants may have been more or only familiar with the meaning to start living somewhere, which might have confused them. In spite of a clear definition of congruency, viz. presence or absence of a literal L1 translation equivalent, the example of settle an invoice suggests that congruency might be less easy to operationalize than previously hypothesized because of polysemy and prototypicality of meaning. Swan (1997) already pointed to the role of prototypicality in translation equivalence. ‘Languages may have exact translation equivalents when used in their central sense but not when they are used in more marginal or metaphorical ways’ (Swan, 1997, p. 158). One could thus tentatively argue that the degree to which a collocation is perceived as congruent might differ from learner to learner and this will undoubtedly be affected by learners’ proficiency and vocabulary knowledge.

Except for the verb–noun collocations in the first form recall test, the findings tentatively suggest that congruent collocations were more likely to be recalled than incongruent ones. A first reason could be learners’ focus on single words, which was manifested in some of the errors they produced in the form recall tests. Erroneous or new combinations (tie up an invoice), L1 induced single word for single word translations (to bring out shares, to conquer the market) in the recall tests and instances of incorrect matching in the recognition test (flat costs, flat shares, perishable profit) were abundant. Thus, in spite of a correct intact model in the word list and the exercises, the collocations were not necessarily learned and stored as chunks. Second, when FL learners tend to focus on word size units (Wray, 2002), learners’ L1 may have an inhibitive effect on learning incongruent collocations and a facilitative effect on the learning process of some congruent combinations, as claimed by Wolter and Gyllstad (2011). They argue that in the case of congruent collocations, FL learners can easily swap out the L1 words with corresponding L2 words (p. 446). However, because no literal equivalent is available for the incongruent collocations, the learning process is more complex and more demanding so FL learners cannot rely on the L1 patterns of the single word in question. Instead they need to make new connections between single words and deal with ‘the new complexities of the L2’ (p. 446). Learners need to know that verbs and adjectives can have a different meaning when they are combined with a particular noun, and that nouns can only be combined with a restricted number of verbs and adjectives. Participants might have been familiar with the meaning of take out in take somebody out to a restaurant, but this knowledge is of little help when asked to supply the correct combination to take out a mortgage. Instead, some participants drew on their L1 and produced the infelicitous collocation to shut down a mortgage, falling into the trap of ‘deceptive compatibility’ (Laufer, 2011), which illustrates that FL learners do rely on their L1 when asked to recall new L2 collocations. The fact that congruency only played a role in the two form recall tests and not in the form recognition test suggests that the main difficulty of incongruent collocations lies in their production and not in their recognition.

It might be interesting to point out that a number of the aforementioned L1-induced errors in the posttests already occurred in the pretest. For instance, seven participants supplied conquer the market (or one of its deviant spelling variants: concur(e), concer, concuer the market) for the collocation corner the market and adhered to this combination in the posttests in spite of the provision of a correct model in the word list and the online exercises. It is not unlikely that these erroneous guesses in the pretest had already resulted in incorrect form–meaning links in the mental lexicon.

From these findings, it is clear that more research is needed before any firm conclusions on the effect of congruency on collocational learning can be drawn. Moreover, a stricter and more precise definition of congruency, taking into account prototypicality and polysemy, might be necessary. The findings presented here should thus be interpreted with great care.

2 Intralexical factors and learners’ vocabulary size

Research question 2 can be answered affirmatively. The findings suggest that the collocate–node relationship is a significant predictor in collocational learning. In each posttest, the highest scores were found for the adjective–noun collocations in spite of the negative role of (some of) the adjectives’ word length (intangible, perishable). The odds in favor of a correct response with an adjective–noun collocation were the greatest in the form recognition test. However, the collocate–node relationship was conditioned by congruency in the first form recall test. Three possible interpretations will be put forward: morphological variation in the collocation’s form, modification of the noun’s meaning and the specific difficulty of the phrasal-verb–noun collocations.

