Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Gisela Jia
Lehman College, The City University of
New York Purpose: This 5-year longitudinal study investigated the acquisition of 6 English
grammatical morphemes (i.e., regular and irregular past tense, 3rd person singular,
Akiko Fuse progressive aspect –ing, copula BE, and auxiliary DO) by 10 native Mandarin-
Brooklyn College and Graduate Center, speaking children and adolescents in the United States (arrived in the United States
The City University of New York between 5 and 16 years of age). The goals were to chart and compare the acquisition
trajectories and levels of mastery across the morphemes, identify when age-related
differences emerged and which forms they took.
Method: Morphological proficiency was measured by the accuracy of these
morphemes in obligatory contexts during spontaneous speech.
Results: The morphemes were mastered by different numbers of participants and
showed different growth trajectories. Performance variance was partially predicted
by age of arrival (AoAr) in the United States, with early arrivals achieving greater
proficiency than late arrivals. However, such AoAr effects took several years to occur
and only existed for 2 of the 6 morphemes (i.e., 3rd person singular and regular past
tense). Growth curve analysis revealed that language environment was a stronger
predictor of individual differences than AoAr. Results did not uncover age-related
differences in the acquisition of tense versus non-tense-related morphemes, nor in
regular versus irregular morphemes, nor in the error types.
Conclusion: Findings support an Environmental account for age-related differences in
2nd language (L2) morphological acquisition. Results also indicate that the acquisition
of some grammatical morphemes by school-aged immigrants takes several years
to complete. As L2 learners exhibit some error types and difficulties similar to
monolingual children with specific language impairment, caution needs to be taken
when interpreting and using morphological errors as indicators of speech /language
learning problems in this population.
KEY WORDS: bilingualism, English language learners, sequential bilingualism
P
revious research has shown consistently that the acquisition of gram-
matical morphemes presents significant challenges to second lan-
guage (L2) learners. Different from native language (L1) learners, L2
learners constitute a dramatically heterogeneous population. They learn
L2 in different settings (e.g., immersion through immigration, or foreign
language instructions at home country). The current study focuses on
the immigration setting. Among the many variables that set L2 learn-
ers apart, age of L2 exposure is an important one. In the immigration set-
ting, those with younger ages of L2 exposure (indexed by ages of arrival
[AoAr] in the L2-speaking country) tend to obtain greater L2 proficiency
than those with older AoAr in the long run. However, the explanations for
1280 Journal of Speech, Language, and Hearing Research • Vol. 50 • 1280–1299 • October 2007 • D American Speech-Language-Hearing Association
1092-4388/07/5005-1280
such age-related differences have been under debate. the nature of the AoAr effect has been controversial, al-
They range from maturational constraints postulated by though it carries much theoretical and applied significance.
the Critical or Sensitive Period account, to language envi- Three general theoretical accounts have been put
ronment differences postulated by the Environmental forth to explain the AoAr effect found in long-term at-
account. tainment studies: the Critical Period account, the L1
To advance our understanding of the applicability of Transfer account, and the Environmental account. Ac-
different theoretical accounts, we examined how 10 na- cording to the Critical Period account, the genetically
tive Mandarin-speaking children and adolescents with guided maturation of a domain specific language learn-
different AoAr in the United States acquired six English ing mechanism underlies the declining ability to learn L2
grammatical morphemes over a 5-year period. The first grammatical morphology (e.g., Lenneberg, 1967; Patkowski,
goal was to identify the timing in which age-related dif- 1990). After a certain time window (i.e., a critical or
ferences occur and to examine how much variance in per- sensitive age period), L2 acquisition becomes severely
formance among participants could be uniquely predicted constrained by maturational factors, which cannot be
by AoAr and language environment. The second goal was overridden by even a rich L2 environment. The L1 Transfer
to investigate the specific forms that age-related differ- account posits that processing of a new language is ham-
ences took in terms of error types and relative difficulty pered by the strength of the linguistic processing routines
with different morphemes. Achieving these goals would automatized in native language processing. In compari-
lead to a deeper understanding of the processes and the son with early arrivals, late arrivals have used their native
underlying mechanisms of L2 morphological acquisition. language for more years and thus experience stronger
Such knowledge can also inform practitioners diagnosing transfer from L1 to L2 that interferes with some aspects of
and treating speech, language, and learning problems of L2 acquisition. According to the Environmental account
bilingual populations, as well as English language edu- (Jia & Aaronson, 2003; Marinova-Todd, Marshall, & Snow,
cators working with these populations. 2000; Snow, 1983; Snow & Hoefnagel-Höhle, 1987), in
the immigration setting, early arrivals are exposed to a
Age-Related Differences in L2 significantly richer L2 environment than late arrivals,
in both quantity and quality (see later sections for details
Morphological Acquisition: of relevant findings). Such systematic environmental
Theoretical Accounts differences lead to higher levels of L2 attainment by early
Age-related differences in L2 morphological acqui- arrivals in the long run (Jia & Aaronson, 2003).
sition have been most explicitly dealt with by a group of In reality, all three accounts can apply to the immi-
studies on the long-term grammatical proficiency of L2 grant populations, in the sense that early arrivals are
learners (e.g., Bialystok & Miller, 1999; Birdsong & Molis, within a sensitive time window, have a richer L2 envi-
2001; Flege, Yeni-Komshian, & Liu, 1999; Jia, Aaronson, ronment, and experience less interference from their L1.
