Professional Documents
Culture Documents
Learning and Individual Differences
Learning and Individual Differences
a r t i c l e i n f o a b s t r a c t
Article history: We examined the association between proficiency in instructed second language acquisition (ISLA) and previous
Received 6 May 2014 bilingualism, starting age of ISLA, language anxiety and attitude. The analyses were conducted on 564 adolescent
Received in revised form 4 April 2015 Australian twins. Additionally, by examining discrepancies within approximately 100 pairs of monozygotic
Accepted 15 August 2015
twins, we sought to specifically identify the environmental effects related to attitude and anxiety on achievement
Available online xxxx
(i.e. with genetic effects removed). We found a clear relationship between attitude towards language learning
Keywords:
and proficiency in the second language. Furthermore, the analyses on the monozygotic twins point to the possi-
Second language acquisition bility that higher language anxiety is associated with higher proficiency. On the other hand, bilingualism and
Attitude starting age of ISLA appear to be unrelated to proficiency in the language being learned.
Bilingualism © 2015 Elsevier Inc. All rights reserved.
Starting age
Anxiety
http://dx.doi.org/10.1016/j.lindif.2015.08.012
1041-6080/© 2015 Elsevier Inc. All rights reserved.
I. Antón-Méndez et al. / Learning and Individual Differences 42 (2015) 44–52 45
(“fraternal”) twins, can be turned to identify environmental factors. In Below, we briefly review the literature pertaining to each of the four
particular, differences between monozygotic twins, as used here, can be key issues in ISLA research that this paper focuses on: the effect of age
especially revealing. on rate of learning, the effect of prior bilingualism, the effect of anxiety
In the paper mentioned earlier (Coventry et al., 2012), we reported and the effect of attitude.
findings from behaviour-genetic analyses on the language learning
achievements of monzygotic (MZ) and dizygotic (DZ) twins. The 1.1. Age
paper also outlines how twins are used to estimate the heritability of ac-
ademic achievement and the relative contributions of environmental Age is perhaps one of the most widely-researched aspects of Second
factors, and the reader is referred to that paper for in-depth explana- Language Acquisition. Since Johnson and Newport's (1989) seminal
tions. Our results showed that, at least as demonstrated by teacher rat- paper extending the hypothesis of a critical period of language acquisi-
ings, success at learning a second language in school is heritable, with tion (Lenneberg, 1967) to the learning of a second language, much work
about two-thirds of the variability among students attributable to has been done investigating the existence or otherwise of a critical or a
genes. Shared environment, that is, environmental factors that twins sensitive period. This refers to the notion that there is a maturational
within a family are likely to share, accounted for about one fifth of var- threshold over which learners are markedly less likely to fully acquire
iability, while non-shared environment, that is, environmental factors certain aspects of a new language, most notably pronunciation, but
unique to individual children within the same family and which in- also morphology and syntax when there are substantial differences be-
cludes measurement error, accounted for about one-quarter. These tween the learner's first and second languages. After a certain age, thus,
non-shared environment effects are simply the differences between learning another language is believed to be, if nothing else, much more
identical twins, which we explore herein. effortful as well as more likely to result in limited success (DeKeyser,
In this paper we adopt a mixed approach to focus on the environ- 2000; DeKeyser & Larson-Hall, 2005; Paradis, 2004; Ullman, 2001).
mental influences of several key factors on instructed second language However, the existence of gradual declines in ultimate attainment
acquisition (ISLA) among adolescent Australian twins where both across the life span rather than a sharp drop after a certain age has led
twins are studying a second language at school. On the one hand, we some authors to question the Critical Period Hypothesis, explaining
ignore the fact that the respondents are twins, referred to hereafter as differences between early and late learners instead as a function of op-
non-twin analyses, to explore associations between different factors. portunities to use the L2 (Birdsong, 2006; Birdsong & Molis, 2001;
On the other hand, we also present one of the first applications of Singleton, 2005). Additionally, the existence of adult learners who
twin design to further explore the specific influence of contextual achieve native-like proficiency (Ioup, Boustagui, El Tigi, & Moselle,
factors in second language achievement in the classroom by examining 1994) and evidence of equivalent neurological markers underlying
discrepancies between MZ twins for clues as to the types of environ- processing of linguistic aspects for first and second language speakers
mental effects that predict achievement. MZ twins share 100% of their (Friederici, Steinhauer, & Pfeifer, 2002) also undermines the whole con-
genes as well as sources of shared environmental influence on a trait cept of a critical period. The lack of consensus extends beyond accepting
(e.g., family attitudes and backgrounds, twins' shared friends), which whether there is a specific critical age to also query what that age may
means that any differences between the two twins in a pair, if not due be, if there is one. In fact, there is evidence that different linguistic as-
to measurement error, should be due to differential non-genetic influ- pects may be associated with different critical ages (Flege, Munro, &
ences, that is, unique environment. Examples of this would be one MacKay, 1995; Hyltenstam & Abrahamsson, 2003; Kuhl, Tsao, & Liu,
twin but not the other suffering an illness, having separate peers, 2003; Sebastián-Gallés, Echevarría, & Bosch, 2005). For example, Kuhl
being instructed by separate teachers, and so on. Thus, examining et al. (2003) found very early effects, around the age of 1 year old, in
which variables correlate with or characterize difference scores be- the case of phonological acquisition. However, puberty is taken by
tween MZ twins for a given trait could give an indication of which many to be a useful cut-off point for most linguistic aspects, including
environmental factors influence that particular trait. syntax and morphology. This is because puberty is a time when the
The question of which factors impact L2 acquisition – with a view to brain undergoes dramatic changes which are likely to have cognitive con-
both explaining differences and improving outcomes in second and sequences, such as a decrease in the ability to adapt to specific features of
foreign language learning – is one that has received a great deal of atten- a new language (Pulvermüller & Schumann, 1994; Ullman, 2004). In this
tion in the instructed second language acquisition and pedagogical liter- (our) contribution to the issue, we followed this general trend in choos-
ature. Some inquiry methods focus on the instructional process itself: ing puberty as marking the end of the critical period and adopted the
the teaching method, the role of the teacher and the merits of form- age of 12, as a conservative estimate of puberty onset, to separate early
focused versus more naturalistic pedagogy (Norris & Ortega, 2000). In starters from late starters (ages between 12 and 14 are common “late-
this study, however, we focused on the learner and how s/he interacts starter” cut-offs in the literature, e.g., Antón-Méndez, 2010, 2012;
with opportunities for learning. Influences on the learner that have Foucart & Frenck-Mestre, 2012; McDonald, 2006).