Because adjectives show less morphological variation than verbs, the presentation of the adjective–noun collocations’ form was always the same, both in the sample sentences in the word list and in the fill-in-the-gap exercises, whereas the collocations containing a (phrasal) verb sometimes had a different form. In the sample sentence he will write off household debts, the collocation appeared with an additional noun household. Other examples are participles (by slashing costs), past tense (the estimated costs … exceeded the budget) or passive voice (our capital is all tied up in property). Laufer (2011) also pointed out that one of the difficulties FL learners face when consulting learner dictionaries is that the verb–noun collocation may appear in a different form in the example given (for an example, see Section II). Thus, the unaltered form of the adjective–noun collocations may have left a stronger imprint on learners’ mental lexicon compared to the (phrasal) verb–noun collocations. This may also explain why the effect of collocate–node relationship was larger in the recognition test compared to the recall tests.

Second, the better retention scores of the adjective–noun collocations can also be explained by the fact that the meaning of the adjective–noun collocations may have been perceived as a meaningful whole compared to the (phrasal) verb–noun collocations. This interpretation is in line with Wolter’s (2006) claim that the new combination of words can lead to conceptual modification and to a single unified concept. Although he does not explicitly distinguish between adjective–noun and verb–noun collocations, the idea of conceptual modification might be more relevant to adjective–noun than to verb–noun collocations, since adjectives are traditionally defined as modifiers of nouns. For instance, in flat sales, flat modifies the meaning of sales by describing what the sales are like. It is not unlikely that learners perceived the meaning of this combination as one single entity. This modification of meaning is even more prominent when the adjective in the collocation is a classifying adjective. Perishable in perishable goods classifies the noun goods as a type of goods. Given this new single unified concept and the unaltered form, the adjective–noun collocations may have been more likely to be learned as chunks compared to (phrasal) verb–noun collocations; an interpretation which is further illustrated in the higher number of incorrect or new (phrasal) verb–noun combinations, e.g. settle the market or slash an invoice, compared to the adjective–noun collocations. Some of these errors could indeed indicate that the words making up the (phrasal) verb–noun collocations were stored separately and then (incorrectly) combined in the posttests. Even if participants may have tried to learn (some of) the collocations as a whole, ‘holistically learned strings are (…) subject to loss of detail’ (Wray, 2002, p. 200).

The difficulty of the phrasal-verb–noun collocations was amongst others manifested in the fact that participants sometimes resorted to one-word synonyms for the phrasal verbs in the two form recall tests, e.g. hire staff for take on staff or raise/increase production for step up production. Although they were instructed to use a collocation from the exercises, some students fell back on familiar single-verb synonyms and adopted a play-it-safe strategy, which is consistent with Hulstijn and Marchena’s (1989) findings. The translation format of the two form recall posttests might also have induced the avoidance of phrasal verbs to some extent, as was demonstrated by Liao and Fukuya (2004). Second, in spite of the formal similarity in the particles between the congruent phrasal-verb–noun collocations and their L1 equivalent (up – op (run up a deficit) and on – aan (take on staff)), a considerable number of errors involved the wrong choice of particles, for instance run, run out, run on, run off a deficit instead of run up a deficit or take, take off, take in, take up staff instead of take on staff. It is not immediately clear why this occurred. One explanation might be found in Hulstijn and Marchena’s (1989) study, which showed that Dutch-speaking learners avoided phrasal verbs that had Dutch equivalents with the same meaning. It might be possible that the congruent phrasal-verb–noun collocations were perceived too ‘Dutch-like’ and too similar to the learners’ L1, as a consequence of which participants may have supplied no or incorrect particles.

