An experimental study of early L3 development: age, bilingualism and classroom exposure

24
This article was downloaded by: [UQ Library] On: 20 November 2014, At: 03:56 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK International Journal of Multilingualism Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/rmjm20 An experimental study of early L3 development: age, bilingualism and classroom exposure Catherine A. Stafford a , Cristina Sanz b & Harriet Wood Bowden c a Department of Spanish and Portuguese , University of Wisconsin- Madison , Madison, WI, USA b Department of Spanish & Portuguese , Georgetown University , 412 ICC, 37 & O Streets NW, Washington, DC, 20057, USA c Department of Modern Languages , University of Tennessee- Knoxville , Knoxville, TN, USA Published online: 15 Jan 2010. To cite this article: Catherine A. Stafford , Cristina Sanz & Harriet Wood Bowden (2010) An experimental study of early L3 development: age, bilingualism and classroom exposure, International Journal of Multilingualism, 7:2, 162-183, DOI: 10.1080/14790710903528122 To link to this article: http://dx.doi.org/10.1080/14790710903528122 PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &

Transcript of An experimental study of early L3 development: age, bilingualism and classroom exposure

Page 1: An experimental study of early L3 development: age, bilingualism and classroom exposure

This article was downloaded by: [UQ Library]On: 20 November 2014, At: 03:56Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

International Journal of MultilingualismPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/rmjm20

An experimental study of early L3development: age, bilingualism andclassroom exposureCatherine A. Stafford a , Cristina Sanz b & Harriet Wood Bowden ca Department of Spanish and Portuguese , University of Wisconsin-Madison , Madison, WI, USAb Department of Spanish & Portuguese , Georgetown University ,412 ICC, 37 & O Streets NW, Washington, DC, 20057, USAc Department of Modern Languages , University of Tennessee-Knoxville , Knoxville, TN, USAPublished online: 15 Jan 2010.

To cite this article: Catherine A. Stafford , Cristina Sanz & Harriet Wood Bowden (2010)An experimental study of early L3 development: age, bilingualism and classroom exposure,International Journal of Multilingualism, 7:2, 162-183, DOI: 10.1080/14790710903528122

To link to this article: http://dx.doi.org/10.1080/14790710903528122

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoeveror howsoever caused arising directly or indirectly in connection with, in relation to orarising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &

Page 2: An experimental study of early L3 development: age, bilingualism and classroom exposure

Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 3: An experimental study of early L3 development: age, bilingualism and classroom exposure

An experimental study of early L3 development: age, bilingualism andclassroom exposure

Catherine A. Stafforda, Cristina Sanzb* and Harriet Wood Bowdenc

aDepartment of Spanish and Portuguese, University of Wisconsin-Madison, Madison, WI, USA;bDepartment of Spanish & Portuguese, Georgetown University, 412 ICC, 37 & O Streets NW,Washington, DC 20057, USA; cDepartment of Modern Languages, University of Tennessee-Knoxville, Knoxville, TN, USA

(Received 9 June 2009; final version received 27 November 2009)

This research investigated Spanish�English bilingual adults’ initial learning of athird language (L3), Latin, comparing the learning processes and outcomes ofearly- and late-onset bilinguals. Thirty-three participants were classified as Earlyor Late Bilinguals according to their age of arrival to the USA, and they wereintroduced to Latin by means of an interactive computer programme thatprovided grammar explanation, input-based practice and explicit feedback.Results indicated that (1) overall, the two groups garnered similar benefits fromthe instructional treatment; (2) the Late Bilinguals maintained improvements inaccurately marking noun case morphology in L3 production somewhat betterthan Early Bilinguals; and (3) there is a role for current age in explaining thevariance observed in initial L3 learning, particularly when a target structure ismorphologically complex and memory cannot be supported by existing L1/L2knowledge.

Keywords: age; input processing; bilingualism; third language acquisition;attention; multilingualism

Introduction

Bilinguals1 are often regarded as particularly talented language learners, able to use

their linguistic experience to their advantage as they undertake learning a new

language. Empirical evidence has demonstrated facilitative effects of bilingualism on

subsequent language learning (e.g. Cenoz & Valencia, 1994; Hernandez, Sierra, &

Bates, 2000; Klein, 1995; Lasagabaster, 2000; Munoz, 2000; Sanz, 2000a, 2007), and

Bialystok (2001, 2007) has forwarded a particularly compelling explanation that may

account for the bilingual advantages observed in these studies. She has argued that

bilinguals, through their unique experience with constantly managing two languages

in daily use, develop exceptional capacity for attentional control which enables them

to focus attention on information that is relevant to a particular task while inhibiting

attention to irrelevant information.

It remains to be seen, however, whether such cognitive advantages of bilingualism

are (1) similar in both early and late bilinguals and (2) available to facilitate not only

fluency in known languages, but also learning of a new language in adulthood.

*Corresponding author. Email: [email protected]

International Journal of Multilingualism

Vol. 7, No. 2, May 2010, 162�183

ISSN 1479-0718 print/ISSN 1747-7530 online

# 2010 Taylor & Francis

DOI: 10.1080/14790710903528122

http://www.informaworld.com

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 4: An experimental study of early L3 development: age, bilingualism and classroom exposure

Bialystok (2007, p. 220) has called for language learning research involving bilinguals

to include examination of such questions given that ‘[t]he ‘‘bilingualism’’ of an

individual coexists with a constellation of other descriptions, including the age,

cognitive level, specific language competence, and a variety of social and motivational

factors that jointly determine the mental capacity and cognitive change that will

follow’.

This study sought to contribute to the growing field of third language (L3)

acquisition research, for which interest has been increasing steadily in recent years

(Sanz & Lado, 2008), and move towards a deeper understanding of bilingualism and

language learning in adulthood by investigating the role of age of onset of L2

learning, L2 proficiency, current age and experience with foreign language learning

(FLL) in the early L3 development of Spanish�English bilinguals learning Latin

through a computer-delivered lesson.

Literature review

The theoretical framework of the Competition Model (CM) developed by Bates and

MacWhinney (1982, 1989; see also MacWhinney, 2005) has been fruitfully adopted

in research that takes a functionalist approach to the investigation of language

learning. According to the CM, the strength of cues in linguistic input governs the

mapping between linguistic forms and their functions. A cue’s strength is determined

by its validity, which refers to availability, or the frequency with which the cue marks

a particular function, and reliability, or the consistency with which an available cue

uniquely marks a function. The relative strength of cues in a given language creates a

cue hierarchy and differences among hierarchies are posited to account for variation

across languages.

The appeal of the CM’s characterisation of linguistic representation as a set of

changeable, weighted mappings between form and function is particularly strong for

research involving bilinguals. To date, research under the CM framework has focused

on comparisons of monolingual and bilingual processing; however, valuable

extensions of this work include investigations of additional language learning by

adult bilinguals with a range of linguistic experience. After all, a framework such asthe CM is advantageous in its ability to capture the inherent dynamism of both

bilingual and multilingual systems. It is for this reason that the CM was selected as

the theoretical framework for this study.

Studies conducted by Bates, MacWhinney and their colleagues have compared

monolingual and bilingual participants’ performance on an agency assignment task

(i.e. deciding who did what to whom). Target sentences consist of two nouns and a

transitive verb and participants decide what is the agent, or subject, of each sentence.

Sentences are manipulated so that the cues they contain variously ‘converge on’ or

‘compete for’ agency assignment. In addition, these converging and competing cues

may be valid in one, the other or both of the bilinguals’ languages. In this way, at

times participants must make their agency choices based on contradictory evidence

from cues that also differ in their strength in the two languages they speak. A cue

hierarchy is thus revealed in the trade-offs that bilinguals make in their choices of

agency, and patterns in task performance are taken as evidence for differentiation

(using one language’s cue hierarchy under some processing conditions and the other

hierarchy under other conditions), transfer (adopting the cue hierarchy of one

International Journal of Multilingualism 163

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 5: An experimental study of early L3 development: age, bilingualism and classroom exposure

language to guide interpretation of the other) or amalgamation of cues in one

hierarchy.