& Wu, 2002). Participants in these studies were adult bi- Where researchers diverge is in the degree to which a
linguals who had arrived in the L2-speaking country at given account can effectively explain the AoAr effects. To
different ages and had resided there for at least several fully understand this issue, we may need findings ob-
years. The studies measured grammatical sensitivity by tained from many sources, including investigations at
having participants detect violations of L2 grammatical different levels of analysis (e.g., behavioral and neuro-
rules (mostly morphological) in spoken or written L2 sen- logical), from different angles (e.g., contrasting different
tences. One predictor of L2 morphological proficiency that L1–L2 populations; contrasting the acquisition of L2
has been consistently found across these studies was AoAr. structures bearing different linguistic relations to L1),
It often predicts the largest amount of variance in mor- and using different approaches (e.g., longitudinal and
phological proficiency (e.g., Flege et al., 1999; Jia et al., cross-sectional). A finding can offer direct support for an
2002). More specifically, those who arrived in the L2- account (e.g., showing the exclusive explanatory power
speaking country at younger ages, usually before 9 years of a factor over others), or be consistent or inconsistent
of age (hereinafter called early arrivals), tended to perform with predictions drawn from a given account, or pose a
at or close to native speaker levels, whereas those who challenge to an account, or simply fail to generate sup-
arrived at older ages ( hereinafter called late arrivals) port for an account. In light of the range of ways that re-
tended to perform significantly poorer than the native search findings can address theoretical accounts, the
speakers and early arrivals. This finding, termed the AoAr current study was intended to illuminate the differential
effects, has been repeatedly obtained from English L2 learn- contributions of maturational constraints (as postulated
ers with various native languages, including Mandarin by the Critical Period account) and language environ-
(Jia et al., 2002), Korean (Flege et al., 1999), Vietnamese ment (as postulated by the Environmental account) to
(McDonald, 2000), and Spanish (Bialystok & Miller, 1999; the AoAr effects in L2 morphological acquisition in the
Jia, Aaronson, Young, Chen, & Wagner, 2005). However, immigration setting. Below, we discuss findings from
1282 Journal of Speech, Language, and Hearing Research • Vol. 50 • 1280–1299 • October 2007
the learning mechanism, needs to be supported by find- morphemes was significantly lower (48.81%) than that for
ings of qualitative differences between younger and older non-tense-related morphemes (70.58%). These findings
learners in specific aspects of morphological learning. In led to the conclusion that the morphological acquisition
contrast, the Environmental account is compatible with a profiles of child L2 learners resembled those of mono-
language learning mechanism that is consistent across lingual children of the same age with SLI in exhibiting
different ages and, thus, does not need to be supported by more difficulties with tense-related morphemes.
evidence of qualitative differences among learners of dif- The current study investigated whether the find-
ferent ages. ings from child L2 learners can be generalized to a sam-
Further, from an applied perspective, only knowl- ple of native Mandarin speakers with a more varied age
edge about the specific challenges encountered by learn- of exposure to English. Particularly, given that tense-
ers of different ages can lead to effective instruction and related morphemes present a larger challenge to child
intervention. Drawing upon existing literature in various L2 learners than non-tense-related morphemes, would
language acquisition populations, the current study ex- this difference be more or less pronounced among older
plored three aspects of L2 morphological acquisition with L2 learners? Would the potential age-related differences
potential age-related differences: (a) the degree of perfor- in this aspect uncover something deeper about the changes
mance discrepancy between tense- and non-tense-related of the L2 acquisition mechanisms across age? No existing
morphemes, (b) error patterns, and (c) the acquisition of findings are available to address this issue. In the current
regular versus irregular morphemes. study, we compared early and late arrivals’ performance
on individual morphological structures, and we observed
The first issue concerns whether age-related differ-
whether the relative difficulty levels of tense- and non-
ences exist in the degree of performance discrepancy be-
tense-related morphemes varied between the two groups.
tween tense- and non-tense-related morphemes. Interest
Consistent with previous studies (e.g., Paradis, 2005;
in this aspect of learning originated in the study of chil-
Rice & Wexler, 1996), the tense-related morphemes in-
dren with specific language impairment (SLI). Monolingual
cluded regular and irregular past tense, third person
children with SLI have particular difficulties acquiring singular, auxiliary DO, and copula BE, and the non-tense-
tense-related morphemes (i.e., regular and irregular past related morpheme was progressive aspect –ing. To en-
tense, third person singular, copula and auxiliary BE, hance the tense and nontense contrast, performance on
and auxiliary DO) but not non-tense-related morphemes another non-tense-related structure, plural, was included
(i.e., plural, possessive –s, articles, and progressive aspect in the nontense group. Except for this analysis, data on
forms; e.g., Bedore & Leonard, 1998; Rice & Wexler, the plural morpheme were not included in other anal-
1996). From a linguistic perspective, some researchers yses of the current report (see Jia, 2003, for other findings).
have suggested that monolingual children with SLI have
The second issue is whether younger and older L2
limitations in connecting the abstract functional category
learners’ error patterns differ. Most morphological pro-
Tense (e.g., rather than Agreement) with its surface mor-
duction errors of L1 and L2 learners are omission errors.
phological forms (Wexler, 1998). Currently, there is a
That is, speakers tend to use nonfinite verbs (verb stems
growing interest in directly contrasting multiple lan-
without inflections) in finite positions in which morpho-
guage learning populations, including L1 and L2 learn-
logical inflections are necessary. The overuse of nonfinite
ers, with and without SLI. The general purpose is to find
verbs (underuse of morphemes) is proposed to be due to
the differences and similarities among these groups of
either missing functional categories underlying the mor-
learners, illuminate and constrain language acquisition
phemes (the impairment view) or problems with mapping
theories that supposedly explain all these similarities from abstract categories to their surface morphological
and differences, and identify markers that distinguish representations (the missing surface inflection hypothesis;
typically developing and nontypically developing learn- see Ionin & Wexler, 2001, for a review). Two findings have
ers in each of the L1 and L2 learning populations. been used as evidence to support the missing surface in-
In line with this interest, the study of tense-related flection hypothesis. First, whereas L2 learners tend to use
and non-tense-related morphemes has been adapted in nonfinite verbs in finite positions, they rarely use finite
L2 acquisition research. Paradis (2005) studied 24 typi- verbs in nonfinite positions (Paradis & Crago, 2000;
cally developing English L2 learners with ages ranging be- Prévost & White, 1999). For example, L2 learners may
tween 50 and 94 months from various L1 backgrounds inappropriately drop the third person singular “s” in
(e.g., Mandarin, Cantonese, Spanish, and Korean). Par- “She like(s) to go to school” (omission errors), but they
ticipants were all child L2 learners (age of English expo- rarely mistakenly add “s” as in “They like(s) to go to
sure ranged from 3 to 6 years of age) and had been school” or “She begin to like(s) to go to school” (commis-
exposed to English for about 2 to 18 months. In obligatory sion errors). Such findings are argued to indicate that L2
contexts of spontaneous speech, accuracy for tense-related learners have the knowledge and thus intact underlying
1284 Journal of Speech, Language, and Hearing Research • Vol. 50 • 1280–1299 • October 2007
arrivals make a similarly high proportion of omission er- language background questionnaire, and observations of
rors? and (c) Do early and late arrivals exhibit the same the child’s language use. All sessions were conducted by
trend of discrepancy in performance on regular versus ir- the first author, a fluent Mandarin–English bilingual.