This study also explored the influence of word length. The research findings revealed that there was a negative correlation between word length and form recall and a positive one with form recognition. An analysis of participants’ errors might explain this mixed finding because it showed that some incorrect collocations in the form recall tests showed formal similarity to the target items. For instance, intangible assets had one of the lowest scores in the two recall tests, but one of the highest scores in the form recognition test. The form was obviously difficult for many of the participants, as was shown in their answers (astingable, instangible, instagneble, tangible assets). However, in spite of being incorrect, these answers show that participants had some partial knowledge of the form intangible assets, being linked to the correct concept. Although it may still have been too difficult for the participants to provide the adjective’s correct form in the recall tests, it was apparently far easier to correctly recognize intangible assets in the form recognition test. Similarly, a number of participants wrote persiable, perniable, persuable goods instead of perishable goods in the form recall tests but they experienced far fewer difficulties when asked to recognize this collocation in the form recognition test. The aforementioned errors in the form recall tests might to some extent be attributed to the word length of these two adjectives. In addition, these two adjectives are clearly more distinctive in form than high-frequency words, such as run or take. As a result, the collocations perishable goods and intangible assets may have attracted learners’ attention more compared to the items run up a deficit and write off debts. The low scores of the former two items in the recall tests and their high scores in the recognition test also explain why the relationship between word length 1 and learners’ responses was negative in the two recall tests and positive in the recognition test.

Unsurprisingly, vocabulary size was positively correlated with participants’ scores but only in the first form recall test. This finding is consistent with other research that found that learners with a larger vocabulary tend to pick up more new words (Zahar, Cobb & Spada, 2001; Montero Perez, Peters, Clarebout, & Desmet, 2014).

3 Pedagogic implications

Teachers wishing to spend classroom time on collocations might take into account that it is more than one factor which is likely to influence the learning burden of collocations:

incongruency;

polysemy of adjectives/verbs in the collocations;

students’ tendency to focus on single words;

the type of collocate–node combination (verb–noun or adjective noun);

the items’ length;

type of knowledge: production vs. recognition.

First, learners should be taught that they cannot always draw on L1 patterns when combining single words. Given the pervasive influence of learners’ L1 in participants’ errors, it might be useful to make learners aware of interlingual differences, for instance via contrastive translation activities (Laufer & Girsai, 2008). Second, learners’ familiarity with or knowledge of the single items’ prototypical meaning (or the collocate’s prototypical meaning) might prevent them from using that item in the collocation. Teachers should thus draw learners’ attention to a word’s polysemy and show that adjectives and verbs in combination with nouns might have a different meaning. As FL learners sometimes tend to resort to single word learning, teachers and material writers should attempt to present new collocations and verb–noun collocations in particular intact (Boers et al., 2014a), even though it is no guarantee of learners’ processing and storing them holistically. In addition, learners’ attention could also be drawn to the collocation as a whole by making the target collocation visually salient (bold, underlined, italics) (Laufer, 2011; Peters, 2012; Sonbul & Schmitt, 2013). This is particularly relevant to verb–noun collocations because they can appear in different forms and show more morphological variation than adjective–noun combinations. Regarding the items’ length, it seems worthwhile to also draw learners’ attention to short or shorter words (e.g. run, take) in the input or in recognition activities, as they are less likely to be noticed than longer, more distinctive words, such as perishable or intangible.

Finally, the findings show that it is productive knowledge where learners’ difficulties with collocations lie. From corpus studies we already know that foreign language learners experience difficulties when using collocations in free production. The present study shows that even controlled productive use presents a serious challenge to learners. Although the present findings look more promising than the findings in studies focusing on incidental learning of collocations (see, for instance, Webb et al., 2013), they also show that form recall of newly learned target collocations remains difficult even after an intensive, explicit learning treatment. In addition, it appeared to be difficult to push participants to use the target collocations instead of familiar, synonymous items. Given that incongruent collocations, (phrasal) verb–noun collocations and collocations consisting of long individual words might need more output practice than others in order to be well entrenched in the mental lexicon, learners should be provided with repeated opportunities to encounter and practice target collocations (Peters, 2014; Webb et al., 2013). Although this study did not examine the type of instruction and form recall remained a challenge for many participants, it does suggest that explicit vocabulary activities, in which collocations are the central focus of attention, seem to be an effective means to make initial form–meaning links in the mental lexicon (Peters, 2014; Webb & Kagimoto, 2011). Participants recalled approximately 40% of the target collocations in the form recall tests and recognized 75% of the items in the recognition test,⁶ which is higher than the learning gains in the incidental learning study by Webb et al. (2013), viz. 15% for form recall and 45% for form recognition. Although it must be acknowledged that participants in the present study needed slightly more time (40–60 minutes) compared to participants in the longest treatment in Webb, Newton, and Chang’s (2013) study (38 minutes), the findings nevertheless suggest that explicit, deliberate teaching activities are beneficial for initial form–meaning mapping of new collocations in a classroom-based course.