been studied include: the optimal starting age for second language
learning (Birdsong, 2006); the role of motivation in achieving proficien- 1.2. Bilingualism
cy (Dörnyei, 1990; Dörnyei & Ushioda, 2009); the effects of various in-
dividual differences such as beliefs, affective factors, cognitive styles Since 40 children in the database came from families where there
and learning strategies on outcomes (Dewaele & Furnham, 2000; were indications of family bilingualism, we were able to investigate
Dörnyei, 2005; Verhoeven & Vermeer, 2002); and whether prior bilin- whether prior bilingualism had any effect on success in ISLA. While
gualism assists the learning of a third or subsequent language (Cenoz, some scholars take the position that the underlying processes are the
2013). Our design allowed us to examine most of these. As mentioned same for learning second or subsequent languages (Mitchell & Myles,
above, we analysed the possible influences of previous bilingualism, 2004; Sharwood-Smith, 1994), in recent years it has been widely recog-
starting age of L2 acquisition, language anxiety, and learner motivation. nized that third and subsequent language acquisition (TLA) is qualita-
These factors were selected due to, on the one hand, their relevance in tively different from SLA since the learner has two language systems
SLA research and, on the other, the feasibility of extracting reliable to use as a basis for acquisition and, furthermore, prior learning experi-
data from the sample under study. Although this list of possible ence is likely to play a role in the new learning task (De Angelis, 2007).
influencing factors is not exhaustive, it is worth noting that we present The Dynamic Model of Multilingualism proposed by Herdina and
here all the factors we did measure, therefore following the scientifically Jessner (2002) suggests that not only does L1 influence L2, as is well-
prudent course of reporting all associations measured even if they are accepted in SLA, but that L2 influences L1, and the combined metacog-
not significant. nition and increased language awareness accruing from knowledge of
46 I. Antón-Méndez et al. / Learning and Individual Differences 42 (2015) 44–52
two languages are brought to bear on the acquisition of a third, resulting Table 1
in a recombinant and dynamic series of interactions. We were therefore Maximum sample sizes, average ages (and S.D.), and percentages of females (F) and
monozygotic (MZ) twins for each of the proficiency measures.
interested to know whether prior bilingualism had an effect on success
in learning an additional language in the school context. Proficiency measure N Age (years.) %F % MZ
Vanderplank, & Strube, 2012; Ross, 1998). As a result, self-ratings are Table 2
commonly used in research on bilinguals and language learners Results of the Principal components analysis for questionnaire items related to attitude
and language learning anxiety.
(e.g., Costa & Santesteban, 2004; Rossi, Gugler, Friederici, & Hahne,
2006; Segalowitz & Frenkiel-Fishman, 2005). Questionnaire item Component
The internal reliabilities were high for all three measures (Cronbach's 1 2
alphas were .89, .95, and .93 for twin ratings, teacher ratings and teacher
a Likes learning languages .873 −.101
rankings respectively). b Likes class .807 −.187
The parents' and twins' questionnaires contained standard demo- c Finds language learning important .797 .047
graphic questions, questions assessing zygosity, and questions regarding d Works hard in class .759 −.105
the following variables4: bilingualism – whether there were any e Likes teacher .681 −.057
f Practices as often as possible .635 .055
languages that the twins either spoke or were regularly exposed to at g Admires speakers of second languages .592 .182
home (if the school language was the same as the home language, the h Associates language learning with good .584 .143
student's data were excluded from the analyses; see Section 2.3 below job
for more specific information on this measure); starting age of SLA – i Tries to speak in new language .407 .231
j Nervous about speaking −.119 .874
age at which student had started learning the second language; SLA
k Worries about mistakes .180 .773
anxiety – student's level of anxiety with respect to practicing the second
language (see Section 2.3 and Table 2 below for more specific informa-
tion on this measure); and SLA attitude – student's attitude towards as the residuals after using regression to adjust for the covariates
second language learning in general, and towards the student's particu- mentioned.