regular morphemes? Because of some problems of scheduling and recording,
Anna (f5), Carl (m7), and Eric (m9) each missed two, one,
and one testing session(s), respectively. The current ar-
ticle reports results from a total of 156 data sessions
Method across all participants.
Participants Target morpheme production measures. Production
of all the target morphemes was coded from spontaneous
Participants were the same 10 native Mandarin-
speech and additionally with an elicitation task for pro-
speaking children and adolescents as described in Jia
gressive aspect. Spontaneous speech samples were col-
(2003): Anna (f 5), Betty (f 6), Carl (m7), Dianna (f 8), Eric
lected through conversations between the interviewer
(m9), Frank (m9), Gary (m12), Hua-lei (f12), Jing-lan
and the participants about their activities in school and
(f 15), and Kang-da (m16). These were 5 girls and 5 boys
at home, their language use in various situations, and
who were from 5 to 16 years of age upon immigration to the
undirected story telling prompted by a series of picture
New York City area. Participants’ pseudonyms are speci-
scripts. The length of the samples of each session ranged
fied with their gender and AoAr in parentheses. For all
from about 30 to 45 min. Only for the progressive aspect,
participants, English immersion began when they en-
an additional elicitation task was used. Participants
tered New York City public schools. Only the two oldest
were shown a set of eight pictures, each depicting a per-
participants had previous exposure to English through
son(s) doing something. Participants were asked to “use
classroom instruction in their native country. All par-
a complete English sentence to describe what is happen-
ticipants attended English-speaking schools where all sub-
ing.” A different set of pictures was used for each testing
jects were taught in English and over 70% of the students
session. The appropriate pictures were first selected and
were native English speakers. All received focused En-
then randomly assigned to different testing sessions. As
glish language instructions from English as a Second Lan-
exploratory statistical analysis did not show differences
guage (ESL) programs, with length of attendance varying
in the performance on the elicitation task and during
from 1 to 3 years across participants. The parents of
spontaneous speech, the data from both sources were com-
the participants all possessed English communication
bined for the progressive aspect structure.
skills, ranging from basic to fluent. The differences in the
length of ESL education and parental English levels ex- Language environment measures. Participants’ lan-
isted evenly across participants of different ages. Lan- guage environment was evaluated by (a) an annual parental
guage use with siblings was an insignificant part of questionnaire written in Chinese, (b) child and parent in-
language environment because 8 participants were only terviews, and (c) the interviewer’s observations of the par-
children, and the other 2 were sisters who exclusively ticipants’ language use at home with their family members
used Chinese with each other. As immigrants with AoAr and peers. At the end of each year, parents filled out a
younger and older than 9 years have been shown to ex- questionnaire to report their children’s language use in
hibit distinctive patterns in the relative levels of L1 and various situations over the previous year. In the question-
L2 obtained in the long run (e.g., Jia et al., 2002; Yeni- naire, parents reported (a) the number of hours of TV their
Komshian, Flege, & Liu, 2000), the 10 participants were children watched in each language during each week,
subdivided into a group of 6 early arrivals (with AoAr as (b) the number of books they read in each language dur-
9 years and younger) and a group of 4 late arrivals (with ing each school semester and the two school breaks, (c) the
AoAr as 12 years and older). number of friends they had who predominantly spoke ei-
ther Chinese or English during the entire year, and (d) the
pecentage of time that they spoke each language at home
Procedures during the year. These four aspects were combined to yield
Testing sessions. The study started 3 months af- an English L2 environment composite score for each par-
ter participants’ English immersion through entering ticipant for each year of the study (see Jia & Aaronson,
English-speaking schools. Sixteen testing sessions were 2003, for details of this measure). Each score ranges from
conducted in participants’ homes during the first 5 years 0 to 100, and a higher score indicates a richer L2 environ-
of their residence in the United States. These included ment and a poorer L1 environment. The parent ques-
seven monthly sessions during Year 1, four quarterly ses- tionnaire data were highly consistent with data obtained
sions during Year 2, two semiannual sessions each during from child interviews and observations. To summarize the
Years 3 and 4, and a single session during Year 5. Each findings, throughout the 5 years of the study, a younger
session lasted about 4 hr and consisted of five basic activ- AoAr predicted a significantly richer L2 environment and
ities: language tasks, child interview, parental interview, poorer L1 environment. From the third year on, early arrivals
1286 Journal of Speech, Language, and Hearing Research • Vol. 50 • 1280–1299 • October 2007
used because its corpus age samples are closest to those Table 1. Number of obligatory contexts (averaged across all testing
of the participants in the current study. One of its fre- sessions) for each participant.
quency value indictors, U, that is scaled in terms of
frequency-per-million was used as the frequency index Participant 3rd PT (r) PT (ir) DO BE –ING
for the past tense verbs.
Anna (f5) 12 29 50 11 27 22
Coding of third person singular in the obligatory con- Betty (f6) 18 15 31 12 20 17
text. The criterion for the obligatory context for third per- Carl (m7) 8 8 16 8 10 17
son singular is the simultaneous presence of a verb in a Dianna (f8) 30. 16 41 30 36 30
present tense context with a third person subject. Cor- Eric (m9) 13 18 24 13 11 16
rect productions included the regular form of adding of Frank (m9) 19 5 17 19 6 19
“s” to a verb, and the irregular form of changing “ have” Gary (m12) 24 10 30 24 18 17
Hua-lei (f12) 10 6 10 10 7 18
into “ has.” Omission errors included the absence of the
Jing-lan (f15) 12 9 19 12 14 21
third person singular morpheme “s,” or not changing
Kang-da (m16) 13 14 21 13 14 22
“ have” into “ has.” The only commission errors were the
use of past tense to indicate third person singular form, Note. 3rd = third person singular; PT (r) = past tense regular; PT (ir) =
as in “She is checking if she had any money.” Formulaic past tense irregular; DO = auxiliary DO; BE = copula BE; –ING =
expressions such as “ Who cares?” were not coded. progressive aspect –ing.