Results of CM studies (e.g. Hernandez et al., 2000; Kempe & MacWhinney, 1998;

Su, 2001) have revealed a general pattern of amalgamation among early bilinguals,

but not without exception. In Hernandez et al.’s (2000) study, for example, early

Spanish�English bilinguals’ patterns of agency choice were comparable to those of

Spanish monolinguals, suggesting differentiation rather than amalgamation. The

researchers attributed this result to the fact that two out of the three cues included in

their task are valid only in the bilinguals’ L1. They argued that this situation

compelled bilinguals to adopt a monolingual processing mode different from

strategies they might have been expected to invoke in the presence of other

combinations of cues.

The present study extends the body of CM research by investigating bilinguals’

patterns of cue use as they interact with a new language. As a study that focuses on

initial L3 learning, its approach differs from that of previous CM studies in the

inclusion not only of cues that carry different weights in the participants’ L1 and L2,but also of a new morphosyntactic cue. Examining initial L3 learning in this way may

help shed light not only on how adult language learners with different experiential

profiles apply their existing knowledge in order to process and encode new language,

but also on how varied L2 experience facilitates or hinders the incorporation of a

new cue into existing hierarchies.

Results of several studies comparing L2 and L3 learners have indicated that early

bilinguals enjoy more successful outcomes than monolinguals (e.g. Cenoz &

Valencia, 1994; Klein, 1995; Sanz, 2000a; but see Gibson, Hufeisen, & Libben,

2001, for results showing no such difference). Furthermore, studies that have focused

on L3 learning among bilingual children and adolescents (e.g. Lasagabaster, 2000;

Munoz, 2000; Sanz, 2007) have found that balanced bilingualism is associated with

more successful L3 outcomes.

The studies comparing L2 and L3 learning have focused on either overall L2/L3

achievement or acquisition of a particular aspect of the L2/L3 and, as noted by

Cenoz (2003), the findings have varied accordingly, with more consistent patterns of

bilingual advantages emerging in the results of studies focusing on overallproficiency. Cenoz and Valencia (1994), for example, investigated L2/L3 achievement

among 320 classroom English learners who were either Spanish monolinguals or

early Spanish�Basque bilinguals. L2/L3 achievement was operationalised as perfor-

mance on a battery of English tests that evaluated a variety of language skills. When

the combined influence of motivation, intelligence, years of formal English

instruction and age was held constant, bilingualism’s contribution to predicting

English language achievement was found to be statistically significant.

The findings of Sanz (2000a) replicated those of Cenoz and Valencia. In her

study, 77 monolingual Spanish and 124 early Catalan�Spanish bilingual secondary

school students were compared in their performance on a variety of English

proficiency measures. Results indicated that bilingualism was a significant predictor

of performance in English when the influence of motivation and language exposure

was controlled.

Klein’s (1995) study focused more narrowly on how 32 monolingual or multi-

lingual adolescents learning English as an L2 or L3/L4 reset the prepositionstranding parameter assumed by Universal Grammar. Monolingual participants had

164 C.A. Stafford et al.

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 6: An experimental study of early L3 development: age, bilingualism and classroom exposure

started acquiring L2 English in early adolescence and the multilinguals had learned

their first L2 as children. Results revealed that, while both monolinguals and

multilinguals appeared to follow the same route to resetting the parameter, the

multilinguals reset it more rapidly than their monolingual counterparts.

Gibson et al. (2001) likewise selected a narrow linguistic focus for their study of

controlled production of prepositional verbs in L2/L3/L4 German. Unlike Klein,

however, they found no advantage for bilingual/multilingual learners over mono-

linguals. Participants in the study were 64 adult learners from a variety of L1/L2

backgrounds, and all were studying German formally and living in Germany at the

time of the study. Results revealed no statistically significant performance difference

on a gap-fill task between learners of L2 German and L3/L4 German.

The incongruous results of Klein’s and Gibson et al.’s studies may be accounted

for by differences in exposure to the target languages as well as methodological

differences; however, they also may be interpreted as suggesting that the bilingual

advantage does not hold equally for lexical and morphosyntactic development.

Clearly, more research is needed, and in particular studies that examine specificfeatures of L3 proficiency, so that the advantages of bilingualism might be more

thoroughly characterised and explained.

In research comparing monolinguals and bilinguals in non-linguistic areas of

cognition, Bialystok and her colleagues (e.g. Bialystok, 1988; Bialystok & Martin,

2004; Martin-Rhee & Bialystok, 2008) have found that bilingual children develop a

capacity to focus attention in the face of distraction earlier than their monolingual

peers. They have also found that this capacity is maintained through adulthood and

may delay declines in cognitive functioning that occur with ageing (Bialystok, Craik,

Klein, & Viswanathan, 2004; Bialystok, Craik, & Ruocco, 2006). The combined

results of these studies suggest that the beneficial effects of bilingualism may be

cumulative, a notion which has received some measure of empirical support in recent

L3 research (e.g. Lasagabaster, 2000; Munoz, 2000; Sanz, 2007).

Munoz (2000) compared proficiency in three languages among 394 Catalan�Spanish/Spanish�Catalan bilingual children and adolescents learning English as a

foreign language in Catalonia, Spain. Participants varied in language dominance and

home use of Catalan and Spanish; however, all participants attended schools inwhich the minority language, Catalan, was the language of instruction. Proficiency

was assessed in all three languages with cloze and dictation tests and results of score

analysis revealed strong, positive correlations, thus supporting the researcher’s

hypothesis that participants’ L3 competence was associated with their L1 and L2

proficiency.

Lasagabaster (2000) likewise examined achievement in instructed L3 English in

relation to L1 and L2 competence among 252 Spanish�Basque bilingual children and

early adolescents. Spanish was the L1 for most of the students2 and Spanish is also

the majority language of the communities involved in the study. Students’ degree of

balance in L1/L2 proficiency varied, as did the proportion of the school day during

which L2 Basque was used as the language of instruction in the school systems.

Results of Basque reading, writing and listening tests indicated that proficiency in

Basque varied as a function of the amount of instruction delivered in Basque, with a

similar relationship obtaining between amount of Basque instruction and perfor-

mance on L3 English tests. Put another way, students’ degree of bilingualism was

positively associated with L3 learning outcomes.

International Journal of Multilingualism 165

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 7: An experimental study of early L3 development: age, bilingualism and classroom exposure

Sanz obtained similar results in her (2007) study of 120 Catalan�Spanish

bilingual adolescents learning L3 English in Catalonia. Specifically, when the effects

of motivation and L3 exposure were controlled, more balanced bilingualism

(including biliteracy) was significantly predictive of performance on various

measures of L3 grammatical (but not lexical) proficiency.

To summarise, research in bilingualism and cognition has shown that the

favourable effects of bilingualism are cumulative. Furthermore, research in L2 and

L3 learning has found that early bilinguals by and large are more successful than

their monolingual counterparts at acquisition of new language in late childhood and

adolescence, so if the benefits of bilingualism are indeed cumulative, it would follow

that early bilinguals will also be more successful than later bilinguals at learning new

language. This is the central hypothesis that the present study set out to test.Findings also suggest that other factors that covary with early and late

bilingualism � such as exposure to formal instruction in non-primary languages �may facilitate successful L3 learning in childhood and adolescence. Interactions

among cognitive and social variables are bound to be more complex among adults,

however, so empirical L3 research that focuses on post-adolescent bilinguals is

needed. An important additional consideration for research involving post-adoles-

cents is the potential influence of chronological age on language learning in

adulthood. Given the processing-oriented nature of our instructional treatment

and the fact that processing speed is identified among cognitive functions affected by

normal ageing (Park, 2000), current age must certainly be taken into account.