lar language class (see Section 2.3 and Table 2 below for more specific
information on this measure). 2.3.2. Bilingualism
To examine whether early home bilingualism confers any advan-
2.3. Data preparation tages at ISLA, we first had to determine which students could be consid-
ered bilingual on the basis of their responses to several items in the
Some of the variables of interest could not be derived from responses questionnaires. We did not include a straight question about whether
to single questionnaire items and thus required preparation before they the students were bilinguals as this term has different meanings for
could be entered in statistical analyses. In this section, we detail which different people. As indicative of bilingualism, we selected a number
data underwent further processing and how they were prepared for of items expected to correlate with degree of family (or individual) bi-
the analyses that follow. lingualism: student reports that their home language was not English,
student reports of other language(s) spoken at home, student reports
2.3.1. Proficiency of being able to speak and understand other languages (apart from
As reported in our previous paper (Coventry et al., 2012), we looked at the one being taught at school), parent reports of speaking another lan-
whether the ratings for the four key learning areas of speaking, listening, guage at (near) native level, parent reports of other language(s) spoken
reading and writing represented a single factor and could therefore be early and regularly to the twins, and student reports of having spent a
combined into one single measure. For this, we carried out a principal substantial amount of time (three months or longer) in a country
components factor analysis of the speaking, listening, reading and where English was not the official language.
writing items for the teacher NC ratings, teacher rankings and twin self- Responses to all of these items except the last one were on a 4 point
ratings. If a single factor is found with this procedure to provide a parsi- Likert scale (“always”, “often”, “sometimes”, “rarely”). Students who
monious account of the data, the use of a single proficiency measure in- had answered “rarely” to all of these and had not lived in another
stead of four is warranted. We found that a single factor accounted for country were considered monolingual (N = 524). There were 40
74%, 88% and 90% of the variance for twin self-ratings, teacher NC ratings twins who had responded positively to different subsets of these
and teacher rankings respectively, suggesting that, for each of these pro- items. These were inspected individually and assigned to one of three
ficiency measures, we could average across the speaking, listening, read- categories according to the probability that the student was an early bi-
ing and writing items to form a single composite measure for each. lingual based on the combined responses to all these items. Students
In the case of students for whom we had proficiency data about were considered bilingual (N = 36) if: a) they reported that their
more than one second language, we chose to carry out the analyses on home language was not English, b) they reported that another language,
the language for which they had the highest (i.e. best) ratings (either in which they were proficient, was spoken at home a substantial
self-ratings or NC ratings), provided this was not a language already amount of time, and this was supported by parent reports of their
spoken at home. We opted for this “best language” because the learners own bilingualism and use of other languages at home, c) parents report-
would have had the most experience with those languages, which ed a history of other languages spoken at home at an earlier age and
enhances the reliability of the assessments (Ross, 1998). students themselves reported being somewhat proficient in that
For these analyses we conducted a square-root transformation on language even if they were no longer using it regularly. There were
the teacher ranking variable in order to approximate a normal distribu- also some students (N = 4) who had merely been exposed to another
tion. The three proficiency variables were then standardized. We also language early on but appeared to have acquired no active knowledge
covaried out the effects of (1) school year, (2) years spent studying an of it. This last group was excluded from analyses of possible effects of
additional language, (3) age and (4) sex, since such effects could have bilingualism.
introduced systematic biases on the proficiency estimates. For example,
the number of years studying a language should affect the NC ratings 2.3.3. Anxiety and attitude
(which should be higher the more years students spent studying a Students' views and feelings about language learning and affinity for
language), while it would be less likely to affect twins' own estimate their language classes were probed by means of a number of question-
(which should be more dependent on comparisons with classmates naire items asking them: if they thought language learning was impor-
than on attainment of specific milestones, as the NC ratings are). The tant, if they liked it, if they admired people who could speak several
covariation procedure recalculates the different proficiency measures languages, if they wanted to learn another language in order to get a
good job, if they took every opportunity to practise outside of class, if
4
Questionnaires also contained other items not relevant to the present paper. Full ques- they tried to say what they wanted even when they didn't know the
tionnaires are available from the authors on request. exact words, if speaking another language outside of class made them
48 I. Antón-Méndez et al. / Learning and Individual Differences 42 (2015) 44–52
nervous, if they worried about making mistakes, if they liked their uncorrelated random effects (Visscher, Benyamin, & White, 2004): the
language class, if they liked their teacher, and if they worked hard for first specified the twins as pairs (i.e., each pair of twins had their own
this subject. All these measures ranged from 0, not at all true, to 3, id number), and the second specified the effect shared by MZ twins
very much true – therefore, higher scores signified students had higher but not by DZ twins (i.e., each MZ pair had their own id number and
levels of anxiety or a better attitude towards language learning. To all DZs were treated as individuals who were unpaired, each having
assess how many psychological traits actually underlay the responses their own id number).
