LI Influence on the Acquisition of L2 Collocations: Japanese ESL Users and EFL Learners Acquiring English | Collocations |

L1 Influence on the Acquisition of L2 Collocations: Japanese ESL Users and EFL Learners Acquiring EnglishCollege Park, Maryland, United States This study investigated first language L1 i

L1 Influence on the Acquisition of L2 Collocations: Japanese ESL Users and EFL Learners Acquiring English

College Park, Maryland, United States

This study investigated first language (L1) influence on the acquisition

of second language (L2) collocations using a framework based on Krolland Stewart (1994) and Jiang (2000), by comparing the performance

on a phrase-acceptability judgment task among native speakers ofEnglish, Japanese English as a second language (ESL) users, andJapanese English as a foreign language (EFL) learners The testmaterials included both congruent collocations, whose lexical compo-nents were similar in L1 and L2, and incongruent collocations, whoselexical components differed in the two languages EFL learners mademore errors with and reacted more slowly to incongruent collocationsthan congruent collocations ESL users generally performed betterthan EFL learners (lower error rate and faster speed), but they stillmade more errors on incongruent collocations than on congruentcollocations Interestingly, however, the L1 effect was not apparent onthe ESL users’ reaction time The results suggested that (a) both L1congruency and L2 exposure affect the acquisition of L2 collocationswith the availability of both maximizing this acquisition; (b) it isdifficult to acquire incongruent collocations even with a considerableamount of exposure to L2; and (c) once stored in memory, L2collocations are processed independently of L1 Possible differences inacquiring congruent and incongruent collocations are discussed

doi: 10.5054/tq.2010.235998

M ultiword units such as formulaic sequences, idioms, andcollocations have been considered to be important for bothlanguage use and language learning Sinclair’s (1991) idiom principle,for example, recognizes that relatively fixed multiword expressions are

widespread in language use and serve as an organizing principle of thelanguage Acknowledging the importance of a suitable balance betweencreative (rule-based) and holistic (chunk-based) processes, Wray andPerkins (2000) also maintain that our baseline strategy in everydaylanguage use is the holistic processing that relies on the expectedlikelihood of lexical sequences rather than unexpected novelties.Furthermore, unanalyzed multiword units create bases for creativelanguage in language acquisition (Wray, 2002).

In the second language acquisition literature, the importance ofmultiword units has also been well recognized (e.g., Krashen & Scarcella,1978; Weinert, 1995; Wray, 2002) Earlier research, for example,documented widespread formula use by second language (L2) learners(e.g., Fillmore, 1979; Hanania & Gradman, 1977) Multiword units havealso received due attention in models of L2 lexical knowledge (e.g.,Nation, 2001; Richards, 1976) and models of L2 acquisition In usage-based models of language acquisition, for example, much of languageacquisition is considered sequence learning, starting from phonologicalstrings in lexical items and then moving to a probabilistic sequence ofword strings or collocations Only later can syntactic and semanticclassification emerge from the patterns of lexical co-occurrence (Ellis,1996)

Recently, the field has witnessed the emergence of empirical studies

on the representation and processing of multiword units The focus ofattention has been on the issue of whether formulaic sequences arestored and retrieved as a whole in language use Holistic representationand processing would be important because, if it is true, multiword unitswould then contribute to efficiency of language processing both inproduction and comprehension For instance, L2 learners may be able

to speak more fluently and naturally by uttering fixed expressions as oneprosodic unit They may also be able to comprehend texts more quickly

by expecting upcoming words in a fixed phrase If multiword units areprocessed quickly without utilizing much of our attention, we can usemore of our cognitive resources for higher level language processing(e.g., organizing messages or understanding implied meanings).The findings of recent empirical studies are mostly confirmative L2researchers have reported that formulaic sequences are processed morequickly (Conklin & Schmitt, 2008; Jiang & Nekrasova, 2007) and moreaccurately (Jiang & Nekrasova, 2007) than nonformulaic sequences.Words in formulaic sequences are read faster than the same words innonformulaic sequences (Underwood, Schmitt, & Galpin, 2004) Bothfigurative and literal meanings of idioms are processed more quicklythan nonformulaic controls (Conklin & Schmitt, 2008) Thus, emergingevidence supports the processing advantage of formulaic units overnonformulaic strings because of the holistic processing of the former