Coding of auxiliary DO in the obligatory context. The
criterion for the obligatory DO context is the presence of
interrogative and negative sentences with a main verb. Intercoder agreement rates ranged from 80.2% to 95.4%,
Correct use of DO refers to the use of DO with correct with an average of 88.2% for morpheme counting based
tense and person agreement with the subject. Omission on evaluations of the middle four rounds of data (7, 8, 9,
errors refer to the absence of DO or DOES in these con- and 10) of each participant. Most disagreements were
texts. Commission errors refer to the use of DO with the resolved by consensus agreement involving a third na-
wrong agreement (all cases involved using “DO” for tive English speaker and the first author. When agree-
“DOES”) or with the wrong tense. Only one instance of ment could not be reached, the items were excluded from
an additional commission error occurred in which the analysis.
use of DO was mistakenly substituted with “ BE” (“I’m
not like the movie”). Formulaic expressions such as “I
don’t know” were not coded. Results
Coding of copula BE. The criteria for the obligatory To address the first goal of the study, age-related dif-
copula BE context is the presence of a subject and its pred- ferences were assessed by (a) examining whether the
icate (an adjective or noun phrase that denotes some number of morphemes mastered by participants varied
properties of the person(s) or things(s) referred to by the according to AoAr, and (b) examining at what time point
subject). Correct use of BE refers to the use of BE with of L2 immersion significant negative correlations be-
correct tense and person agreement with the subject. tween AoAr and performance on each morpheme (indi-
Omission errors refer to the absence of BE in these con- cating a younger learner advantage) started to emerge.
texts. Commission errors refer to the use of BE with Further, growth curve analysis was conducted to com-
wrong agreement (e.g., “She are late”) or with wrong pare the growth rates and trajectories of individual mor-
tense (e.g., “Yesterday she is late”). Formulaic expres- phemes of each participant and investigate whether AoAr
sions such as “It’s like” and “I’m like” were not coded. uniquely explained a significant portion of performance
Coding of progressive aspect –ing. The progressive as- variance when language environment was taken into ac-
pect denotes an ongoing action. In obligatory context, cor- count. To address the second goal of the study, perfor-
rect cases refer to a verb inflected with –ing. Omission mance across individuals (thus AoAr) was assessed by
errors were bare verb stems without –ing (e.g., “He was do comparing scores on tense- and non-tense-related mor-
something”). No commission errors were found. phemes, and on regular and irregular morphemes, as well
as by analyzing the error types (i.e., commission vs. omis-
Table 1 lists the number of obligatory context for each
sion errors) made.
structure averaged across all the testing sessions for each
participant. The correlations between AoAr and number of
obligatory contexts for all structures were not statistically Mastery of the Morphemes
significant, indicating that differences in number of oblig- This section of the data analyses examined the ac-
atory contexts for all structures were not related to AoAr. quisition profiles of individual morphemes shown by each
Intercoder reliability. The coding was conducted in- participant using mastery of morpheme as an indicator of
dependently by two trained native English speakers. proficiency. Figures 1–6 present the percentage correct in
obligatory contexts per participant over the 16 testing Individual differences also existed in the mastery of
sessions by token counts for each of the six morphologi- the seven structures. The number of structures each
cal structures. For clarity, data for each structure were participant mastered ranged from six (Betty [f 6] and
charted on two separate panels, one for the 6 younger and Dianna [f 8]) to one (Kang-da [m16]). The correlation be-
one for the 4 older participants. Mastery of a structure is tween AoAr and the number of structures mastered was
defined by over 80% correct use of the morpheme in ob- not statistically significant, r(8) = –.53. The correlations
ligatory contexts across three consecutive testing sessions between AoAr and performance on various structures
as is traditional in monolingual literature (e.g., Lahey, were obtained for each data session, with significant neg-
Liebergott, Chesnick, Menyuk, & Adams, 1992). Table 2 ative correlations indicating AoAr effects. Though most of
lists the time at which participants mastered the six struc- the correlations were not statistically significant, the
tures, with the data on plurals adopted from a previously trend was consistent. Across the structures, over time,
published report (Jia, 2003).
the correlations changed from positive to negative. Ini-
The number of participants who mastered each struc- tially, there were consistently positive values (indicat-
ture varied dramatically. Present progressive –ing was ing an older learner advantage). Then, between the 1st
mastered by all participants, followed by plural (by 7 par- and 2nd year of L2 immersion, the values became and
ticipants), auxiliary DO (by 6 participants), and copula BE stayed consistently negative (indicating a younger learner
(by 5 participants). Third person singular and irregular advantage). It was not till the end of the study that a
past tense were mastered by only 3 and 4 participants, re- younger AoAr predicted significantly better average per-
spectively, and regular past tense –ed by none. formance on third person singular, r(8) = –.87, p < .01, and
1288 Journal of Speech, Language, and Hearing Research • Vol. 50 • 1280–1299 • October 2007
Figure 2. The scatterplot of length of second language (L2) immersion and the progressive aspect
(–ing) production accuracy of the younger arrivals (top) and older arrivals (bottom).
the correlations approached a significant level for past the group mean. In the present study, this was accom-
tense regular, r(8) = –.60, p = .09, and for the average per- plished through the use of a mixed model approach that
formance on all structures, r(8) = –.59, p = .08. allows one to obtain intercept and slope coefficients unique
to each participant. These are then compared with the
average intercept and slope characterizing the sample
Growth Curve Analyses as a whole. Differences between the average trajectory
To further examine the developmental trajectories and developmental trajectories of each child were sum-
of the structures, growth curve analyses were conducted marized in single estimates of the variability of individual
to answer the following questions: (a) Did the accuracy of intercepts and slopes around the sample averages. The
participant’s morpheme production grow linearly or non- software used to conduct the analyses was SAS PROC
linearly? (b) Were there individual differences in the rate MIXED, a program for fitting multilevel models, hierar-
and type of growth? (c) How were such individual dif- chical linear models, and individual growth models
ferences predicted by variables such as age and language (Singer, 1998; Singer & Willett, 2003).
environment? Three nested models were used. Model 1 examined
The analyses were done using hierarchical linear mod- whether the overall growth trajectory was linear or non-
eling (HLM) procedures that analyze individual differences linear, and whether there were significant intercept and
in growth and change processes (Bryk & Raudenbush, slope differences in the individual growth trajectories.