4 Limitations and future research

There are of course a number of limitations in this study. Since authentic collocations from students’ course materials were targeted, only a limited number of target items, viz. 18, were selected because it appeared more difficult than expected to find congruent collocations that were unknown to a sufficiently large group of participants. These 18 target collocations were divided into congruent/incongruent collocations and into adjective–noun or (phrasal) verb–noun collocations. The selected items in each subgroup were expected to represent one of these general categories. Although the GEE analysis is based on the number of observations and not on participants’ total score (max = 18), it cannot be ruled out that the limited number of different items played a role.

A second limitation regards the formal (exhaust – exceed; take on – take out) and semantic similarity of some of the collocations (settle an invoice, write off debts, run up a deficit). Since data were collected in a classroom-based experiment (business English) and ecological validity was, thus, considered important, the presentation of formally similar or semantically related verbs was not explicitly avoided (see also Boers et al., 2014a). Future studies could take these two limitations (number of items and similarity) into account by setting up a tightly controlled study, in which semantically and formally unrelated, low-frequency collocations that are likely to be unknown to a large group of participants are selected from a corpus instead of from specific course materials. However, this procedure might be at the expense of ecological validity since it would not represent a learning situation that – at least – these students are very likely to encounter.

Finally, an analysis of participants’ errors suggests that form recall tests might not always be sensitive enough to show all learning gains such as very partial knowledge of the form. Future studies might want to include other measures to assess learning gains, such as free recall or tests tapping into implicit knowledge of collocations (for an example, see Sonbul & Schmitt, 2013).

VII Conclusions

This study aimed to investigate a number of variables that might influence the learning burden of EFL learners’ initial form–meaning mapping of new collocations. The results suggest that depending on the posttest all variables under investigation tend to influence learners’ initial form–meaning mapping of new collocations. Mixed findings were found for congruency because it only played a role when learners had to recall the form of the collocations and not when they had to recognize the form. Second, the findings suggest that adjective–noun collocations tend to be easier to learn than (phrasal) verb–noun collocations. The results also indicate that word length and learners’ vocabulary size might play a role in the initial form–meaning mapping of new collocations. Finally, it seems that it is productive knowledge where learners’ main difficulties with collocations lie, as even controlled productive use presented a serious challenge to the participants in this study.

Footnotes

Acknowledgements

I would like to thank my colleagues Raf Erzeel, Tine Van Houtven and Dirk Van Massenhove for their help with the data collection and Peter Flynn and Paul Pauwels for translating the collocations. I am also grateful to Kris Heylen and Koen Jaspaert for their invaluable help with the statistical analyses. Finally, I would like to thank the anonymous reviewers for their insightful comments.

Declaration of conflicting interests

The author declares that there is no conflict of interest.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Notes

References

Agresti

(2002). Categorical data analysis. 2nd edition. Hoboken, NJ: Wiley-Interscience.

Benson

Ilson

R.F.

(1997). The BBI dictionary of English word combinations. Amsterdam/Philadelphia, PA: John Benjamins.

Boers

Lindstromberg

(2012). Experimental and intervention studies on formulaic sequences in a second language. Annual Review of Applied Linguistics, 32, 83–110.