These recent studies, however, have mostly focused on formulaicsequences Another important type of multiword unit—collocations—has received little attention Collocations, that is, multiword units thatconsist primarily of open class items, such as heavy traffic, strong tea, andkill time, may share certain characteristics with formulas For example,both types of multiword units may be stored holistically as single lexicalunits, at least in the case of highly frequent, and probably short (Schmitt

& Carter, 2004), collocations However, collocations are also differentfrom formulaic sequences (e.g., at first, in front of, on the other hand) insome important ways

First, collocations are looser combinations of words than formulaicsequences, in the sense that a component word in a collocation maycollocate frequently with many other words to form other collocations.For example, ‘‘heavy’’ not only collocates with ‘‘traffic’’ but also with

‘‘stone’’ and ‘‘smoker’’, while ‘‘in front of’’ makes a unit in which theconstituent words cannot be replaced by another However, suchflexibility is limited For instance, ‘‘argument’’ collocates both with

‘‘strong’’ and ‘‘powerful’’ (strong argument, powerful argument), but

‘‘car’’ collocates only with ‘‘powerful’’ (*strong car, powerful car) and

‘‘tea’’ collocates only with ‘‘strong’’ (strong tea, *powerful tea; Carter &McCarthy, 1988, p 34)

Second, formulaic sequences are often language specific, in the sensethat a formulaic expression in one language may not readily have acounterpart consisting of identical or similar lexical items in anotherlanguage For instance, the Japanese language does not have multiwordexpressions corresponding word by word to English formulaic sequencessuch as look forward to or out of order The same concepts are expressed bydifferent lexical items, and learning a formulaic sequence resembleslearning a new lexical item in the L2 In contrast, collocations are oftencross-linguistic, in that a collocation in one language usually has acounterpart in another language except when culture-specific conceptsare involved Thus it is likely that many languages have a collocationsimilar to strong wind, although whether it is represented as a singlesemantic unit may differ across languages However, the specific lexicalitems used to refer to the concept vary between languages Twolanguages may share identical lexical items or have different componentwords For example, both English and Japanese have the identicalcollocation of hot tea, but what is referred to as strong tea in English iscalled dark tea in Japanese (other possible translations would be dense teaand thick tea) Similarly, English and Chinese share the collocation greentea, but black tea in English becomes red tea in Chinese Based on thiscross-linguistic relationship, a distinction may be made betweencongruent collocations and incongruent collocations The former share

identical lexical elements between two languages, and the latter involvedifferent words.

These two unique features of collocations, that is, the flexibility oftheir component words in recombining to form other collocations andtheir cross-linguistic nature, carry significant consequences for learning.The flexibility makes them less salient as multiword units As a result, alearner may treat a collocation as a regular phrase without noticing itand thus without learning its particular combination of lexicalcomponents As a result, the learner may combine words too liberally,producing inappropriate collocations More important, their cross-linguistic nature, in combination with their flexibility, makes learningcollocations particularly susceptible to first language (L1) influence.Such influence may be facilitative at times but inhibitory at other times(Wolter, 2006) When an L2 learner first encounters a new collocation,

he or she can understand its meaning immediately in the case of acongruent collocation if the component words are known However, anincongruent collocation may be difficult for learners to understand andthus take much longer to learn