to the different questions, responses to these 11 items were submitted For the MZ twin analyses, we selected all MZ twin pairs where both
to a principal components factor analysis with Oblimin rotation. Kaiser's twins were studying the same language (provided they did not speak
rule suggested three factors, which explained 63.46% of the variance, the same language at home as they were learning at school), and creat-
but the scree plot suggested one factor. Accordingly, we ran a one, two ed difference scores for the three proficiency measures by subtracting
and three factor solution and, of these, chose the two factors solution the adjusted and standardized scores of the two twins (i.e. twin 1
as this gave the simplest structure (i.e. the fewest cross loadings). This minus twin 2). We also created difference scores for: age at which the
solution exhibited a very clean structure with pattern-matrix loadings two twins had started to learn a second language; performance anxiety;
greater than .50 and no cross-loadings above .18 for all items except and attitude towards language learning. The fourth factor, bilingualism,
whether students tried to say what they wanted in the second language could not be considered since there were no twin pairs for which one
even when they did not know all the words (see Table 2). This item, twin was monolingual while the other was bilingual. We also found
item i in Table 2, was removed. The loadings indicated that the first that, in most of the twin pairs, both twins had started to learn the lan-
component was what could be broadly considered attitude towards lan- guage at the same age (80%, 78%, and 82% of the pairs for self-ratings,
guage learning and the language class (with high loadings, N .58, for NC ratings, and teacher rankings respectively), so the distribution of dif-
questions about whether students liked learning languages, liked their ference scores for starting age did not deliver enough variance to pursue
language class, thought language learning was important, worked the analyses. As a consequence, for the MZ twin analyses, the multiple
hard for this subject, liked their teacher, took every opportunity to prac- regression analyses contained only two predictors: difference scores
tise outside of class, admired people who could speak several languages, for attitude and anxiety.
and thought knowing a second language could lead to a good job), and By choosing only MZ twins, sample sizes are reduced with respect to
the second was related to anxiety (with high loadings (N .77) for items those included in the non-twin analyses. However, the resulting
on whether students felt nervous about speaking another language out- samples (see Section 3.2 below) are not considered small for this kind
side of class, and whether they worried about making mistakes). The of analysis since discordant MZ analyses have substantially more
two factors proved reliable with Cronbach's alphas of .86 and .65 for power than regular analyses (Christian & Kang, 1972).
the attitude and anxiety factors respectively. We therefore constructed
two simple measures –attitude and anxiety – by averaging the items 3. Results
found to relate to each of the two components of the principal compo-
nents analysis (PCA). The correlation between these two new constructs 3.1. Non-twin analyses
of anxiety and attitude was .264 (p b .001).
Descriptive statistics regarding the variables included in the MRs are
2.4. Analyses presented in Table 3. Note that the sample, though mostly overlapping
for the three dependent variables of proficiency, was different for each
Data were subject to two different sets of analyses. In the first set, the since data for one or the other proficiency measure could be missing
non-twin analyses, all the students were included, and each twin was for some individuals. In particular, many teachers chose not to include
treated as an individual subject (so each family contributes two scores, the student's ranking within the language class.
one for each twin). In the second set, the MZ twin analyses, only the MZ A summary of the results for the three multiple regression analyses
twin pairs were included, and each pair contributed just a single differ- is presented in Table 4. Altogether the predictors explained between
ence score between the two twins in a pair (so each family contributes a 17 and 43% of the variance, explaining the most amount of variance
single score). for the twins' self-ratings of proficiency. The coefficients revealed a
For the non-twin analyses, we ran standard multiple regression very consistent pattern across all three criterion variables, with only
(MR) analyses to assess the contribution on each of the three different attitude towards the language class explaining significant amounts of
proficiency measures as criteria (twin self-ratings, teacher NC ratings, variance in proficiency: the better the attitude, the higher the proficien-
and teacher rankings; see Section 2.3.1 for descriptions) of the four cy. The effect size for this predictor is medium to large in the case of
factors of interest as predictors: 1) bilingualism, 2) starting age of SLA, teacher rankings and twins' self-ratings (sr = .27 & .43 respectively),
3) language anxiety (ranging from 0, low anxiety, to 3, high anxiety), while its effect on NC ratings was relatively small (sr = .11).
and 4) attitude towards language learning (ranging from 0, very poor To further explore whether starting to learn a second language
attitude, to 3, very good attitude). MR analyses were chosen since earlier resulted in better outcomes, we decided to conduct more
they assess the unique contribution of each predictor on each proficien- targeted analyses on monolingual students only. Because the relation
cy measure after having accounted for the effects of the other
predictors.
Additionally, because starting SLA instruction before or after the crit- Table 3
Means and standard deviations of the variables of interest for the subset of the sample
ical period could be considered a discrete variable, independent sample used in tests of each proficiency measure.
t-tests were also conducted on the possible effect of age of starting SLA
on proficiency. Proficiency Starting Starting age Starting age SLA SLA
measure age (years) early startersa late startersa anxietyb attitudec
The non-twin analyses have a drawback: including both twins of a
pair in the regression violates the assumption of independence. This Self-ratings 8.86 (2.86) 7.64 (2.40) 12.13 (0.34) 1.29 (0.92) 1.83 (0.67)
NC ratings 9.04 (2.86) 7.69 (2.38) 12.13 (0.33) 1.33 (0.90) 1.88 (0.67)
inflates the p-values but not the test statistic. Therefore, to control for
Teacher 9.40 (2.81) 7.09 (2.41) 12.14 (0.34) 1.35 (0.91) 2.13 (0.72)
the interdependence within twin pairs, we calculated the p values rankings
using mixed linear model analyses (MLM) (which, unlike general linear a
Early starters are children who had started learning the second language before 12 years
models, model not only the means but the variances and covariances of age, late starters had started after that age.
as well), specifying the method of estimation as restricted maximum b
Anxiety scores range from 0, low anxiety, to 3, high anxiety.
likelihood (REML), and entering two variables into the model as c
Attitude scores range from 0, very poor attitude, to 3, very good attitude.