Boers

Demecheleer

Coxhead

Webb

(2014a). Gauging the effects of exercises on verb–noun collocations. Language Teaching Research, 18, 54–74.

Boers

Lindstromberg

Webb

(2014b). Further evidence of the comparative memorability of alliterative expressions in second language learning. RELC Journal, 45, 85–99.

Campoy

(2008). The effect of word length in short-term memory: Is rehearsal necessary? Quarterly Journal of Experimental Psychology, 61, 724–734.

Cowie

A.P.

Mackin

(1993). Oxford dictionary of phrasal verbs. Oxford: Oxford University Press.

Dagut

Laufer

(1985). Avoidance of phrasal verbs: A case for contrastive analysis. Studies in Second Language Acquisition, 7, 73–79.

Davies

(2004–). The British National Corpus (BYU-BNC): 100 million words, 1980–1993. [Corpus available online.] Retrieved from http://corpus.byu.edu/bnc (December 2014).

10.

Hulme

Maughan

Brown

G.D.A.

(1991). Memory for familiar and unfamiliar words: Evidence for a long-term memory contribution to short-term memory span. Journal of Memory and Language, 30, 685–701.

11.

Hulstijn

J.H.

(1992). Retention of inferred and given word meanings: experiments in incidental vocabulary learning. In: Arnaud

P.L.J.

Béjoint

(Eds.), Vocabulary and applied linguistics (pp. 113–125). London: Macmillan Academic and Profession.

12.

Hulstijn

J.H.

(2001). Intentional and incidental second language vocabulary learning: A reappraisal of elaboration, rehearsal and automaticity. In: Robinson

(Ed.), Cognition and second language instruction (pp. 258–286). Cambridge: Cambridge University Press.

13.

Hulstijn

J.H.

(2003). Incidental and Intentional Learning. In: Doughty

C.J.

Long

M.H.

(Eds.), Handbook of second language acquisition (pp. 349–381). Malden, MA: Blackwell.

14.

Hulstijn

J.H.

Marchena

(1989). Avoidance: Grammatical or semantic causes? Studies in Second Language Acquisition, 11, 241–255.

15.

Laufer

(1997). What’s in a word that makes it hard or easy: some intralexical factors that affect the learning of words. In: Schmitt

McCarthy

(Eds.), Vocabulary: Description, acquisition and pedagogy (pp. 140–155). Cambridge: Cambridge University Press.

16.

Laufer

(2005). Focus on form in second language vocabulary learning. In: Foster-Cohen

S.H.

Garcia-Mayo

Cenoz

(Eds.), Eurosla yearbook: Volume 5 (pp. 223–250). Amsterdam: John Benjamins.

17.

Laufer

(2011). The contribution of dictionary use to the production and retention of collocations in a second language. International Journal of Lexicography, 24, 29–49.

18.

Laufer

Girsai

(2008). Form-focused instruction in second language vocabulary learning: A case for contrastive analysis and translation. Applied Linguistics, 29, 1–23.

19.

Laufer

Rozovski-Roitblat

(2011). Incidental vocabulary acquisition: The effects of task type, word occurrence and their combination. Language Teaching Research, 15, 391–411.

20.

Laufer

Waldman

(2011). Verb–noun collocations in second language writing: corpus analysis of learners’ English. Language Learning, 61, 647–72.

21.

Lea

Crowther

Dignen

(2002) (Eds.). Oxford collocations dictionary for students of English. Oxford: Oxford University Press.

22.

Liao

Fukuya

Y.J.

(2004). Avoidance of phrasal verbs: The case of Chinese learners of English. Language Learning, 54, 191–226.

23.

Milton

(2009). Measuring second language vocabulary acquisition. Bristol: Multilingual Matters.

24.

Montero Perez

Peters

Clarebout

Desmet

(2014). Effects of captioning on video comprehension and incidental vocabulary learning. Language Learning & Technology, 18, 118–141.