The difficulty collocations create for L2 learners are well ted It can be seen in frequent discussions of collocations in pedagogicalcontexts (e.g., Brown, 1974; Holmes & Brown, 1987; Lewis, 2000a;Richards, 1976; Schmitt, 1999) There is also research evidence showingthat even advanced L2 learners often have difficulty with collocations(e.g., Altenberg & Granger, 2001; Dechert & Lennon, 1989; Hyltenstam,1988; Lennon, 1996; Nesselhauf, 2003, 2005)

documen-The purpose of the present study is to examine the acquisition ofcollocations by L2 learners, with a special focus on the role of L1influence The dependence of L2 lexical processing on L1 lexicon hasbeen well demonstrated (e.g., Jiang, 2002, 2004; Kroll & Stewart, 1994;Sunderman & Kroll, 2006; Talamas, Kroll, & Dufour, 1999) This findingprovides good grounds for hypothesizing that L1 collocation patternsmay influence the acquisition of L2 collocations However, seriousattempts to test L1 influence on the acquisition of L2 collocations are yetvery limited Some corpus-based research has emphasized the L1influence on the production of L2 collocations (e.g., Altenberg &Granger, 2001; Nesselhauf, 2003) The present study adopted a moreexperimental approach examining the error rate and reaction time,which has some advantages over corpus-based studies First, becauselinguistic expressions are prepared beforehand according to thetheoretical interest in L1–L2 differences, the study may provide a morestraightforward ground to test the possibility of L1 influence, whereas inthe corpus analysis the causal interpretation is usually post hoc Second,the data are not likely to be affected by learners’ avoidance strategies,which could be the case with respect to corpus-based data Third, we can

obtain insights not only into the learners’ knowledge representation ofcollocations but also into how they process collocations.

Our study was conceived and designed within a conceptual frameworkbased on Kroll and Stewart (1994) and Jiang (2000) The revisedhierarchical model (Kroll & Stewart, 1994) postulates three memoryrepresentations (concepts, L1 lexicon, and L2 lexicon) and connectionsamong them There are two levels in the connection: the conceptuallevel (i.e., the link between concepts and either the L1 or L2 lexicon)and the lexical level (i.e., the link between the L1 and L2 lexicons) Thelink between the L1 lexicon and concepts is assumed to be strong,whereas the link between the L2 lexicon and concepts is relatively weak.The lexical link from L2 to L1 (i.e., the association of L2 words withcorresponding L1 words) is hypothesized to be stronger and morereliable than the link from L1 to L2 This is because the L1 lexiconnormally contains a larger number of lexical items than the L2 lexicon

It is, therefore, relatively simple and quick to find an L1 word thatcorresponds to an L2 word (Kroll & Tokowicz, 2001) Because of thesedifferences in connections between concepts and independent lexicalrepresentations in the L1 and L2, we can hypothesize that, when L2learners try to understand an L2 word, they first translate it into the L1and use the strong link from the L1 word to its concept The linkbetween the L2 lexicon and concepts will become stronger withincreased exposure to the L2, and gradually L2 learners will be able toaccess meaning directly from L2 words

In his model of the L2 mental lexicon, Jiang (2000) also postulatedthree memory representations and connections among them, but wasmore explicit about the role of L1 transfer, which is the centralmechanism in the acquisition of L2 vocabulary Lemma and lexeme aredistinguished, and L1 lemma (syntax and semantics) is transferred tothe L2 lexicon and influences L2 processing in the course ofdevelopment The model specifies three developmental stages The firststage is word association, where a new L2 word form enters the L2lexicon but is strongly connected with its L1 translation There is no linkbetween the L2 lexicon and the concept The second stage is L1 lemmamediation, where L1 lemma information is transferred into the L2lexicon The transferred lemma information links L2 words andconcepts, albeit weakly Activation of L1 translation in L2 use maydecrease in this stage, but transferred L1 lemma information alwaysmediates L2 use With increased exposure to the L2, the link betweenthe concept and the L2 lexicon develops and L2-specific lemmainformation emerges in the L2 lexicon The third and final stage isfull integration, where L2-specific information replaces transferred L1information, and the L2 lexicon is linked to the concept as strongly as isthe L1 lexicon