I. Antón-Méndez et al. / Learning and Individual Differences 42 (2015) 44–52 49
Table 4
Multiple regression analyses on all twins. The three proficiency measures were the criteria. The predictors were bilingualism, starting age of ISLA, attitude and anxiety. The p-values have
been corrected for the interdependence between twins⁎.
Self-ratings 461 .43 .18 Bilingualism −.23 .19 −.05 .585 −.01 −.05
Starting age .00 .02 .01 .792 .05 .01
Attitude .58 .06 .45 .001 .42 .43
Anxiety −.09 .05 −.09 .359 .04 −.08
NC ratings 256 .17 .01 Bilingualism .51 .29 .11 .321 .13 .11
Starting age .01 .02 .02 .874 .04 .02
Attitude .14 .08 .11 .043 .13 .11
Anxiety −.00 .07 −.00 .256 .04 −.00
Teacher rankings 188 .29 .06 Bilingualism .24 .35 .05 .493 .07 .05
Starting age −.02 .03 −.05 .576 −.03 −.05
Attitude .38 .10 .29 .016 .27 .27
Anxiety −.09 .08 −.08 .403 .00 −.08
Notes. b = unstandardized beta, SE = standard error, b = standardized beta, r = zero-order correlation, sr = semi-partial correlation.
⁎ Confidence intervals are not provided because they would not be adjusted for interdependence and would be at odds with the adjusted p values.
between starting age and ease of acquisition of a second language is 42) for NC ratings; 0.45 (S.D. = 0.50, N = 27) for teacher rankings. Re-
thought to be discontinuous, i.e., dependent on the student having garding the predictors, the mean of difference scores was 0.87 (S.D. =
had their first contact with the second language before a certain critical 0.78) for performance anxiety, and 0.55 (S.D. = 0.48) for attitude to-
age (see Section 1 for discussion of the critical period hypothesis and wards language learning. Table 6 presents a summary of the MR results.
justification of cut-off age), we divided monolingual students into two The two predictors explain between 39% of the variance for differences
groups: one composed of students who had started learning the second in self-ratings, and 54% of the variance for differences in NC ratings and
language before the age of 12 y.o. (M = 7.64, SD = 2.40); and one com- teacher rankings. Once more, the coefficients for the attitude difference
posed of students who had started learning the second language after scores were significant, explaining between 39 and 50% of the variance
the age of 12 y.o. (M = 12.13, SD = 0.34). We performed independent of the three criterion variables. Surprisingly, differences in anxiety levels
samples t-tests to compare the proficiency of the two groups. The also appear to significantly account for the variance in difference scores
results of these t-tests showed no significant effects of this variable on for NC ratings such that the bigger the differences in anxiety levels, the
any of the proficiency measures: self-ratings, NC ratings, or teachers' bigger the differences in proficiency according to the NC ratings. Addi-
rankings (see Table 5). tionally, although the coefficient for anxiety difference scores did not
It is possible, however, that a linear relationship between starting explain a significant amount of the variance in teacher rankings, it was
age and proficiency exists for one of the two groups, either the early associated with a medium effect similar in size to that of the same pre-
or the late starters, but not the other. Given the different critical periods dictor in NC ratings. Since the sample for the teacher rankings was the
postulated for different linguistic dimensions, it may be, for example, smallest (N = 27), the lack of a significant effect here may be due to
that starting to learn a new language at age 7 instead of at age 10 does lack of power.
matter, but an equivalent age difference after the final critical age
around puberty, say between age 13 and age 16 does not. Or, on the 4. Discussion
contrary, it could be that all learners starting before puberty are equally
able to acquire a new language, but this ability starts gradually decreas- Of the factors that we considered as possible influences on the devel-
ing after puberty (Birdsong, 2006; Singleton, 2005). Therefore, we also opment of proficiency in ISLA, attitude towards language learning
carried out separate MRs for each group, i.e., early and late starters, provided the most robust findings, in line with prevailing findings in
with each of the three proficiency measures as the dependent variables the literature (Dörnyei, 2007; Masgoret & Gardner, 2003). Additionally,
and with the age at which they had started to learn the second language anxiety appeared significant when examining differences between
as a continuous fixed effect. Again, we found no statistically significant monozygotic twins which, to remind the reader, targets the effects of
effect of age for any of the three proficiency measures in either late or the environment by looking for differences in a population, MZ twins,
early starters (all p's N .05). that is genetically identical. Bilingualism and starting age of ISLA, on
the other hand, showed no effect in the analyses.
3.2. MZ twin analyses Attitude is a relatively difficult trait to define and measure. The
literature on the related concept of motivation has identified at least
The means of the absolute difference-scores between twins in a pair three distinct constructs related to it (i.e., instrumental, integrative
were: 0.76 (S.D. = 0.64, N = 76) for self-ratings; 0.29 (S.D. = 0.46, N = and intrinsic motivation), which we tried to measure independently
with separate questions targeting aspects of each separately – i.e., our
Table 5
questionnaire included items to assess the extent to which students con-
T-tests of the difference between early and late SLA beginners on second language sidered learning another language useful (items c and h in Table 2); items
proficiency with, specifically, early and late SLA beginners' sample sizes, average reported to gauge the student's desire to belong to the community of the L2
proficiency (standard deviations), and effect size estimates (Cohen's d) of early SLA start speakers (item g); and their general enthusiasm for learning a language
on standardized scores of the three proficiency measures (self-ratings, NC ratings, and
(items a, d, and f) as well as their more specific affinity for the language
teacher ratings) with p-values adjusted for non-independence.