25.

Nation

I.S.P.

Webb

(2011). Researching and analyzing vocabulary. Boston, MA: Heinle.

26.

Nesselhauf

(2003). The use of collocations by advanced learners of English and some implications for teaching. Applied Linguistics, 24, 223–242.

27.

Nesselhauf

(2005). Collocations in a learner Corpus. Amsterdam: John Benjamins.

28.

Paquot

Granger

(2012). Formulaic language in learner corpora. Annual Review of Applied Linguistics, 32, 130–149.

29.

Parkinson

(2005). Oxford business English dictionary for learners of English. Oxford: Oxford University Press.

30.

Peng

C.-Y. J.

T.-S. H.

(2002). Logistic regression analysis and reporting: A primer. Understanding statistics, 1, 31–70.

31.

Peters

(2012). Learning German formulaic sequences: the effect of two attention-drawing techniques. Language Learning Journal, 40, 65–79.

32.

Peters

(2014). The effects of repetition and time of post-test administration on EFL learners’ form recall of single words and collocations. Language Teaching Research, 18, 75–94.

33.

Schmitt

(2008) Review article: instructed second language vocabulary learning. Language Teaching Research, 12, 329–363.

34.

Schmitt

Redwood

(2011). Learner knowledge of phrasal verbs. A corpus-informed study. In: Meunier

De Cock

Gilquin

Pacquot

(Eds.), A taste for corpora. In honour of Sylviane Granger (pp. 173–208). Amsterdam: John Benjamins.

35.

Schmitt

Clapham

(2001). Developing and exploring the behavior of two new versions of the vocabulary levels test. Language Testing, 18, 55–88.

36.

Sinclair

(1991). Corpus, concordance, collocation: Describing English language. Oxford: Oxford University Press.

37.

Sinclair

(2001). Collins Cobuild English dictionary for advanced learners. Glasgow, HarperCollins.

38.

Siyanova

Schmitt

(2007). Native and nonnative use of multi-word vs. one-word verbs. IRAL, 45, 119–139.

39.

Sonbul

Schmitt

(2013). Explicit and implicit lexical knowledge: Acquisition of collocations under different input conditions. Language Learning, 63, 121–159.

40.

Swan

(1997). The influence of the mother tongue on second language vocabulary acquisition and use. In: Schmitt

McCarthy

(Eds.), Vocabulary: Description, acquisition and pedagogy (pp. 156–180). Cambridge: Cambridge University Press.

41.

Waibel

(2007). Phrasal verbs in learner English: A corpus-based study of German and Italian students. Unpublished PhD dissertation, Freiburg University, Germany. Retrieved from: http://www.freidok.uni-freiburg.de/volltexte/3592 (December 2014).

42.

Webb

Kagimoto

(2011). Learning collocations: Do the number of collocates, position of the node word, and synonymy affect learning? Applied Linguistics, 32, 259–276.

43.

Webb

Newton

Chang

A. C.-S.

(2013). Incidental learning of collocation. Language Learning, 63, 91–120.

44.

Wolter

(2006). Lexical network structures and L2 vocabulary acquisition: The role of L1 lexical/conceptual knowledge. Applied Linguistics, 27, 741–747.

45.

Wolter

Gyllstad

(2011). Collocational links in the L2 mental lexicon and the influence of L1 intralexical knowledge. Applied Linguistics, 32, 430–449.

46.

Wolter

Gyllstad

(2013). Frequency of input and L2 collocational processing: A comparison of congruent and incongruent collocations. Studies in Second Language Acquisition, 35, 451–482.

47.

Wray

(2002). Formulaic language and the lexicon. Cambridge: Cambridge University Press.

48.

Zahar

Cobb

Spada

(2001). Acquiring vocabulary through reading: Effects of frequency and contextual richness. The Canadian Modern Language Review, 57, 541–572.