Applying these models, we assume that the concept referred to by anL2 collocation already exists in a learner’s mind.1Therefore, the essence

of learning an L2 collocation is to store a new collocation in the L2lexicon and link it to an existing concept This process may entail at leastthe following three specific steps The first step is the recognition of aword string as a legitimate collocation in the L2, which is typicallyaccompanied by the initial comprehension of the meaning of the newcollocation The second is the integration of the collocation in one’slong-term memory or L2 mental lexicon, as a result of repeatedexposures During the first two steps, an L2 collocation may be linkedprimarily to its counterpart in the L1, or to both its L1 counterpart andthe existing concept, but with a stronger link to the former than thelatter The third step is the establishment of a direct connection betweenthe new collocation and the concept that is strong enough to allow it tofunction automatically and autonomously

This framework helps us to make some specific predictions about thelearning of congruent and incongruent collocations in examining therole of the L1 We can reasonably hypothesize that the learning ofincongruent collocations will lag behind that of congruent ones Unlikecongruent collocations, whose meaning is more readily understandablebecause of assistance from their L1 counterparts, the meaning of anincongruent collocation has to be obtained through the analysis of themeanings of its component words, often with help from its linguistic ornonlinguistic contexts However, contexts may not provide enoughinformation for the learner to make a successful inference Therefore,much more input or more frequent encounters with an incongruent L2collocation would be necessary in order for L2 learners to recognize thelegitimacy of an incongruent collocation than a congruent one

In the same vein, we can also anticipate faster processing of congruentcollocations than incongruent ones at early stages of acquisition.Assuming that its meaning has to be activated in order for a collocation

to be recognized and understood, the recognition of a congruentcollocation is much faster because of its link to an existing L1collocation, which is linked to meaning The processing of anincongruent collocation, however, can be a more deliberate and slowerprocess, in the sense that it involves the rejection of its word-for-word L1translation and the finding of an L1 collocation that shares the samemeaning, before a strong direct connection is established between theL2 collocation and its concept

1 Some collocations may express a concept that is culture or language specific Narrow escape, which does not have a ready counterpart in Chinese, can be said to express a concept that

is not present, or at least not lexicalized, in the conceptual system of a Chinese native speaker.

Such involvement of L1 collocations in L2 processing has beenpowerfully demonstrated by Ueno (2006), who showed that it is not onlyL1 translation equivalents of single words but also L1 lexical networksthat are activated when L2 learners recognize an L2 word It wasreported, as a case in point, that the English prime word forgivefacilitated the processing of marriage by Japanese English as a foreignlanguage (EFL) learners, whereas there was no such facilitation fornative speakers of English Forgive marriage is not a legitimate expression

in English, but it is in Japanese According to Ueno, when the EFLlearners saw forgive, its translation equivalent in Japanese was activatedand primed marriage through an L1 collocational link

It seems reasonable to hypothesize further that, with increasing input,L2 collocations, both congruent and incongruent, may become multi-word units that are no longer dependent on the L1 lexicon and that aredirectly connected to concepts As a result, advanced learners wouldshow little difference in processing time between the two types ofcollocations

The present study used a phrase-acceptability judgment task ing both accuracy and speed of the performance of native speakers ofEnglish and native speakers of Japanese processing English collocations

examin-In order to see the L2 developmental change, two groups of Japanesespeakers who largely differed with respect to their exposure to Englishwere included In terms of the length or span of collocations, onlyimmediate connections of words (short collocations consisting ofbasically two words) were targeted This was because our primary focuswas L1 influence, and there was no specific need to involve longer wordcombinations, which would have made the task more difficult forparticipants

In light of the role of L1 congruency in L2 collocation acquisitiondiscussed earlier, several possible scenarios can be expected when we testnative speakers of English, EFL learners, and English as a secondlanguage (ESL) users on congruent and incongruent collocations First,English native speakers should show no reliable difference betweencongruent and incongruent collocations in both response time anderror rate, because the test materials were matched in length, lexicalfrequency, and collocation frequency Second, lower-proficiency EFLlearners should produce a congruency effect, in that they should makemore errors on and respond more slowly to incongruent collocationsthan congruent ones Third, ESL users of higher proficiency mightproduce a congruency effect similar to EFL learners if the L1 continues

to affect the representation and processing of L2 collocations However,

it is also possible, as hypothesized earlier, that, given the increasedamount of input, ESL users are able to develop more autonomousrepresentations for L2 collocations and, as a result, both congruent and

incongruent collocations can function independently with little ence from their congruency with their L1 counterparts In this case, nocongruency effect would be observed on ESL speakers.