class (items b and e). Nevertheless, our PCA analysis result indicated
Proficiency measure Early starters Late starters t p d that all these potential motivational dimensions were, in this particular
N M (SD) N M (SD) case, best treated as only one factor which we labelled attitude (as the
Self-ratings 383 2.69 (0.75) 143 2.76 (0.72) 0.44 .974 0.04 reader may remember, attitude is here considered a subset of the larger
NC ratings 202 3.32 (1.37) 88 3.98 (1.55) 0.98 .500 0.12 construct of motivation; see Section 1). As discussed in Section 1, our
Teacher rankings 130 60.77 (25.04) 71 60.57 (26.45) 0.13 .447 0.02 learners may have been too young to show integrative or instrumental
Note. Proficiency means and standard deviations refer to the original scores, not to the motivation and it is likely that what we have thus measured is more
covaried and standardized scores. related to students intrinsic motivation (see highest loading item in
50 I. Antón-Méndez et al. / Learning and Individual Differences 42 (2015) 44–52
Table 6
Multiple regression analyses on monozygotic twin pairs where all variables entered consist of difference scores associated with each twin in a pair. The three proficiency measures were
the criteria. The predictors were attitude and anxiety.
Self-ratings 83 .39 .13 .001 Att .47 .23 to .72 .13 .39 b.001 .39 .39
Anx .00 −.17 to .18 .09 .00 .988 .01 .00
NC ratings 38 .54 .25 .002 Att .36 −.58 to −.14 .11 .49 .002 .42 .48
Anx .23 −.42 to −.04 .10 .35 .022 .25 .34
Teacher rankings 38 .54 .23 .017 Att .46 .13 to .78 .16 .51 .008 .45 .50
Anx .24 −.05 to .53 .14 .30 .102 .19 .29
Notes. b = unstandardized beta, SE = standard error, b = standardized beta, ci = confidence interval, r = zero-order correlation, sr = semi-partial correlation; Att = attitude, Anx =
anxiety.
Table 2), and the influence that their classroom experiences may have on Instead of looking for a correlation between anxiety and proficiency,
it (see high loading items b, d, e, and f in Table 2). As such our results the analysis on MZ twins looks specifically at how differences between
would be in agreement with those of Nikolov (1999) who found that the twins in a pair on anxiety are related to differences in their profi-
the motivations of similarly-aged children (6–14) were more related to ciency. As already explained (Section 1), by limiting the data to geneti-
the classroom experience and the teacher than to other factors. This cally identical twins, the focus is on differences brought about by the
raises the question of whether our finding that attitude predicts profi- environment. This more targeted test tends to be more sensitive,
ciency means that positive attitudes lead to increases in proficiency or which could explain why a relationship between anxiety and proficien-
the reverse, since how well students do in their language learning could cy was only found in the MZ twins' analyses but not in those done on the
very well be driving their attitude rather than their attitude towards whole population. This also means that any anxiety effects found should
language learning ensuring they study and learn better. be the result of an interaction of the student with his or her environ-
While we may not be able to answer that question with total certain- ment which could explain why we did not find anxiety to be significant-
ty, the fact that the same results were obtained in our analyses of the MZ ly related to twins' self-ratings as well as to teachers' ratings. The
twins should help eliminate some of the factors often cited as confound- construct of L2 anxiety is believed to stem from students' low self-
ing the interpretation of this kind of correlational data. One of these esteem (Onwuegbuzie, Bailey, & Daley, 1999) and from students' unre-
possible confounding factors is natural or innate ability, which is likely alistic expectations about language learning leading to disappointment
to be highly associated with attitude and directly affects achievement. with their own progress (Gregersen & Horwitz, 2002), both of which
In this case, given the genetic identity of these twins, MZ discordancy would be expected to result in low self-estimations of proficiency. How-
must be driven by environmental differences, not natural abilities. ever, if anxiety due to low self-esteem is an innate personality trait, the
This extends our non-twin findings in an important way since, with MZ twin analyses would not allow us to detect its effects. Alternatively,
significant non-twin results alone, there is no way to rule out the if the anxiety effect detected in the MZ twin analysis is related to the
claim that our association between attitude and proficiency is simply a classroom environment, it may have more to do with how students per-
consequence of the same genes influencing both. Our finding at both form in class than how they assess their own knowledge. Furthermore,
the non-twin and MZ discordant level shows the effect has to be envi- since the positive correlation means the higher the anxiety levels the
ronmental rather than genetic in nature. That is, whatever the direction higher the proficiency, it is possible that students' anxiety in this case
of causation, at least some of the motivational differences associated was only high enough to make them work more for the subject, rather
with proficiency gains must be driven by the environment. It could be than so high as to impair their performance in class (MacIntyre &
that the language class itself is influencing students' attitudes, whether Gardner, 1994). This could then influence the teachers' evaluations pos-
or not that influence is mediated by achievement. itively, while the students themselves may be rating their L2 knowledge
The other potential confounding factors we can eliminate are envi- at a more general level unrelated to how much they work for the
ronmental in nature: L1–L2 differences (i.e., linguistic distance between subject.