1992). HLM models investigate whether there are sta- The existence of linear and nonlinear (quadratic) changes
tistically significant departures among participants in over time varied across the structures (see Table 3). Reg-
their initial statuses and/or their rates of change from ular past tense –ed showed no linear or quadratic growth.
Auxiliary DO showed linear but nonquadratic growth. Model 1 existed only for the quadratic part of the function.
Irregular past tense, present progressive –ing, copula BE, That is, during the early period of accelerated growth,
and third person singular showed both linear and qua- participants did not differ in their speed of development.
dratic growth (including marginally significant results). When growth approached asymptotic levels as a group,
Significant and marginally significant slope differences the developmental functions started to vary in trajecto-
were found for third person singular and progressive –ing, ries. The case for the third person singular was the op-
respectively. Significant intercept was found for all six posite. Mere elimination of the quadratic function left
morphological structures (marginally significant for reg- the slope significant. This indicates that the slope dif-
ular past tense), reflecting the fact that, at the third month ferences for this structure existed during the early growth
of L2 immersion, participants already showed different of the structure but not when the growth reached asymp-
levels of accuracy. totic levels.
Model 2 examined whether the slope differences ex- Model 3 examined whether the differences in the in-
isted for only the nonlinear part of the function or both tercept and slope found for five of the six structures were
the nonlinear and linear part by eliminating the linear predicted by age-related and language environment–
term and the quadratic term of the function separately related variables (see Table 3). The two explanatory var-
(see Table 3). For progressive –ing, eliminating the linear iables entered included AoAr and L2 environment score
term of the function still resulted in significant slope dif- averaged across all the testing sessions. L2 environment
ferences, whereas eliminating the quadratic term of the was a significant predictor of the intercept differences for
function resulted in nonsignificant slope differences. This past tense irregular and the slope and intercept differences
indicates that the significant slope differences detected in of third person singular. Performance on auxiliary DO and
1290 Journal of Speech, Language, and Hearing Research • Vol. 50 • 1280–1299 • October 2007
Figure 4. The scatterplot of length of second language (L2) immersion and copula BE (BE)
production accuracy of the younger arrivals (top) and older arrivals (bottom).
copula BE were not accounted for by any of the two var- (within-subjects variable), and between the younger and
iables (see Table 3). older group (between-subjects variable). With this lay-
out of analysis, we can observe performance on the two
non-tense-related morphemes in relation to performance
Tense-Related and Non-Tense-Related on the other five tense-related morphemes at individual
Morphemes morpheme level.
One approach to compare participants’ performance There was a significant main effect of structure,
on the tense- and non-tense-related morphemes is to ob- F(6, 48) = 35.27, effect size = .82, p < .001. This reflects the
tain the average percentage correct scores separately for finding that average percentage correct scores (collapsed
all the five tense-related structures and the two non- across participants) varied significantly across structures.
tense-related structures for each data session and for The ranking order of the percentage correct scores
each participant. However, as the current study only in- (from the highest to lowest) is the following: auxiliary
cluded two non-tense-related morphemes (i.e., progressive DO (83.87%), progressive –ing (83.13%), copula BE
aspect –ing and plural) in contrast to five tense-related (71.36%), plural (68.97%), past tense irregular (65.65%),
morphemes, such a comparison lacks sufficient validity. third person singular (44.82%), and past tense regular
Therefore, we chose to conduct a mixed two-way analysis (40.26%).
of variance, with a 2 (age group) × 7 (structure) design, There was no main effect of age, indicating no sta-
using percentage correct as the dependent variable. In tistically significant differences between the average scores
this analysis, each morpheme was treated individually, (collapsed across structures) of the younger and older group
and performance was contrasted across the morphemes (69.52% for the older vs. 61.35% for the younger).
There was a significant interaction effect between as these ranking order differences are not relevant to the
age group and structure, F(6, 48) = 2.47, effect size = .24, tense and nontense issue we intended to address, no fur-
p < .05. Inspection of the percentage scores for each struc- ther analysis was conducted to find out exactly which
ture separately for the younger and older group indicates pairs of structures were significantly different from each
that the same structures fell into the low, high, and me- other within each age group.
dium accuracy levels for both age groups. More specifi-
cally, for both groups, the structure with the lowest
accuracy was regular past tense (younger group: 47.03%;
Error Analysis
older group: 33.48%), followed by third person singular Participants made two types of errors. One type in-
(younger group: 51.48%; older group: 38.16%). The two cluded omission errors for which the required morpheme
easiest structures were progressive –ing (younger group: was dropped, and the other type included commission er-
78.83%; older group: 87.41%) and auxiliary DO (younger rors for which the required morpheme was substituted
group: 85.11%; older group: 82.63%). With medium level by another morpheme. Table 4 lists the averaged ratio of
accuracy were plural ( younger: 71.68%; older: 66.26%), omission errors out of the total number of errors for each
past tense irregular ( younger: 74.46%; older: 56.84%), data session. Evidently, for all six morphemes, omission
and copula BE (younger: 78.05%; older: 64.67%). The sig- errors constituted most of the errors, though the omission
nificant interaction stemmed from the fact that, within error rates varied dramatically. Four of the structures
each of the three performance levels (i.e., low, high, and (regular and irregular past tense, third person singular,
medium), the specific ranking orders of the structure and progressive –ing) had over 90% of omission errors.