influ-METHODS

Participants

The participants included 20 native speakers of English, 24 JapaneseESL users, and 23 Japanese EFL learners The native speakers werestudents at a university in the United States, and the ESL users werestudents, researchers, or instructors at four universities in the UnitedStates Participants in these groups were residing in the United States atthe time of testing and were tested in the United States The EFLlearners were students at a university in Japan They were residing inJapan at the time of testing and were tested in Japan They had neverlived in English-speaking countries, and their exposure to English hadbeen limited largely to language classes where they studied English as asubject All the Japanese EFL and ESL participants were native speakers

of Japanese who were born in Japan and who started formal education ofEnglish in the seventh grade in Japan A 58-item cloze test wasadministered to the Japanese participants for the purpose of estimatingthe difference in their L2 proficiency This cloze test has been used inYamashita (2007) and Yamashita and Ichikawa (2010) and demonstratedhigh internal consistency (Cronbach’s alpha 0.90) The mean scoreswere significantly different between the ESL and EFL groups (t(45) 59.19, p , 0.01) Table 1 summarizes the Japanese participants’ English

TABLE 1

English Learning Background of Japanese Participants

ESL users EFL learners Mean SD Mean SD Cloze 26.92 6.62 12.96 3.11 Age starting English study 12.21 0.78 12.00 1.24 Years of formal English education 8.42 1.74 6.87 0.92 Years of residence in English-speaking countries 5.39 4.62 0.00 0.00 Age 31.42 7.73 18.91 0.79 Self-ratingsa

Speaking 6.08 1.64 2.00 1.14 Listening 6.88 1.70 2.33 1.28 Reading 6.83 0.92 3.00 1.64 Writing 6.21 1.10 2.67 1.56 Note ESL 5 English as a second language; EFL 5 English as a foreign language; SD 5 standard devation.aRatings scored on a 10-point scale with 1 5 minimum proficiency and 10 5 near- native proficiency.

learning background obtained in a questionnaire administered eitherimmediately before or after the experiment As shown in the table, thetwo groups were not very different in the length of formal education ofEnglish (approximately 7 vs 8 years), but the exposure to English out ofthe classroom was considerably different (0 vs approximately 5 years onaverage).

Materials

The materials consisted of 24 congruent collocations, 24 incongruentcollocations, and 48 implausible word combinations The 24 congruentcollocations have counterparts in Japanese (i.e., those collocations can

be directly translated word for word between English and Japanese, e.g.,make lunch and heavy stone) The incongruent collocations cannot betranslated directly word for word from English to Japanese (e.g., kill timeand slow learner) In order to express the same meaning of Englishcollocations, the Japanese language has to use different word combina-tions (e.g., crush/break time for kill time) or paraphrasing (e.g., a person wholearns slow for slow learner)

The materials were constructed using the following procedures First,

a number of potential English collocations were identified by the firstauthor using three English–Japanese dictionaries Then the most typicaltranslation of each component word of the English collocations waschecked, because congruency is determined by the translation relation-ships between the languages All the single words constituting thecandidate collocations were listed and given to five native speakers ofJapanese with high proficiency in English They were asked to writedown as many Japanese words as they could think of as a translation of agiven English word The English words remained in the pool of possibleitems if (a) the meanings were known to at least four of the translators,(b) the translations by at least four translators were equal to the meaning

of a critical word in congruent collocations, and (c) the translations by atleast four translators were different from the meaning of a critical word

in incongruent collocations For example, in the case of kill animals(congruent) and kill time (incongruent), the critical word kill was firsttranslated by all translators into korosu (5 cause the death) as in killanimals, whereas only one added (jikan-o) tsubusu (5 past [time]) as thesecond translation as in kill time If Japanese speakers go through the L1-mediation process when they understand these English collocationsincluding kill, the L1 word corresponding to cause the death (korosu) ismost likely to be the one to which they first associate the English wordkill If this is true, kill animals may be processed more quickly than kill

time, because learners should be able to more easily associate kill with themeaning of cause the death via the L1 word korosu.