the native language and the language being learned) and differences in The finding that bilingualism had no effect on proficiency was
teacher skills. Different teachers could clearly influence either achieve- surprising since much of the literature on third and subsequent
ment, attitude, or both. Similarly, it seems safe to assume differences language acquisition supports the claim that prior bilingualism gives
between the first language and the language being learned, if uncon- learners advantages in learning a third language (Cenoz, 2001;
trolled, could impact either achievement or attitude. In the present Lasagabaster, 2000, 2001). However, it is possible that it is not being
study, L1–L2 differences can be eliminated because the same language bilingual but more specifically being biliterate that provides the advan-
was being studied by each member of a twin pair in the MZ twin anal- tages to further language learning. As well as the studies indicating that
yses, and differences in teacher skill can be (almost) eliminated because it is the biliterate learners that show advances in proficiency in the third
in all but 6 pairs the twins had the same teacher as their co-twin. If atti- language (Keshavarz & Astaneh, 2004; Sanz, 2000), Thomas (1988)
tude is in fact driving achievement differences, it may result from such found that bilinguals who had received formal instruction in their sec-
things as friendship groups not shared by members of a twin pair, or ond language outperformed – in learning of the third language – those
personal affinities for particular languages. If, on the other hand, who had only learned the second language in the home environment.
achievement is driving attitude differences, that may come about, for While it is not clear why this should be so, we can speculate that the bi-
example, by preferential treatment by a teacher for one twin over the lingual advantages seen in SLA may stem from the increased metalin-
other. guistic awareness which, although found to be stronger in bilinguals
Apart from attitude towards language learning, the other factor that in general (e.g., Jessner, 1999), is likely to be greatly enhanced by having
also appears to correlate positively with attainment is anxiety. Although had explicit instruction about more than one language.
this result only appeared significant when proficiency was estimated by In any case, our lack of significant findings relating to bilingual-
the teacher according to the NC rating scale, the fact that this is the most ism could stem from the fact that the bilingual twins in the study
standardized and possibly the most nuanced of the three proficiency had all acquired bilingualism in the home and were not likely to
measures included, together with the substantial size of the effect have had any formal instruction in the home language. In fact, only
(sr = .34) and a similar effect size for the teacher rankings (sr = .29), 11 of 62 bilingual twins claimed to be somewhat literate in the
gives some credence to the result. home language.
I. Antón-Méndez et al. / Learning and Individual Differences 42 (2015) 44–52 51
Herdina, P., & Jessner, U. (2002). A dynamic model of multilingualism. Clevedon: Multilin- Nikolov, M. (1999). ‘Why do you learn English?’ ‘Because the teacher is short’. A study of
gual Matters. Hungarian children's foreign language learning motivation. Language Teaching
Horwitz, E. (2001). Language anxiety and achievement. Annual Review of Applied Research, 3(1), 33–56.
Linguistics, 21, 112–126. Norris, J. M., & Ortega, L. (2000). Effectiveness of L2 instruction: A research synthesis and
Horwitz, E., Horwitz, M. B., & Cope, J. (1986). Foreign language classroom anxiety. Modern quantitative meta-analysis. Language Learning, 50(3), 417–528. http://dx.doi.org/10.
Language Journal, 70, 125–132. 1111/0023-8333.00136.
Hyltenstam, K., & Abrahamsson, N. (2003). Maturational constraints in SLA. In C. J. Doughty, Oliver, B., Harlaar, N., Hayiou Thomas, M. E., Kovas, Y., Walker, S. O., Petrill, S. A., et al.
& M. H. Long (Eds.), Handbook of second language acquisition (pp. 539–588). Malden, (2004). A twin study of teacher-reported mathematics performance and low perfor-
MA: Blackwell. mance in 7-year-olds. Journal of Educational Psychology, 96(3), 504–517. http://dx.doi.
Ioup, G., Boustagui, E., El Tigi, M., & Moselle, M. (1994). Reexamining the critical period hy- org/10.1037/0022-0663.96.3.504.
pothesis: A case of successful adult SLA in a naturalistic environment. Studies in Second Onwuegbuzie, A. J., Bailey, P., & Daley, C. E. (1999). Factors associated with foreign
Language Acquisition, 16(01), 73–98. http://dx.doi.org/10.1017/S0272263100012596. language anxiety. Applied Psycholinguistics, 20, 217–239.
Jessner, U. (1999). Metalinguistic awareness in multilinguals: Cognitive aspects of third Ortega, L. (2009). Understanding second language acquisition. London: Hodder (Vol.).
language learning. Language Awareness, 8(3 & 4), 201–209. Paradis, M. (2004). A neurolinguistic theory of bilingualism. Amsterdam John Benjamins
Johnson, J. S., & Newport, E. L. (1989). Critical period effects in second language learning: Publishing.
the influence of maturational state on the acquisition of English as a second language. Plomin, R., Kovas, Y., & Haworth, C. M. A. (2007). Generalist genes: Genetic links between
Cognitive Psychology, 21(1), 60–99. brain, mind, and education. Mind, Brain, and Education, 1(1), 11–19. http://dx.doi.org/
Johnson, J. S., & Newport, E. L. (1991). Critical period effects on universal properties of lan- 10.1111/j.1751-228X.2007.00002.x.
guage: The status of subjacency in the acquisition of a second language. Cognition, Plomin, R., & Walker, S. O. (2003). Genetics and educational psychology. British Journal of
39(3), 215–258. http://dx.doi.org/10.1016/0010-0277(91)90054-8. Educational Psychology, 73(1), 3–14. http://dx.doi.org/10.1348/000709903762869888.