varied between the younger and older groups. However, Similarly, the rate of omission errors for plurals was also
1292 Journal of Speech, Language, and Hearing Research • Vol. 50 • 1280–1299 • October 2007
Figure 6. The scatterplot of length of second language (L2) immersion and irregular past tense
morpheme production accuracy of the younger arrivals (top) and older arrivals (bottom).
very high, from 84.0% to 97.2% (M = 92.5, SD = 4.0; Jia, frequencies of correctly produced regular and irregular
2003). However, for copula BE and auxiliary DO, the verbs, and the incorrectly produced regular and irreg-
percentage of omission errors was lowered by almost ular verbs. For 8 of the 10 participants, the average fre-
30%. quency value for the correctly produced regular verbs
was significantly lower than the irregular verbs (see
Regular and Irregular Past Tense Table 5). The remaining 2 participants showed nonsignifi-
cant but similar trends.
For each participant, performance on regular and
irregular past tense forms was compared with paired
sample t tests (see Table 5). For 7 of the 10 participants,
performance on irregular verbs was significantly better Discussion
than on regular verbs. Two older arrivals and 1 younger The current study examined the acquisition of six
arrival showed the same though insignificant trend. English grammatical morphemes by 10 native Mandarin-
To examine how the performance on regular and ir- speaking children and adolescents over their first 5 years
regular past tense forms was related to word frequency, of residence in the United States. We charted and com-
for each data session for each participant, an average score pared the acquisition trajectories and levels of mastery
of frequency (U values) was obtained for the correctly reg- of the morphemes by each participant. By the end of
ularized and irregularized past tense verbs separately. the 5th year of English immersion, only one structure
Paired sample t tests were conducted to compare the (progressive aspect –ing) was mastered by all participants
Participant 3rd PT (r) PT (ir) DO BE Plur – ING Variable Model 1 Model 2 Model 3
Auxiliary DO
(defined as over 80% accuracy across three consecutive Linear 9.67*
testing sessions), and one structure (regular past tense –ed) Intercept 2.14*
was mastered by none of the participants. Performance
Copula BE
on other structures fell in between. In other words, all
Linear 10.35*
participants still spoke English with various degrees of Quadratic 4.70yy
morphological errors, after having lived in an English- Intercept 1.66*
speaking country for 5 years. None of them used regular
Present progressive –ING
past tense correctly over 80% of the time when it was re-
Linear 67.61***
quired, and most of them could not pass this criterion
Quadratic 44.98*** 1.94*
when using irregular past tense and third person sin- Intercept 1.87*
gular. In contrast, typically developing monolingual En- Slope 1.61yyy
glish speakers can produce these morphemes with at least
80% accuracy (over 90% for most morphemes) in sponta- Blank cells represent values of nonsignificant results.
neous productions by the age of 5 years (Rice & Wexler, y
p = .07. yyp = .06. yyyp = .05. *p < .05. **p < .01. ***p < .001.
1996; Rice, Wexler, & Hershberger, 1998). These results
reinforce previous findings of long-term attainment stud-
ies (e.g., Bialystok & Miller, 1999; Birdsong & Molis, structure, showed a clear upward trend that did not pla-
2001; Flege et al., 1999; Jia et al., 2002) or of longitudinal teau during the 5-year period. The most difficult struc-
studies (e.g., Hakuta, 1978; Lardiere, 1998; Paradis, ture, regular past tense, showed no significant growth
2005), indicating the difficulty of morphological acquisi- over time, with its acquisition trajectories characterized
tion by L2 learners. by tremendous fluctuations. However, visual inspections
The morphemes were mastered by drastically differ- of the graphs indicate that the younger arrivals might be
ent numbers of participants, and the acquisition trajectories at the beginning of a growth curve that will take more
of the morphemes also varied substantially. Progressive
aspect –ing, on average the easiest structure to acquire, Table 4. The averaged ratio of omissions errors over the total number
showed accelerated improvement in the beginning fol- of errors for each data session.
lowed by a period of leveling off. When findings of plural
morpheme acquisition from Jia (2003) with the same Variable M SD Range
participants are incorporated, plural was the second eas-
iest, with 7 out of 10 participants having mastered the Third person 92.80 12.45 50 –100
structure. The growth trajectories of plurals and –ing Regular past tense 95.58 9.12 55.56–100
also resembled each other. Auxiliary DO, copula BE, and Irregular past tense 92.30 12.76 50 –100
Auxiliary DO 62.60 36.73 16.67–100
irregular past tense presented medium level difficulty,
Copula BE 59.71 30.21 5.88–100
each mastered by 6, 5, and 4 participants, respectively.
Progressive aspect –ING 100.00 N/A N/A
Growth curve analysis showed a general pattern of accel-
erated growth in the beginning of the study followed by N/A = not applicable.
a plateau. Third person singular, the second hardest
1294 Journal of Speech, Language, and Hearing Research • Vol. 50 • 1280–1299 • October 2007
Table 5. Percentage correct and word frequency values for regular and irregular past tense.
y
p < .10. *p < .05. **p < .01. ***p < .001.
years to complete, whereas the older arrivals were on a First, the age-related differences took time to emerge
flat curve with no discernable advancing pattern. Thus, and only existed for the relatively difficult morphemes. It
only a longitudinal study of more than 5 years may be was not until the end of the study that younger AoAr pre-
long enough to capture the full growth trajectories of the dicted significantly better performance on third person
more difficult morphological structures. singular, and close to significantly better performance
The relative difficulty levels and acquisition order on regular past tense. That is, with time, early arrivals
of the morphemes indicated by these growth patterns gained an edge over late arrivals in acquiring the most
are generally consistent with past research with na- difficult morphological structures. This pattern resem-
tive Chinese speakers and speakers of other languages. bles that of phonological acquisition by native Mandarin
They basically conform to the order posited by Krashen’s (Jia et al., 2006) and Korean speakers (Flege, Birdsong,
(1977) natural order hypothesis and those findings from Bialystok, Mack, Sung, & Tsukada, 2005). For example,
long-term attainment studies (e.g., Flege et al., 1999; Jia Jia et al. (2006) found that an early arrival advantage
et al., 2002). Importantly, the current study uncovered the gradually emerged in L2 (English) vowel perception and
approximate growth trajectories that delineate the acqui- production after 3–5 years of English immersion. No-
sition rate and pattern for each morpheme. When a large tably, results from this study were obtained from a much
enough database like this is developed by future re- larger sample size (more than 200 participants) with a
search, progress in L2 morphological acquisition will be much wider AoAr range (from under 10 to over 40 years
able to be predicted on the basis of the specific morphemes of age) than those of the current study.