This study investigated whether the age of onset of L2 acquisition had any

bearing on L3 learning later in life, and also examined the degree to which other

factors that covary with age of onset (current age, L2 proficiency and formal

language learning) moderate successful L3 outcomes. The following research

questions guided the study: (1) Do early and late Spanish�English bilingual adults

exhibit similar patterns of cue reliance at the initial stages of L3 learning in a

computer-based instructional context? (2) Do bilinguals’ age, L2 proficiency and

experience with classroom FLL moderate successful adjustment of cue hierarchies in

the L3? To answer these questions we compared early and late Spanish�English

bilingual adults’ initial L3 learning under computer-delivered, explicit instructional

conditions, focusing on morphosyntactic cues related to the assignment of thematic

agent/patient roles to nouns in Latin.

Method

Participants

Participants were biliterate, bilingual Latino adults living in the USA. They were

recruited through advertisements in Spanish-language media, listservs, posted flyers

and by word of mouth, and they were paid for their participation. The 33

participants who comprised the final sample varied in age of onset (AoA) of L2

learning, which was operationalised for this study as the age participants reported

they arrived to the USA. Those who reported an AoA of 12 years or younger were

classified as Early Bilinguals (n�15), and those who reported an AoA of 16 years or

older were classified as Late Bilinguals (n�18). These groupings are in line with

previous research investigating age effects on SLA (e.g. Birdsong & Molis, 2001;

166 C.A. Stafford et al.

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 8: An experimental study of early L3 development: age, bilingualism and classroom exposure

DeKeyser, 2000; Johnson & Newport, 1989), in which ‘early’ bilingualism is broadly

taken to have a pre-pubertal onset of development and includes both simultaneous

and sequential bilinguals, and ‘late’ bilingualism is understood to have a post-

pubertal onset. AoA ranged from birth to age 12 among Early Bilinguals and from

age 16 to 39 among Late Bilinguals.

All participants reported that Spanish was the language they spoke first at home

and that they had no knowledge of Latin and no functional knowledge of other case

inflection languages.3 Thus, all participants were on equal footing in terms of being

naıve learners of noun case morphology. Participants rated their listening, reading,

writing and speaking skills in English and Spanish on seven-point Likert-type scales.4

Participants who rated themselves 5 (‘good’) or below in current L2 proficiency in

any of the four language modalities were additionally required to complete the

reading section of the Test of English for International Communication (TOEIC) to

ensure that their proficiency was of a level that would enable them to benefit from

treatment instructions, explanations and feedback, all of which were provided in

English. This 100-question subtest of the TOEIC includes discrete-point grammar

items, error identification and reading comprehension tasks. Three participants were

removed from the study due to low test performance.

In addition, all participants completed a test of written Spanish in order to

control for L1 attrition. The test, which was developed by one of the researchers,

consisted of four gap-fill tasks adapted from textbook exercises designed for heritage

Spanish speakers (Foerster, Lambright, & Alfonso-Pinto, 1999). Two participants

were removed from the study on the basis of test performance which suggested

substantial language attrition over the course of their residence in the USA.

Using participants’ self-reports, the total number of instructional hours was

calculated for each participant’s study of English and of additional languages. Group

data for these variables as well as for age, language proficiency and formal education

variables are summarised in Table 1.

Target form

The linguistic target of the study was Latin morphosyntax related to the assignment

of thematic agent/patient roles to nouns. Latin was the language of choice because its

structure allows for the examination of three morphosyntactic cues associated with

Table 1. Descriptive statistics by group.

Group AoA AoTL1 prof.

(Max�7)L2 prof.

(Max�7)Education(Max�5)a

Instr.ESL

Instr.FLL

Early(n�15)

8.3(2.3)

25.1(8.1)

6.4(0.6)

6.8(0.4)

3.1(0.6)

288(760)

439(433)

Late(n�18)

25.1(6.8)

32.5(7.9)

7.0(0.1)

5.4(0.8)

3.3(1.1)

662(677)

169(242)

aEducational level was classified as follows: 1�high school diploma or GED; 2�technical school,associate’s degree or equivalent; 3�undergraduate degree or equivalent; 4�Master’s or law degree; and5�doctoral candidate.Note: Means are presented with SD in parentheses; AoA�age of arrival to USA; AoT�age at time ofexperiment; L1 prof.�self-assessed L1 Spanish proficiency; L2 prof.�self-assessed L2 English proficiency;Instr. ESL�hours of formal English instruction; Instr. FLL�hours of formal foreign language instruction.

International Journal of Multilingualism 167

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 9: An experimental study of early L3 development: age, bilingualism and classroom exposure

thematic role assignment: subject�verb agreement, subject�verb�object (SVO) word

order and noun case morphology. While the first two of these cues are analogous to

strong cues available in Spanish and English, the third represents not only a new cue

to be incorporated into participants’ existing hierarchies, but also is the only one of

the three that consistently maps onto thematic roles in Latin, making it an essential

cue for extracting and encoding meaning related to the semantic function of Latin

nouns.

Practice and critical test items, examples of which are presented in (1) below, eachconsisted of two human nouns5 and a transitive verb. Verb agreement and word order

were manipulated to enable comparison of participants’ cue use in the presence of

familiar (i.e. verb agreement and SVO word order) and unfamiliar (i.e. noun case

morphology) means of assigning thematic roles. In sentences in which one noun was

singular and the other plural, as in example (1b) below, verb agreement was

informative, whereas in sentences in which both nouns were singular or plural, as in

example (1c) below, it was not. Similarly, word order of items was varied so that it,

too, would be an inconsistently reliable cue for assigning thematic roles.

(1) a. Potentissimus auscultat reginam.

king-nom. sing. listen-3rd sing. queen-acc. sing.

The king listens to the queen.

b. Potentissimi reginam auscultant.

kings-nom. pl. queen-acc. sing. listen-3rd pl

The kings listen to the queen.

c. Reginas auscultant potentissimi.queens-acc. pl. listen-3rd pl. kings-nom. pl.

The kings listen to the queens.

Target sentences were generated from a list of 35 nouns and 11 verbs, all of which aremorphophonologically regular. In addition, each grammatical function (e.g. feminine

singular agent) is marked by unique inflection (e.g. -a), making noun case

morphology a highly reliable cue for assigning thematic roles correctly.6

Treatment

The study’s treatment was highly explicit in its instructional approach. It was selected

in order to draw participants’ attention as much as possible to the most efficient

means of assigning agent/patient roles in Latin, that is, relying on noun case

morphology cues. Participants worked through an interactive computer-delivered

treatment that began with a vocabulary lesson in Latin, followed by an explanation

of Latin sentence structure, and finally task-essential practice (Loschky &

Bley-Vroman, 1993) that incorporated explicit positive and negative feedback. Thesefeatures of the treatment, which are detailed below, were designed to increase the

salience of noun morphology by presenting it in a structured way and then

encouraging it to be processed as part of meaningful input practice.7 Given our

research questions, it was desirable to eliminate as much linguistic ‘noise’ as possible

in order to focus in on the morphosyntactic cues of interest and this is what

motivated our choice of computer-delivered instruction that provided all participants

with the same amount and type of L3 input. While such a learning environment is

168 C.A. Stafford et al.

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 10: An experimental study of early L3 development: age, bilingualism and classroom exposure

undeniably quite different from the way language learning happens in classroom and

naturalistic settings, what was lost in ecological validity was gained in the high degree

of experimental control our instructional treatment afforded us.