Keshavarz, M. H., & Astaneh, H. (2004). The impact of bilinguality on the learning of En- Pulvermüller, F., & Schumann, J. H. (1994). Neurobiological mechanisms of language
glish vocabulary as a foreign language (L3). International Journal of Bilingual Education acquisition. Language Learning, 44(4), 681–734.
and Bilingualism, 7(4), 295–302. http://dx.doi.org/10.1080/13670050408667814. Ross, S. (1998). Self-assessment in second language testing: A meta-analysis and analysis
Kormos, J., & Csizér, K. (2008). Age-related differences in the motivation of learning of experiential factors. Language Testing, 15, 1–20.
English as a foreign language: Attitudes, selves, and motivated learning behavior. Rossi, S., Gugler, M., Friederici, A., & Hahne, A. (2006). The impact of proficiency on
Language Learning, 58(2), 327–355. http://dx.doi.org/10.1111/j.1467-9922.2008. syntactic second-language processing of German and Italian: Evidence from event-
00443.x. related potentials. Cognitive Neuroscience, 18, 2030–2048. http://dx.doi.org/10.1162/
Krashen, S., Long, M. H., & Scarcella, R. (1979). Accounting for child–adult differences in jocn.2006.18.12.2030.
second language rate and attainment. TESOL Quarterly, 13, 573–582. Ryan, R. M., & Deci, E. L. (2000). Self-determination theory and the facilitation of intrinsic
Kuhl, P. K., Tsao, F. M., & Liu, H. M. (2003). Foreign-langauge experience in infancy: motivation, social development and well-being. The American Psychologist, 55(1),
Efffects of short-term exposure and social interaction on phonetic learning. 68–78.
Proceedings of the National Academy of Science, 100(15), 9096–9101. Sanz, C. (2000). Bilingual education enhances third language acquisition: Evidence from
Lasagabaster, D. (2000). Three languages and three linguistic models in the Basque Catalonia. Applied Psycholinguistics, 21(1), 23–44.
educational system. Bilingual Education and Bilingualism, 179–197. Sebastián-Gallés, N., Echevarría, S., & Bosch, L. (2005). The influence of initial exposure on
Lasagabaster, D. (2001). Bilingualism, immersion programmes and language learning in lexical representation: Comparing early and simultaneous bilinguals. Journal of
the Basque country. Journal of Multilingual and Multicultural Development, 22(5), Memory and Language, 52, 240–255.
401–425. http://dx.doi.org/10.1080/01434630108666443. Segalowitz, N., & Frenkiel-Fishman, S. (2005). Attention control and ability level in a
Lenneberg, E. (1967). Biological foundations of language. New York: Wiley. complex cognitive skill: Attention shifting and second-language proficiency.
MacIntyre, P. D., Clément, R., Dörnyei, Z., & Noels, K. A. (1998). Conceptualizing willing- Memory and Cognition, 33, 644–653.
ness to communicate in an L2: A situational model of L2 confidence and affiliation. Sharwood-Smith, M. (1994). Second language learning. London: Longman.
Modern Language Journal, 82, 545–562. Singleton, D. (2005). The critical period hypothesis: A coat of many colours. International
MacIntyre, P. D., & Gardner, R. C. (1994). The subtle effects of language anxiety on cogni- Review of Applied Linguistics in Language Teaching, 43(4), 269–285. http://dx.doi.org/
tive processing in the second language. Language Learning, 44, 283–305. 10.1515/iral.2005.43.4.269.
Masgoret, A. M., & Gardner, R. C. (2003). Attitudes, motivation, and second language Thomas, J. (1988). The role played by metalinguistic awareness in second and third
learning: A meta-analysis of studies conducted by Gardner and associates. Language language learning. Journal of Multilingual and Multicultural Development, 9, 235–246.
Learning, 53(1), 123–163. http://dx.doi.org/10.1111/1467-9922.00212. Ullman, M. T. (2001). The neural basis of lexicon and grammar in first and second
McDonald, J. L. (2006). Beyond the critical period: Processing-based explanations for poor language: The declarative/procedural model. Bilingualism: Language and Cognition,
grammaticality judgment performance by late second language learners. Journal of 4(1), 105–122.
Memory and Language, 55(3), 381–401. http://dx.doi.org/10.1016/j.jml.2006.06.006. Ullman, M. T. (2004). Contributions of memory circuits to language: The declarative/
Mitchell, R., & Myles, F. (1998). Theoretical perspectives on second language learning. procedural model. Cognition, 92, 231–270.
Arnold Publishers. Verhoeven, L., & Vermeer, A. (2002). Communicative competence and personality dimen-
Mitchell, R., & Myles, F. (2004). Second language learning theories (2nd ed.). New York: sions in first and second language learners. Applied Psycholinguisitics, 23, 361–474.
Arnold. Visscher, P. M., Benyamin, B., & White, I. (2004). The use of linear mixed models to
Muñoz, C. (Ed.). (2006). Age and the rate of foreign language learning. Clevedon, UK: estimate variance components from data on twin pairs by maximum likelihood.
Multilingual Matters. Twin Research and Human Genetics, 7(6), 670–674.