under consideration, a learner’s age at initial L2 expo- The Critical Period account and the Environmental
sure, his/ her native language background, and language account both predict a long-term younger learner ad-
environment. vantage. Both accounts postulate that younger learners
have some advantages over older learners: a richer L2
Age-Related Differences in L2 environment according to the Environmental account, or
Morphological Acquisition: When They an intact language learning mechanism (operated within
a sensitive time window) according to the Critical Period
Emerge During the Learning Process account. However, the nature of the younger learner ad-
The first goal of the current study was to identify vantage leads to different predictions of the timing of the
the time at which age-related differences occurred and to occurrence of the age effects. A richer L2 environment,
examine how much of the performance variance among par- the advantage of younger learners proposed by the Envi-
ticipants could be predicted by AoAr when language envi- ronmental account, is by nature cumulative. We do not
ronment was taken into account. By including a sample expect that greater L2 use by younger learners immedi-
with a wide range of AoAr, and systematically document- ately confers greater L2 proficiency. Therefore, the En-
ing their language environment over the entire 5-year vironmental account is fully consistent with findings of
period of L2 immersion, the current study obtained two the gradual emergence of a younger learner advantage
sets of findings that address the first goal of the study with increasing L2 acquisition experiences. In contrast,
and illuminate the nature of the AoAr effects. the Critical Period account is challenged to explain this
1296 Journal of Speech, Language, and Hearing Research • Vol. 50 • 1280–1299 • October 2007
tense, and third person singular; and around 60% for Taken together, the current study documented how
auxiliary DO and copula BE. This finding is consis- the acquisition of six English grammatical morphemes
tent with that from child L2 learners (Paradis & Crago, by 10 native Mandarin speakers varied across structures
2000). Importantly, the current findings indicate that and individuals. The morphemes were mastered by dra-
there were no age-related differences in the degree of the matically different numbers of participants and showed
dominance of omission errors. Wexler (1998) explained different growth trajectories. Variance in performance
the dominance of the omission errors as evidence for the variance across the participants was partially predicted
possession of functional categories by L2 learners who by AoAr in the United States, with early arrivals obtain-
only have problems connecting the functional categories ing greater proficiency in some morphological structures.
with the surface morphological structures. If we adopt However, the AoAr effects took several years to occur,
this view, our findings indicate that younger and older whereas, considerably earlier than that, early arrivals all
L2 learners all have the representations of the function- switched their preference to using L2 and were, thus, ex-
al categories, and they resort to the same mechanisms to posed to a significantly richer L2 environment than late
reach the production targets. arrivals. Further, when AoAr and its documented covar-
Finally, we examined whether age-related differences iate, L2 environment, were incorporated in the same
were manifested in the acquisition of regular versus growth curve analysis, language environment but not
irregular past tense forms. The D/P model (Ullman, 2001) AoAr was a significant predictor of individual differences.
predicts that older learners should learn irregular forms From a theoretical perspective, these findings illu-
better than regular forms by a large margin, and that such minate the nature of age-related differences in the L2
superiority may be smaller or even nonexistent among morphological acquisition mechanism. The study failed
younger learners. In the current study, inconsistent with to generate any evidence for the discontinuity of the lan-
these predictions, all participants performed better on the guage learning mechanism that is predicted by the Crit-
irregular than on the regular past tense, and such a dif- ical Period account. Specifically, no age-related differences
ference was statistically significant for 7 out of the 10 par- were found regarding the three specific aspects of L2 mor-
ticipants. On the other hand, as mentioned earlier, for phological acquisition (error types, regular vs. irregular
most of the participants, the correct irregular verbs all forms, tense- vs. non-tense-related). Further, the Critical
had a significantly higher frequency than regular verbs. Period account is challenged to offer an explanation for the
Frequency of verb past tenses had been shown to predict late emerging feature of the AoAr effects. Regarding the
the production accuracy of both typically and atypically Environmental account, there is strong supporting evi-
developing monolingual English-speaking children (on dence from the finding that when AoAr was used in the
average 8 years of age) (Marchman, 1997; Marchman, same analysis with its covariate, language environment,
Wulfeck, & Weismer, 1999). Pairs of irregular past and variations in language environment were a better predictor
present tense verbs also show larger phonological var- of performance. The finding of no age-related differences
iations than pairs of regular past and present tense verbs in the three specific aspects of morphology is consistent
(–ed). Therefore, if we assume a dual-mechanism model with the Environmental account postulating continuity
for morphological acquisition, the younger and older learn- in the language learning mechanism.
ers in the current study did not show a difference in the From an applied point of view, even in an L2 immer-
extent to which they relied on one mechanism or the sion setting, the acquisition of some grammatical mor-
other. Rather, a uniform set of factors, such as word fre- phemes by school-aged immigrant children takes several
quency and perceptual salience, seemed to influence all years to complete (if ever completed). Long-term systematic
participants. and targeted instruction is necessary for this population.