Vocabulary lesson and quiz

The vocabulary lesson introduced participants to the 46 lexical items used

throughout the grammar lesson and practice. Vocabulary learning was assessed by

means of a multiple choice quiz for which immediate onscreen feedback was

provided. Participants were required to pass the vocabulary quiz to a criterion of

100% and the computer application cycled back through the lesson until the criterion

score was met. This score was stipulated to ensure that errors made during practice

and testing were the result of difficulties with the target structures rather than with

lexical knowledge. Participants also completed brief multiple-choice format vocabu-

lary reviews before each of the language post-tests. Answers were provided for

incorrect responses and each vocabulary review was repeated until a criterion score of

100% was met.

Grammar explanation

The computerised grammar lesson was self-paced and provided explicit metalinguis-

tic information in English about how thematic agent and patient roles are assigned to

Latin nouns. First, the functional difference between nominative (referred to in the

explanation as ‘subject’) and accusative (referred to as ‘object’) case was explained

and participants interacted with several examples to reinforce the concepts presented.

Next, the Latin morphemes used to mark nominative and accusative case of

masculine and feminine, singular and plural nouns were presented individually,

followed by a series of interactive examples. In addition, Latin’s flexible word order

was highlighted and the difference between singular and plural verb forms was

pointed out. Total time spent on this portion of the treatment was approximately 20

minutes.

Practice and feedback

The input-based, task-essential practice session that followed the grammar explana-

tion consisted of three written and three aural interpretation tasks. Tasks were

presented in both modalities in order to appeal to different learning styles; all

participants completed the same six practice tasks to control the amount and type of

input exposure. Practice was ‘task-essential’ in the sense that successful task

completion required correct interpretation of the target structures. For two practice

tasks participants read or listened to a Latin sentence and interpreted it by selecting

one of two onscreen photographs as shown in Figure 1. For two other tasks,

participants read or listened to a Latin sentence and interpreted it by choosing its

English translation. For the third written interpretation task, participants read two

Latin sentences and decided which one accurately described an onscreen photograph.

For the final aural interpretation task, participants listened to a Latin sentence and

decided whether or not it accurately described an onscreen photograph.

International Journal of Multilingualism 169

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 11: An experimental study of early L3 development: age, bilingualism and classroom exposure

Sentences in all practice tasks consisted of two human nouns and a transitive

verb. Word order varied throughout the practice session and participants were

exposed to a total of 93 target sentences, 39 of which were presented in written form

and 54 of which were presented aurally.8

The explicit feedback provided by the computer programme was designed to be as

similar as possible to that which a conversation partner or instructor might offer in

that it attempted to focus participants’ attention on what they needed for successful

interpretation of the L3 input9; however, computer-delivered feedback is unique in

that it can control the amount and type of input that learners receive. Feedback

included item-specific metalinguistic information designed to reinforce rules of Latin

morphosyntax (see Figure 1) and was provided for both correct and incorrect

responses, again to ensure that all participants were exposed to the same amount and

type of input. The feedback remained onscreen for 5 seconds, after which the next

practice item was presented.10 Total time spent on the practice session was

approximately 1 hour.

Latin test battery

In order to provide as comprehensive an operationalisation of L3 learning as

possible, three different Latin language tests were included in a test battery, written

interpretation, aural interpretation and sentence production. The interpretation tests

were designed to be similar in format to the practice tasks that participants

completed during treatment (see Figure 1) so that we could observe how successfully

participants could apply in a parallel format what they had learned during treatment.

The interpretation tests differed from practice tasks, however, in the inclusion of a

third ‘I don’t know’ option to avoid the 50% chance of a dichotomous choice design.

Figure 1. Practice item with feedback.

170 C.A. Stafford et al.

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 12: An experimental study of early L3 development: age, bilingualism and classroom exposure

A production test was also included in the battery in order to assess to what

degree knowledge of Latin morphosyntax acquired through input-based practice

could be transferred to sentence production. As Wong (2004) suggested, if after

treatment that did not include any output practice participants could successfully

apply L3 knowledge for productive use, we would have evidence of non-negligible

interlanguage change. On the production test, participants constructed Latin

sentences to describe onscreen photographs by dragging and dropping noun and

verb stems and endings into an onscreen field (see Figure 2). Participants were free tobuild sentences as they wished; however, the programme did not allow noun

morphemes to be affixed to verb stems or vice versa.

Critical items on the language tests consisted of two animate nouns and a

transitive verb. Distractor items containing just one noun and a verb were included

on the production test, and three-word distractors which focused on lexis rather than

morphosyntax were included on the interpretation tests. Three versions of the three-

test battery were created and participants completed a different version as the pre-

test, post-test and delayed test. Order of completion of the written and auralinterpretation tests was randomised, but all participants completed the production

test last at all testing sessions. Of the 55 items included on the three tests of a given

battery, 34 were critical items and 21 were distractors. Total time for completion of a

testing session was approximately 30 minutes.

Procedure

A Web-based application combining Flash and ColdFusion programming delivered

treatment and tests and participants’ test answers were recorded to a database.

Participants completed the study in three sessions, meeting individually or in pairs

with one of the researchers. During the first session of the study, participantscompleted informed consent paperwork, a language background questionnaire, the

Figure 2. Written production test item.

International Journal of Multilingualism 171

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 13: An experimental study of early L3 development: age, bilingualism and classroom exposure

Spanish language test, Latin vocabulary lesson, vocabulary quiz and pre-tests.

Participants who scored above 70% on any of the three tests included in the pre-test

battery were eliminated from the sample. During the second session, approximately 1

week after the first, participants completed the lesson, practice and immediate post-

tests. During the final session, 3 weeks after the second session, participants

completed delayed post-tests and, if applicable, the reading section of the TOEIC.

Scoring

The written and aural interpretation tests each consisted of 20 items, 12 of which

were critical items. Participants received one point for each correct response to a

critical item for a maximum score of 12 on these two tests. The sentence production

test consisted of 15 items, 10 of which were critical items. For each critical productionitem, participants earned one point each for correctly marking the case of both

nouns, the number of both nouns, and verb agreement.11 Thus, each critical

production item had a total possible score of 3, making 30 the maximum score on

the production test.

Analysis and results

This study investigated initial L3 learning among Spanish�English bilinguals tocompare the influence of early- and late-acquired bilingualism on subsequent

language learning in adulthood. L3 learning was operationalised broadly as follows:

(1) overall accuracy on post- and delayed tests of Latin written interpretation, aural

interpretation and sentence production; (2) accuracy in interpretation of Latin

sentences containing different combinations of morphosyntactic cues; and (3)

patterns of use of the three cues to mark thematic roles in Latin sentence production.

These operationalisations represent the dependent variables considered in the

analyses and will be considered in turn in the following presentation of results.Except where noted, alpha was set at pB0.05, and values of partial Eta squared are

included to report effect sizes.

Overall L3 interpretation and production accuracy

As shown in Tables 2 and 3, participants’ test scores indicated a general trend of post-

treatment improvement in the ability to both interpret and produce Latin sentences,

with some regression to the mean at the delayed testing session. One-way ANOVAs

were performed on pre-test scores and results indicated no between-groups

differences prior to treatment on any measure of L3 performance with the exception

Table 2. Descriptive statistics for accuracy on L3 interpretation by group.

Group WI pre WI post WI del. AI pre AI post AI del.

Early (n�15) 6.1 (1.5) 9.9 (1.6) 8.5 (2.5) 5.6 (1.8) 9.7 (2.3) 7.9 (2.5)Late (n�18) 6.2 (1.6) 9.4 (2.3) 7.9 (1.8) 5.9 (1.8) 8.9 (1.9) 6.1 (2.2)

Note: Means are presented with SD in parentheses; WI�written interpretation; AI�aural interpretation;Max�12 on all tests. Scores on aural interpretation are based on n�10 Early Bilinguals and n�16 LateBilinguals due to technical problems with that test.