After the translation of each word was sorted out, the translations ofcollocations were checked A list of collocations made up of theremaining words was given to another group of five native speakers ofJapanese with high English proficiency They were asked to translategiven English collocations into Japanese so that the resulting Japaneseexpressions sounded as natural as possible The English collocationsremained in the pool of possible items if (a) the meanings ofcollocations were known to at least four of the translators, (b) thetranslation by at least four of the translators was a literal word for wordrendering in the case of congruent collocations, and (c) the translation

by at least four of the translators was a rendering composed of differentlexical items in Japanese or paraphrasing in the case of incongruentcollocations Both length and frequency of the materials were matched

as closely as possible between the lists of congruent and incongruentcollocations Lexical and phrasal frequencies were obtained fromShogakukan Corpus Network, British National Corpus, available onthe Web When computing phrasal frequencies, no lexical item wasallowed between the two words in adjective + noun combinations, but apossible insertion of one word was allowed between two words in verb +noun combinations in order to accommodate possible insertion of anarticle T-tests confirmed there were no differences between thecongruent and incongruent conditions in terms of length; t(46) 520.060, p 0.05, word frequency; t(46) 5 20.152, p 0.05, or phrasalfrequency; t(46) 5 20.828, p 0.05

Most individual words in the final material were within the mostfrequent 3,000-word level in the JACET List of 8,000 Basic Words(JACET Committee of Basic Words Revision, 2003).2This is the level ofhigh-school textbooks in Japan Words belonging to lower frequencylevels included three words at the 4,000-word level (the beginning level

of university), a derivative of a high-frequency word (learner), and aword that exists in Japanese as a loan word (tempo) Therefore, weassumed that individual words were known to both EFL learners andESL users Implausible word combinations served as fillers, inducingnegative responses from participants These word combinations weresemantically implausible, such as abstract fruits and begin a bed Theinclusion of the implausible items was for task requirement only, andthese items were not analyzed, because their data were not relevant tothe purpose of the study

2 This vocabulary list was constructed by a research team from the Japan Association of College English Teachers.

The participants were tested individually in a quiet room The itemswere presented one at a time on a computer screen in random order Thetask was to decide whether a test item was an acceptable Englishexpression The participants responded by pressing one of the two keys on

a keyboard, one for YES and the other for NO They were asked to respond

as quickly and as accurately as possible Twenty practice items were givenprior to the 96 test items The criteria of acceptability were easily mastered

by all participants by going through the practice items The experimentalsession started with instructions, followed by the practice session, and thetest session Both the practice and test items began with a fixation point,

an asterisk, presented at the center of the monitor for 500 milliseconds.The asterisk was followed by an item, to which the participant responded

A questionnaire was given to the Japanese participants, either before orafter the experiment, to obtain information about their English learningbackground The presentation of test materials and the recording of thereaction times and error rates were accomplished by using DMDX, acomputer program developed by K Forster and J Forster at the University

of Arizona (Forster & Forster, 2003)

RESULTS

In the computation of reaction time means, all incorrect responseswere excluded Based on the practice in the field and our earlierexperience, any response that was two standard deviations away fromeach participant’s mean, shorter than the low cutoff point at 400 milli-seconds, or longer than the high cutoff point at 3,000 milliseconds was

TABLE 2

Mean Reaction Times (in Milliseconds) and Error Rates (in Percent) on Congruent and Incongruent Collocations by Native Speakers of English (NS) and Japanese ESL and EFL Groups

Mean SD Mean SD Mean SD Reaction time Congruent 986 175 1,388 184 1,542 300

Incongruent 980 167 1,385 221 1,597 287 Difference 26 22 55*

Error rate Congruent 6.79 7.31 8.35 6.50 14.31 10.38

Incongruent 9.39 9.16 19.83 8.30 38.80 10.78 Difference 2.60 11.48* 24.49*

Note Difference 5 the difference between congruent and incongruent items.

*p , 0.05.