These findings of no significant age-related differ- As L2 learners exhibit some error types and difficulties
ences in the three specific aspects of acquisition fail to similar to monolingual children with SLI, these findings
generate any evidence for the Critical Period account that also call for caution in interpreting and using morpho-
posits discontinuity of the language learning mechanism. logical errors as indicators of speech, language, or learn-
Further, the findings that learners of all ages (a) pre- ing disorders in the immigrant children population.
dominantly made omissions errors, (b) found the same
morphemes easier or more difficult to learn, and (c) per-
formed better on irregular than regular verbs to the same Acknowledgments
degree lend support to a language learning mechanism This research was funded by Support of Continuous
that is continuous in nature. Though the current study Research Excellence/ National Institute of Child Health and
does not have a large enough sample size to detect all dif- Human Development Grant 41353-11-19/20/21 to the first
ferences in the aspects of learning studied, these findings author. We thank the following students for transcribing and
build a foundation for future research of a more extensive coding the data: Cynthia Hu, Siridatar Khalsar, Angela Wang,
nature. Kim Wang, and Michael Young. Gary Winkel guided us through
Flege, J. E., Yeni-Komshian, G. H., & Liu, S. (1999). Age Marchman, V. A., Wulfeck, B., & Weismer, S. E. (1999).
constraints on second-language acquisition. Journal of Morphological productivity in children with normal lan-
Memory and Language, 41, 78–104. guage and SLI: A study of the English past tense. Journal of
Speech, Language, and Hearing Research, 42, 206–219.
Francis, W., & Kuč era, H. (1982). Frequency analysis of
English usage: Lexicon and grammar. Boston: Houghton Marinova-Todd, S. H., Marshall, D. B., & Snow, C. E.
Mifflin. (2000). Three misconceptions about age and L2 learning.
TESOL Quarterly, 34, 9–34.
Hakuta, K. (1978). A report on the development of gram-
matical morphemes in a Japanese girl learning English as a McDonald, J. L. (2000). Grammaticality judgments in a
second language. In E. M. Hatch (Ed.), Second language second language: Influences of age of acquisition and native
acquisition: A book of readings (pp. 132–147). Rowley, MA: language. Applied Psycholinguistics, 21, 395–423.
Newbury House Publishers. Opitz, B., & Friederici, A. D. (2003). Interactions of the
Hall, W., Nagey, W., & Linn, R. (1984). Spoken words: Effects hippocampal system and the prefrontal cortex in learning
of situation and social group on oral word usage and language-like rules. NeuroImage, 19, 1730–1737.
frequency. Hillsdale, NJ: Erlbaum. Paradis, J. (2005). Grammatical morphology in children
Ionin, T., & Wexler, K. (2001). L1-Russian children learning learning English as a second language: Implications of
English: Tense and overgeneration of be. In X. Bonch- similarities with specific language impairment. Language,
Bruevich, W. J. Crawford, J. Hellermann, C. Higgins, & Speech, and Hearing Services in the Schools, 36, 172–187.
H. Nguyen (Eds.), The past, present, and future of second Paradis, J., & Crago, M. (2000). Tense and temporality: A
language research: Selected proceedings of the 2000 Second comparison between children learning a second language
Language Research Forum (pp. 76–94). Somerville, MA: and children with SLI. Journal of Speech, Language, and
Cascadilla Press. Hearing Research, 43, 834–847.
Jia, G. (2003). The acquisition of the English plural mor- Patkowski, M. S. (1990). Age and accent in a second lan-
pheme by native Mandarin Chinese-speaking children. Jour- guage: A reply to James Emil Flege. Applied Linguistics
nal of Speech, Language, and Hearing Research, 46, 1297–1311. 11, 73–89.
1298 Journal of Speech, Language, and Hearing Research • Vol. 50 • 1280–1299 • October 2007
Prévost, P., & White, L. (1999). Finiteness and variability in Snow, C. E., & Hoefnagel-Höhle, M. (1987). The critical
SLA: More evidence for missing surface inflection. In A. period for language acquisition: Evidence from second
Greenhill, H. Littlefield, & C. Tano (Eds.), Proceedings of the language learning. Child Development, 49, 1115–1128.
23rd Annual Boston University Conference on Language
Ullman, M. T. (2001). The neural basis of lexicon and gram-
Development (pp. 575–586). Somerville, MA: Cascadilla
mar in first and second language: The declarative/procedural
Press.
model. Bilingualism: Language and Cognition, 4, 105–122.
Rice, M. L., & Wexler, K. (1996). Toward tense as a clinical
Wexler, K. (1998). Very early parameter setting and the
marker of specific language impairment in English-speaking
unique checking constraint: A new explanation of the
Children. Journal of Speech and Hearing Research, 39,
optional infinitive stage. Lingua, 106, 23–79.
1239–1257.
Wolfson, N. (1979). The conversational historical present
Rice, M. L., Wexler, K., & Cleave, P. (1995). Specific
alternation. Language, 55, 168–182.
language impairment as a period of extended optional
infinitive. Journal of Speech and Hearing Research, 38, Yeni-Komshian, G., Flege, J. E., & Liu, S. (2000). Pronun-
850–863. ciation proficiency in the first and second languages of
Korean–English bilinguals. Bilingualism: Language and
Rice, M. L., Wexler, K., & Hershberger, S. (1998). Tense
Cognition, 3, 131–149.
over time: The longitudinal course of tense acquisition in
children with specific language impairment. Journal of Zobl, H., & Liceras, J. (1994). Functional categories and
Speech, Language, and Hearing Research, 41, 1412–1431. acquisition orders. Language Learning, 44, 159–180.
Singer, J. D. (1998). Using SAS PROC MIXED to fit multi-
level models, hierarchical models, and individual growth Received March 15, 2006
models. Journal of Educational and Behavioral Statistics,
24, 323–355. Revision received July 24, 2006
Singer, J. D., & Willett, J. B. (2003). Applied longitudinal Accepted January 24, 2007
data analysis: Modeling change and event occurrence. DOI: 10.1044 /1092-4388(2007/ 090 )
New York: Oxford University Press.
Contact author: Gisela Jia, Department of Psychology,
Singleton, D., & Ryan, L. (2004). Language acquisition: Lehman College, City University of New York, 250 Bedford
The age factor. Clevedon, United Kingdom: Multilingual Park Boulevard West, Bronx, NY 10468.
Matters. E-mail: giselajia@ yahoo.com.
Snow, C. E. (1983). Age differences in second language
acquisition: Research findings and folk psychology. In K. M.
Bailey, M. H. Long, & S. Peck (Eds.), Second language acqui-
sition studies (pp. 141–150). Rowley, MA: Newbury House.