172 C.A. Stafford et al.

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 14: An experimental study of early L3 development: age, bilingualism and classroom exposure

Table 3. Descriptive statistics for groups’ overall accuracy and cue use in written production.

Group Overall pre Overall post Overall del. SVO pre SVO post SVO del. AGR pre AGR post AGR del. CASE pre CASE post CASE del.

Early(n�11)

12.6(5.7)

23.6(4.4)

21.0(5.0)

6.9(3.5)

6.6(2.9)

7.3(3.0)

6.9(1.6)

9.4(1.3)

9.5(0.69)

1.3(1.4)

7.6(2.1)

5.1(3.6)

Late(n�13)

12.3(3.9)

21.3(7.2)

19.6(4.0)

6.5(3.6)

5.3(2.4)

6.3(2.8)

5.0(2.0)

8.9(1.9)

9.0(1.5)

1.5(1.3)

5.8(3.8)

4.1(1.8)

Note: Means are presented with SD in parentheses; Max�30 for overall scores; SVO�use of SVO word order in production (Max�10); AGR�accurate marking of verbagreement in production (Max�10); CASE�accurate morphological marking of agent and patient roles in production (Max�10); production scores from four EarlyBilinguals and five Late Bilinguals are not included due to missing data resulting from technical problems with the computer application.

Intern

atio

na

lJo

urn

al

of

Mu

ltiling

ua

lism1

73

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 15: An experimental study of early L3 development: age, bilingualism and classroom exposure

of Early Bilinguals’ significantly more accurate use of verb agreement in sentence

production, F(1, 26)�10.05. Pre-test scores on this measure were thus entered as a

covariate in the corresponding analyses reported below.

Next, scores were submitted to a series of 3 (Time)�2 (Group) repeated-

measures ANOVAs. Results revealed significant main effects for Time with large

effect sizes on accuracy on all three measures, written interpretation, F(2, 58)�27.54,

partial h2�0.49, aural interpretation, F(2, 52)�30.66, partial h2�0.54, and

sentence production, F(2, 44)�29.90, partial h2�0.58. There was no statisticallysignificant main effect for Group and no significant Time�Group interaction.

Pairwise comparisons indicated significant pre-test to post-test improvement on all

three L3 measures, and pre-test to delayed test improvement in written interpretation

and sentence production but not in aural interpretation, suggesting that effects of the

instructional treatment were more lasting for written tests than for the aural test.

L3 interpretation accuracy by sentence type

The explicit instructional treatment was designed to move participants towards

restructuring their cue hierarchies for assigning semantic function to nouns such that

they would rely less on word order, a particularly strong L1/L2 cue, and rely more on

noun case morphology, the new and more reliable L3 cue. In order to see to what

degree participants altered their sentence processing strategies as a result ofinstructional treatment, we examined interpretation accuracy on sentences in which

different combinations of morphosyntactic cues were available to guide assignment

of thematic roles. Test items of the following types were considered (all of which

always included informative and reliable noun case morphology cues): (1) SVO items,

which were presented in subject�verb�object word order and included an informative

verb agreement cue (through contrastive noun number) about 50% of the time; (2)

AGR items, which were not presented in SVO order, but always included an

informative verb agreement cue; and (3) CASE items, which were not presented inSVO order and did not include informative verb agreement. In order maximise the

number of tokens of each sentence type, written and aural test items were combined

for the purposes of these analyses. Group accuracy data are summarised in Table 4.

ANOVA results revealed statistically significant main effects for Time with large

effect sizes on accurate interpretation of AGR items, F(2, 48)�47.42, partial h2�0.66, and CASE items, F(2, 48)�14.40, partial h2�0.38. There was no significant

main effect for Time on accurate interpretation of SVO items, no main effect for

Group and no significant Time�Group interaction. Pairwise comparisons indicatedsignificant pre-test to post-test improvement in interpreting both AGR and CASE

items, and pre-test to delayed test improvement in interpreting AGR, but not CASE,

items.

Cue preferences in sentence production

These analyses were conducted to examine how participants chose to mark

thematic roles on the production test. Production test answers were coded

according to which of the three possible cues for marking thematic roles they

used, and the number of times (Max�10) they used each cue at each test session

was tallied. Use of SVO word order across groups did not change substantially as a

174 C.A. Stafford et al.

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 16: An experimental study of early L3 development: age, bilingualism and classroom exposure

Table 4. Descriptive statistics for accuracy of L3 interpretation by sentence type and group.

Group SVO pre SVO post SVO del. AGR pre AGR post AGR del. CASE pre CASE post CASE del.

Early (n�10) 0.82 (0.29) 0.92 (0.14) 0.90 (0.16) 0.23 (0.18) 0.81 (0.22) 0.73 (0.30) 0.53 (0.16) 0.83 (0.20) 0.52 (0.24)Late (n�16) 0.82 (0.22) 0.90 (0.15) 0.87 (0.12) 0.27 (0.17) 0.73 (0.22) 0.55 (0.23) 0.53 (0.19) 0.71 (0.19) 0.51 (0.18)

Note: Means are expressed as proportions with SD presented in parentheses; SVO�items presented in subject�verb�object word order; AGR�items not presented in SVOword order, but with verb agreement and noun case morphology cues available; CASE�items with noun case morphology as only cue available to guide thematic roleassignment. Scores are based on N�26 due to technical problems with the aural interpretation test.

Intern

atio

na

lJo

urn

al

of

Mu

ltiling

ua

lism1

75

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 17: An experimental study of early L3 development: age, bilingualism and classroom exposure

result of treatment F(2, 44)�0.68, p�0.05, partial h2�0.03, but participants’

accurate use of both verb agreement and case morphology cues increased after

treatment (see Table 3).

Results revealed significant main effects for Time with large effect sizes on

accurate marking of verb agreement, F(2, 42)�35.61, partial h2�0.63, and case

morphology, F(2, 44)�25.96, partial h2�0.54. Pairwise comparisons indicated that

improvements were significant from pre-test to both post-test and delayed test for

accurate marking of both cues, indicating that the effects of the instructionaltreatment were lasting. Finally, results of independent-sample t-tests indicated that

the Early Bilinguals’ initial advantage in accuracy at marking verb agreement, shown

by a significant t-test result for pre-test scores, t(26)�3.17, disappeared after

instructional treatment, as indicated by a non-significant t-test result, t(26)�0.66,

p�0.05.

To summarise the results thus far, consistent findings of significant main effects

for Time suggest across-the-board improvements on all measures of initial L3

learning except interpretation of SVO sentences and use of SVO word order insentence production due to ceiling effects. Importantly, knowledge gained was

retained 3 weeks after treatment in all cases but overall aural interpretation accuracy

and interpretation of sentences in which the only cue available to assign thematic

roles was the new case morphology cue. Finally, the Early Bilinguals’ pre-treatment

advantage in accuracy of marking verb agreement in L3 production disappeared after

treatment.

The lack of significant differences between groups may have been the result of

factors that covary with the AoA variable we used to distinguish Early from LateBilinguals so we conducted further analyses to control several covariates. One-way

ANOVAs performed to examine between-group differences revealed no significant

difference in level of education, F(1, 31)�0.21, p�0.65, or amount of formal

instruction in English, F(1, 31)�2.23, p�0.15. Significant between-groups differ-

ences were revealed, however, for age at time (AoT), F(1, 31)�7.01, L1 proficiency,

F(1, 31)�13.25, L2 proficiency, F(1, 31)�37.70, and hours of classroom FLL, F(1,

31)�5.11. Early Bilinguals were on average younger, rated themselves as less

proficient in the L1 and more proficient in the L2, and had more experience withclassroom FLL than the Late Bilingual group. In light of these results, analyses of

covariance (ANCOVA) were performed in which AoT, L2 proficiency12 and

classroom FLL were controlled statistically. Results are presented for each of the

three dependent variables as outlined above.

Overall L3 interpretation and production accuracy

With the effects of AoT, L2 proficiency and classroom FLL controlled statistically,

results revealed a significant main effect for Time with a medium effect size onaccuracy in written interpretation, F(2, 52)�4.89, partial h2�0.16. Pairwise

comparisons showed that improvement in written interpretation accuracy was

significant from pre-test to both post-test and delayed test, suggesting that

participants retained what they learned as a result of interacting with the Latin

lesson and practice. There was no other significant main effect for either Time or

Group and no significant Time�Group interaction for overall accuracy on the three

L3 tests.

176 C.A. Stafford et al.

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 18: An experimental study of early L3 development: age, bilingualism and classroom exposure

L3 interpretation accuracy by sentence type

The ANCOVA results revealed significant main effects for Time with small effect

sizes on interpretation accuracy for SVO items, F(1.3, 27)�4.37, partial h2�0.17

and AGR items, F(2, 42)�3.26, partial h2�0.14. Pairwise comparisons indicated no

statistically significant difference before and after treatment in accuracy of

interpretation of SVO items; however, significant improvement was revealed in

accuracy of interpretation of AGR items from pre-test to both post-test and delayed

test. Although the main effect for Time on interpretation accuracy for CASE items

was not statistically significant, pairwise comparisons were examined and they

indicated significant improvement from pre-test to post-test in interpreting this

sentence type, with significant loss in accuracy from post-test to delayed test. No

significant main effect for Group or Time�Group interaction was found for

interpretation accuracy of any sentence type.

Cue preferences in sentence production

Results revealed a significant main effect for Time with a large effect size on accurate

marking of verb agreement, F(2, 38)�8.90, partial h2�0.32. Pairwise comparisons

showed that across-the-sample improvement in accurate marking of verb agreement

was significant from pre-test to both post-test and delayed test, indicating that the

effects of instructional treatment were lasting for this facet of L3 development.

A significant Time�Group interaction with a medium effect size was found for

accurate marking of noun case morphology, F(2, 38)�3.48, partial h2�0.16. To

examine this interaction more closely, post hoc paired-samples t-tests were performed

(Bonferroni-corrected significance levels were set at p�0.017 to adjust for multiple

comparisons). Results showed that while Late Bilinguals improved their accuracy in

marking noun case morphology significantly from pre-test to both post-test, t(12)��3.70, and delayed test, t(12)��4.18, Early Bilinguals improved significantly from

pre-test to post-test, t(10)��6.71, but only marginally so from pre-test to delayed

test, t(10)��2.79, p�0.019. This was due to a significant loss in accuracy from

post-test to delayed test, t(10)�2.87. These results suggest that the Late Bilinguals

were somewhat more successful than the Early Bilinguals at retaining what they

learned through interaction with the instructional treatment.

The ANCOVA results additionally revealed a significant Time�AoT interaction

with a medium effect size for accurate marking of noun case morphology, F(2, 38)�4.62, partial h2�0.20, suggesting that current age accounted for a non-trivial

proportion of variance in performance on this L3 measure. Correlations were run in

order to investigate this interaction further, and results revealed that while there was

no relationship between AoT and accuracy in marking noun case morphology at pre-

test or post-test, r(24)�0.11 and r(24)�0.10, respectively, there was a significant,

negative correlation at delayed test, r(24)��0.41.

To summarise, ANCOVA results were similar to those for ANOVAs in suggesting

that Early and Late Bilinguals garnered broadly comparable benefits from the

explicit instructional treatment. All participants continued to invoke SVO word

order � arguably the strongest cue to assigning semantic function among Spanish�English bilinguals � while simultaneously employing verb agreement and case

morphology cues more accurately after treatment in sentence interpretation and

International Journal of Multilingualism 177

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 19: An experimental study of early L3 development: age, bilingualism and classroom exposure

production. With ANCOVAs allowing for statistical control of factors (i.e. AoT,

English proficiency and classroom FLL) that covaried with AoA in our sample, Late

Bilinguals showed somewhat better retention over time of accuracy in marking the

new noun case morphology cue in sentence production. However, a significant

Time�AoT interaction revealed that current age had a role to play in retention over

time, with younger participants outperforming older participants at delayed test in

accurate marking of noun case morphology in production.

Discussion, conclusions and future research

Previous research has identified advantages for bilinguals over monolinguals as well

as for more balanced bilingualism in learning additional non-primary languages. Do

the cumulative effects of bilingualism influence early L3 development such that early-

onset bilinguals will show an advantage over later-onset bilinguals? And are there

other contributing factors that covary with age of onset of L2 learning? To answerthese questions, we examined the extent to which initial L3 learning among early and

late Spanish�English bilinguals was influenced by age of onset of L2 acquisition,

current age, L2 proficiency and exposure to formal language instruction.

Results revealed similar tendencies across the sample, suggesting no apparent

advantage at the initial stages of L3 learning for language experience accumulated

through bilingualism. Specifically, before participants interacted with the instruc-

tional treatment, when their L3 knowledge was limited to vocabulary, both Early and

Late Bilinguals relied per force on SVO word order � likely, the foremost cue in our

participants’ hierarchies given their L1/L2 background � to assign thematic roles in

L3 interpretation and production. The explicit instructional treatment helped both

groups move beyond word order-based strategies to use L3 verb agreement and noun

case morphology cues more successfully to assign thematic roles to Latin nouns.

Furthermore, despite the input-based nature of the practice, participants improved

significantly at not only interpreting but also producing L3 sentences. In other words,

participants were able to apply what they learned as a result of instructional

treatment to a context that was quite different from the input-based context in which

they had completed the lesson and practice.The pattern of results for interpretation by sentence type indicated that

participants continued to rely on SVO word order when it was available; however,

interpretation of sentence types in which SVO word order was not available improved

significantly as a result of treatment, and improvement was retained for a period of at

least 3 weeks in the case of successful use of the verb agreement cue. Results for the

production test patterned with those for interpretation in that after instructional

treatment participants continued to encode thematic roles with SVO word order

while at the same time including verb agreement and noun case morphology more

accurately and more frequently in production. Improved accuracy in use of both of

these cues in production was retained for a period of at least 3 weeks. We interpret

these results to mean that while participants continued to perceive word order as a

useful cue in the L3 context, they were nonetheless successful in creating form-

meaning conjunctions between the verb agreement and noun case morphology cues

and assignment of thematic roles in Latin.

At the early stage in learning that was the focus of the present study, there was

little difference in the L3 outcomes observed between early- and late-onset bilinguals.

178 C.A. Stafford et al.

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 20: An experimental study of early L3 development: age, bilingualism and classroom exposure

However, a significant difference did emerge in Early and Late Bilinguals’ ability to

retain what they learned about the new and complex noun case morphology cue. This

difference is notable because it suggests a slight advantage for Late Bilinguals over

Early Bilinguals, a result that stands in contrast with previous research that has

found advantages in non-primary language learning for more experienced language

learners and that led us to hypothesise that Early Bilinguals would have an advantage

over Late Bilinguals. Our results, however, disconfirm this hypothesis.

Why did Late Bilinguals outperform Early Bilinguals in this area of L3

development? Our answer is tentative given the small data set and rather isolated

finding upon which it is based, but it may be that the highly structured and explicit

instructional treatment was more in tune with Late Bilinguals’ learning strategies.

Apart from L1 acquisition, the Late Bilinguals in this study have always undertaken

language learning as cognitively mature adults. A number of researchers (e.g.

Cochran, McDonald, & Parault, 1999; Ellis, 2005; Kersten & Earles, 2001;

MacWhinney, 2001; Skehan, 1989) have argued that adult language learners in

particular benefit from the environmental support that characterises our treatment,

i.e. highly structured input, multi-modal presentation, task-essential practice andfocused feedback, so it may be that our Late Bilinguals responded better than Early

Bilinguals to the highly explicit instructional conditions.

Finally, the finding that AoT had a specific role to play in longer-term retention

of the new noun case morphology cue can be accounted for straightforwardly by

appealing to age effects on memory function. Given that participants had to

distinguish and encode eight discrete morphemes associated with noun case at

treatment and then retrieve them at the delayed test without the benefit of having had

any opportunity for L3 exposure or practice in the three intervening weeks, it is

unsurprising that older participants would be more subject to memory decay than

younger participants.

In a related vein, our results provide some evidence for the claim that the learning

of syntax on the one hand and the morphology with which that syntax is realised on

the other vary with regard to their relative complexity and do not necessarily proceed

at the same rate (e.g. DeKeyser, 2005; Sanz, 2000b). In the case of a new, complex cue

like noun case morphology, it is clear that more opportunity to engage in its

meaningful processing is essential for triggering learning mechanisms such aschunking and automatisation, which, as Ellis (2005) has suggested, are integral to

the strengthening of associations between form and meaning. Debriefing ques-

tionnaire data support this interpretation; several participants indicated that at the

delayed tests they remembered the form�meaning connection between case mor-

phology and thematic roles in Latin, but could no longer recall the specific function

marked by each of the eight morphemes.

In conclusion, we set out in this study to investigate the role of Spanish�English

bilinguals’ age of onset of L2 acquisition in initial L3 development in an explicit,

computer-assisted instructional context. The controlled laboratory setting enabled us

to focus on three morphosyntactic cues and compare whether and how early and late

bilinguals initiated the restructuring of cue hierarchies that defines acquisition within

the framework of the CM. Results point to more similarities than differences in

learning outcomes, which we attribute to the likelihood that the highly explicit nature

of the instructional condition served to level the playing field among our

participants. Sanz, Lin, Lado, Bowden, and Stafford (2009) similarly found that

International Journal of Multilingualism 179

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 21: An experimental study of early L3 development: age, bilingualism and classroom exposure

thinking aloud during interaction with an explicit lesson comparable to ours had no

effect on adult English�Spanish bilinguals’ L3 learning outcomes while, in contrast,

thinking aloud under a less explicit instructional condition enhanced learning

outcomes.

Notwithstanding the broad similarities observed in our participants’ learning

outcomes, the Late Bilinguals showed a slight advantage over Early Bilinguals at

retaining what they learned about case morphology, with current age also playing a

role in retention of this new and complex L3 cue. We suggest that explicit instructionmay resonate especially well among later bilinguals, whose learning strategies may

make its high degree of explicitness particularly beneficial, and that even in highly

explicit and supportive instructional contexts, age effects on memory function

inevitably play a role in how well learners retain what they learn over time.

Future studies should continue to tackle the challenge of bilinguals’ inherent

heterogeneity by considering experiential, social and cognitive factors such as

individuals’ motivations for becoming bilingual, identity within the L1/L2 commu-

nity, and aptitude for language learning, and should start with sizable samples. In ourcase, homogeneity and control through careful screening for L1 attrition, L2

proficiency and level of formal education came at a cost to statistical power. It will

also be important for future research to examine the interaction between pedagogical

conditions and the numerous internal variables that covary with bilingualism as well

as how such interactions affect the development of different aspects of non-primary

language in adulthood.

Acknowledgements

The authors thank the friends, family and colleagues who made the Latin Project, of which thisstudy is a part, come to fruition. They are considerably indebted to Bill Garr, Ru San Chen andGorky Cruz for their programming and statisticial expertise. The Latin Project was developedby the aurhors with support from Georgetown University Graduate School grants to CristinaSanz. The authors thank Diana Frantzen, Sally Magnan and three anonymous InternationalJournal of Multilingualism reviewers for their helpful comments on earlier versions of thispaper. Any remaining errors are our own.

Notes

1. The term ‘bilingual’ is used inclusively in this study to refer to individuals who use bothSpanish and English in some capacity on a daily basis while allowing for a range ofproficiencies in both languages.

2. The author reported that, though Spanish was the L1 for most of the students in hissample, Basque was the L1 for some. He did not provide further details regardingdistribution of the L1 Basque speakers among groups.

3. The languages reported (with the numbers in parentheses representing the number ofparticipants who reported some level of exposure to or knowledge of the precedinglanguage) were French (16), Portuguese (8), Italian (7), Japanese (2), Danish (1), Kekchi-Kakchikel (1), Nahuatl (1), Swedish (1) and Yucatec-Maya (1). Exposure to Romancelanguages had occurred in all cases in the context of formal instruction, and participantsreported at best limited productive knowledge of these languages. Exposure to non-Romance languages (all of those listed here include case inflection in their grammarsexcept for Swedish) occurred in informal contexts (as with domestic help in the home orthrough vacations or military service abroad), and participants reported having neitherreceptive nor productive knowledge of any of these languages, so their exposure wasconsidered unproblematic for the study. Of those who reported such exposure to case

180 C.A. Stafford et al.

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 22: An experimental study of early L3 development: age, bilingualism and classroom exposure

inflection languages, two were Early Bilinguals and four were Late Bilinguals (one of theLate Bilinguals had been exposed to both Nahuatl and Japanese).

4. Proficiency ratings included on the scale were as follows: 1�none or almost none; 2�poor; 3�fair; 4�functional; 5�good; 6�very good; 7�like a native speaker.

5. All nouns included in the stimuli are human in order to avoid animacy as a potentiallyinformative clue to assigning thematic roles.

6. A masculine singular agent is marked by the morpheme �us in Latin. Given that a word-final �s marks plural rather than singular nouns in Spanish and English, this potentiallyconfusing difference was addressed explicitly during the grammar explanation.

7. The design of the treatment is similar to VanPatten’s (2005) processing instruction, apsycholinguistically motivated instructional technique intended to draw learners’attention to less salient, more complex features of the L2 input so that they are noticedand processed more elaborately.

8. Each sentence in the aural practice exercises was repeated once; thus, the total of 54sentences presented aurally represents 27 different sentences repeated once each.

9. For example, if a participant gave an incorrect answer for the Latin object�verb�subjectsentence ‘Stultum salutat potentissimus’ (The king greets the fool), s/he received thefollowing feedback: Uh-oh! �us is a subject ending. �um is an object ending. Also, rememberthat in Latin, the subject does not have to be the first noun. In this way, the most reliableway of assigning agent and patient roles was reinforced while at the same time pointingout that using SVO word order in Latin can be a misleading processing strategy. Weacknowledge that participants may have made an incorrect answer choice for a reasonthat the computerised feedback did not address, but feedback was designed to addresshighly likely reasons for incorrect responses.

10. Two participants reported that 5 seconds had not been enough time to read and absorbthe feedback before the programme advanced to the next item. Thus, reading speed andskill is likely to have influenced the degree to which participants were able to take fulladvantage of the feedback provided.

11. For example, the sentence ‘Potentissimi reginam auscultant’ (the kings listen to the queen)would score 3 points for correctly marking the number of both nouns, the case of bothnouns and subject�verb agreement; the sentence ‘Potentissimos reginam auscultant’would score 2 points for correctly marking both nouns’ number and subject�verbagreement (the point is not scored for noun case in this example because both nouns aremarked for accusative case); the sentence ‘Potentissimos reginam auscultat’ would score 1point for correctly marking both nouns’ number (points are not scored for noun case orsubject�verb agreement in this case). We decided to award 1 point for correct case andnumber marking of both nouns rather 1 point for each correct morpheme becausemarking both nouns correctly was the most unambiguous demonstration that partici-pants were using noun morphemes appropriately to contrast thematic roles.

12. L2 proficiency was selected over L1 proficiency as a control variable because there was abroader range in the L2 proficiency self-ratings, suggesting superior discriminationamong participants than with the L1 proficiency ratings, in which there was less variationacross the sample.

References

Bates, E., & MacWhinney, B. (1982). Functionalist approaches to grammar. In E. Wanner &L.R. Gleitman (Eds.), Language acquisition: The state of the art (pp. 173�218). New York:Cambridge University Press.

Bates, E., & MacWhinney, B. (1989). Functionalism and the competition model. InB. MacWhinney & E. Bates (Eds.), The cross-linguistic study of sentence processing (pp.3�73). New York: Cambridge University Press.

Bialystok, E. (1988). Levels of bilingualism and levels of linguistic awareness. DevelopmentalPsychology, 24, 560�567.

Bialystok, E. (2001). Bilingualism in development: Language, literacy, and cognition. New York:Cambridge University Press.

International Journal of Multilingualism 181

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 23: An experimental study of early L3 development: age, bilingualism and classroom exposure

Bialystok, E. (2007). Cognitive effects of bilingualism: How linguistic experience leads tocognitive change. The International Journal of Bilingual Education and Bilingualism, 10,210�223.

Bialystok, E., Craik, F.I.M., Klein, R., & Viswanathan, M. (2004). Bilingualism, aging, andcognitive control: Evidence from the Simon task. Psychology and Aging, 19, 290�303.

Bialystok, E., Craik, F.I.M., & Ruocco, A.C. (2006). Dual-modality monitoring in aclassification task: The effects of bilingualism and ageing. The Quarterly Journal ofExperimental Psychology, 59, 1968�1983.

Bialystok, E., & Martin, M.M. (2004). Attention and inhibition in bilingual children: Evidencefrom the dimensional change card sort task. Developmental Science, 7, 325�339.

Birdsong, D., & Molis, M. (2001). On the evidence for maturational constraints in second-language acquisition. Journal of Memory and Language, 44, 235�249.

Cenoz, J. (2003). The additive effect of bilingualism on third language acquisition: A review.The International Journal of Bilingualism, 7, 71�87.

Cenoz, J., & Valencia, J.F. (1994). Additive trilingualism: Evidence from the Basque country.Applied Psycholinguistics, 15, 195�207.

Cochran, B.P., McDonald, J.L., & Parault, S.J. (1999). Too smart for their own good: Thedisadvantage of a superior processing capacity for adult language learners. Journal ofMemory and Language, 41, 30�58.

DeKeyser, R.M. (2000). The robustness of critical period effects in second languageacquisition. Studies in Second Language Acquisition, 22, 499�533.

DeKeyser, R.M. (2005). What makes learning L2 grammar difficult? A review of issues.Language Learning, 55, 1�25.

Ellis, N.C. (2005). At the interface: Dynamic interactions of explicit and implicit languageknowledge. Studies in Second Language Acquisition, 27, 305�352.

Foerster, S.W., Lambright, A., & Alfonso-Pinto, F. (1999). Punto y Aparte: Spanish in review,moving toward fluency. New York: McGraw-Hill College.

Gibson, M., Hufeisen, B., & Libben, G. (2001). Learners of German as an L3 and theirproduction of German prepositional verbs. In J. Cenoz, B. Hufeisen, & U. Jessner (Eds.),Cross-linguistic influence in third language acquisition: Psycholinguistic perspectives (pp.138�169). Buffalo, NY: Multilingual Matters.

Hernandez, A.E., Sierra, I., & Bates, E.A. (2000). Sentence interpretation in bilingual andmonolingual Spanish speakers: Grammatical processing in a monolingual mode. SpanishApplied Linguistics, 4(2), 179�213.

Johnson, J.S., & Newport, E.L. (1989). Critical period effects in second language learning: Theinfluence of maturational state on the acquisition of English as a second language.Cognitive Psychology, 21, 60�99.

Kempe, V., & MacWhinney, B. (1998). The acquisition of case marking by adult learners ofRussian and German. Studies in second language acquisition, 20, 543�587.

Kersten, A.W., & Earles, J.L. (2001). Less really is more for adults learning a miniatureartificial language. Journal of Memory and Language, 44, 250�273.

Klein, E.C. (1995). Second versus third language acquisition: Is there a difference? LanguageLearning, 45, 419�465.

Lasagabaster, D. (2000). Three languages and three linguistic models in the Basque educationalsystem. In U. Jessner & J. Cenoz (Eds.), English in Europe: The acquisition of a thirdlanguage (pp. 179�197). Buffalo, NY: Multilingual Matters.

Loschky, L., & Bley-Vroman, R. (1993). Grammar and task-based learning. In G. Crookes &S. Gass (Eds.), Tasks and language learning: Integrating theory and practice (pp. 123�167).Buffalo, NY: Multilingual Matters.

MacWhinney, B. (2001). The competition model: The input, the context, and the brain. InP. Robinson (Ed.), Cognition and second language instruction (pp. 69�90). New York:Cambridge University Press.

MacWhinney, B. (2005). New directions in the competition model. In M. Tomasello & D.I.Slobin (Eds.), Beyond nature-nurture: Essays in honor of Elizabeth Bates (pp. 81�110).Mahwah, NJ: Lawrence Erlbaum.

182 C.A. Stafford et al.

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014

Page 24: An experimental study of early L3 development: age, bilingualism and classroom exposure

Martin-Rhee, M.M., & Bialystok, E. (2008). The development of two types of inhibitorycontrol in monolingual and bilingual children. Bilingualism: Language and Cognition, 11,81�93.

Munoz, C. (2000). Bilingualism and trilingualism in school students in Catalonia. InU. Jessner & J. Cenoz (Eds.), English in Europe: The acquisition of a third language(pp. 157�178). Buffalo, NY: Multilingual Matters.

Park, D.C. (2000). The basic mechanisms accounting for age-related decline in cognitivefunction. In D.C. Park & N. Schwarz (Eds.), Cognitive aging: A primer (pp. 3�21).Philadelphia, PA: Psychology Press.

Sanz, C. (2000a). Bilingual education enhances third language acquisition: Evidence fromCatalonia. Applied Psycholinguistics, 21, 23�44.

Sanz, C. (2000b). What form to focus on? Linguistics, language awareness, and the educationof L2 teachers. In J.F. Lee & A. Valdman (Eds.), Meaning and form: Multiple perspectives.Boston: Thomson Learning.

Sanz, C. (2007). Predicting enhanced L3 learning in bilingual contexts: The role of biliteracy.In C. Perez-Vidal, M. Juan-Garau, & A. Bel (Eds.), A portrait of the young in the newmultilingual Spain (pp. 220�240). Buffalo, NY: Multilingual Matters.

Sanz, C., & Lado, B. (2008). Third language acquisition research methods. In K. King & N.H.Hornberger (Eds.), Encyclopedia of language and education (Vol. 10, pp. 1�23). Boston:Kluwer.

Sanz, C., Lin, H., Lado, B., Bowden, H.W., & Stafford, C. (2009). Concurrent verbalizations,pedagogical conditions and reactivity: Two CALL studies. Language Learning, 59, 33�71.

Skehan, P. (1989). Individual differences in second-language learning. New York: EdwardArnold.

Su, I.R. (2001). Transfer of sentence processing strategies: A comparison of L2 learners ofChinese and English. Applied Psycholinguistics, 22, 83�112.

VanPatten, B. (2005). Processing instruction. In C. Sanz (Ed.), Mind and context in adultsecond language acquisition: Methods, theory, and practice (pp. 267�281). Washington, DC:Georgetown University Press.

Wong, W. (2004). The nature of processing instruction. In B. VanPatten (Ed.), Processinginstruction: Theory, research, and commentary (pp. 33�63). Mahwah, NJ: LawrenceErlbaum.

International Journal of Multilingualism 183

Dow

nloa

ded

by [

UQ

Lib

rary

] at

03:

56 2

0 N

ovem

ber

2014