Johnson Newport 1989 Cog Psy

COGNITIVE PSYCHOLOGY 21, 60%’ (1989)
Critical Period Effects in Second Language Learning:

The Influence of Maturational State on the Acquisition
of English as a Second Language
JACQUELINE %JOHNSON AND ELISSA L. NEWPORT
University of Illinois
Lenneberg (1967) hypothesized that language could be acquired only within a
critical period, extending from early infancy until puberty. In its basic form, the
critical period hypothesis need only have consequences for first language acqui-
sition. Nevertheless, it is essential to our understanding of the nature of the
hypothesized critical period to determine whether or not it extends as well to
second language acquisition. If so, it should be the case that young children are
better second language learners than adults and should consequently reach higher
levels of final proficiency in the second language. This prediction was tested by
comparing the English proficiency attained by 46 native Korean or Chinese speak-
ers who had arrived in the United States between the ages of 3 and 39, and who
had lived in the United States between 3 and 26 years by the time of testing. These
subjects were tested on a wide variety of structures of English grammar, using a
grammaticality judgment task. Both correlational and r-test analyses demon-
strated a clear and strong advantage for earlier arrivals over the later arrivals. Test
performance was linearly related to age of arrival up to puberty; after puberty,
performance was low but highly variable and unrelated to age of arrival. This age
effect was shown not to be an inadvertent result of differences in amount of
experience with English, motivation, self-consciousness, or American identifica-
tion. The effect also appeared on every grammatical structure tested, although the
structures varied markedly in the degree to which they were well mastered by
later learners. The results support the conclusion that a critical period for lan-
guage acquisition extends its effects to second language acquisition. 8 1989ACT-
demic Press, Inc.
In most behavioral domains, competence is expected to increase over

development, whether gradually or in stages. However, in some domains,
it has been suggested that competence does not monotonically increase
with development, but rather reaches its peak during a “critical period,“’
This research was supported in part by NIH Grant NS16878 to E. Newport and T.
Supalla, and by NIH Training Grant HD07205 to the University of Illinois. We are grateful
to Geoff Coulter, Henry Gleitman, and all of the members of our research group for dis-
cussion of the issues raised here, to Lloyd Humphreys for advice on statistical matters, to
Marcia Linebarger for the loan of test materials, and to Carol Dweck, John Plavell, Dedre
Gentner, Doug Medin, and two anonymous reviewers for helpful comments on earlier drafts
of this paper. Reprint requests should be sent to Jacqueline S. Johnson, Department of
Psychology, Cornell University, Uris Hall, Ithaca, NY 14850.
’ In this paper we use the term crirical period broadly, for the general phenomenon of
60
OOlO-0285/89$7.50
Copyright 0 1989 by Academic Press, Inc.
All rights of reproduction in any form reserved.
CRITICAL PERIOD 61
which may be relatively early in life, and then declines when this period
is over. For example, in the development of early visual abilities, the
development of attachment, or-in the case considered here-the acqui-
sition of language, it has been suggested that learners are best able to
achieve the skill in question during a maturationally limited period, early
in life. Elsewhere we have presented evidence that first language learning
is indeed limited in this way (Newport & Supalla, 1987). The present
paper focuses on the acquisition of a second language, asking whether this
type of learning, undertaken only after a native language is already ac-
quired, is nevertheless still maturationally constrained.
We will begin by reviewing prior evidence on this hypothesis, for both
first and second language learning, and will then present a new empirical
study which we believe shows evidence for a maturational function in
second language learning. Such evidence leaves open, however, whether
the underlying maturational change occurs in a specific language faculty,
or rather in more general cognitive abilities involved in language learning.
We will conclude by considering the types of mechanisms which are
consistent with our findings,
Evidence for a Critical Period Effect in First Language Acquisition

The critical period hypothesis, as advanced by Lenneberg (1967), holds
that language acquisition must occur before the onset of puberty in order
for language to develop fully. As will be detailed in the subsequent sec-
tion, Lenneberg’s hypothesis concerned only first language acquisition;
he left open the question of whether this critical period extended to sec-
ond language acquisition, which would occur after a first language was
already in place.
Lenneberg’s argument contained two parts. First, he reviewed avail-
able behavioral evidence suggesting that normal language learning oc-
curred primarily or exclusively within childhood. At the time his book
was written, no direct evidence for the hypothesis (from normal individ-
uals who had been deprived of exposure to a first language for varying
lengths of time in early life) was available. His review therefore included
changes over maturation in the ability to learn (in the case under consideration in this paper,
to learn language). We therefore include within this term maturational phenomena which
other investigators have called sensitive, rather than critical, periods. By using the term in
this broad fashion, we mean to avoid prejudging what the degree or quality of such matu-
rational change may be (e.g., is it a sharp qualitative change vs. a gradual quantitative one?)
and what the nature of the underlying maturational mechanism may be (e.g., is it a change
in a special language faculty vs. a more general change in cognitive abilities?). These further
questions will be addressed in part by the nature of our findings, and in part by future
research.
62 JOHNSON AND NEWPORT
various types of indirect evidence, for example, differences in recovery

from aphasia for children vs. adults, and differences in progress in lan-
guage acquisition, before vs. after puberty, in the mentally retarded.
Second, he proposed a mechanism which might be responsible for a
maturational change in learning abilities. The proposed mechanism was
fundamentally neurological in nature. He suggested that the brain, having
reached its adult values by puberty, has lost the plasticity and reorgani-
zational capacities necessary for acquiring language. Subsequent research
has questioned whether all of the neurological events he cited occur at an
appropriate time for them to serve as the basis for a critical period
(Krashen, 1975). Nevertheless, the hypothesis that there is such a critical
period for language learning has remained viable.
Since Lenneberg’s writing, behavioral studies approximating a direct
test of the critical period hypothesis for first language acquisition have
become available. One such study is a well-known case of Genie, a girl
who was deprived of language and social interaction until her discovery at
the age of thirteen (Curtiss, 1977). Her lack of linguistic competence,
particularly in syntax, after seven years of rehabilitation supports the
critical period hypothesis. However, the abnormal conditions under
which Genie was reared, including nutritional, cognitive, and social de-
privation, have led some investigators to question whether her language
difficulties have resulted only from lack of linguistic exposure during
early life.
More recently, Newport and Supalla (Newport, 1984; Newport & Su-
palla, 1987) have studied language acquisition in the congenitally deaf, a
population in which exposure to a first language may occur at varying
ages while other aspects of social and cognitive development remain nor-
mal. Their data come from congenitally deaf subjects for whom American
Sign Language (ASL) is the first language. However, since 90% of the
congenitally deaf have hearing (speaking) parents, only a few deaf indi-
viduals are exposed to this language from birth. The majority of deaf
people are exposed to ASL only when they enter residential school for the
deaf and first associate with other deaf individuals; this can be as early as
age four or as late as early adulthood.
Newport and Supalla separated subjects by their age of exposure into
three groups: native learners, who were exposed to ASL from birth by
their deaf parents; early learners, who were first exposed to ASL between
the ages of 4 and 6; and late learners, who were first exposed to ASL at
age 12 or later. Wishing to test asymptotic performance (i.e., ultimate
command of the language), they chose subjects who had at least 40 years
of experience with the language as their primary, everyday communica-
tion system. The subjects were tested on their production and compre-
hension of ASL verb morphology. The results show a linear decline in
CRITICAL PERIOD 63
performance with increasing age of exposure, on virtually every mor-

pheme tested. That is, native learners scored better than early learners,
who scored better than late learners, on both production and comprehen-
sion.
This study thus provides direct evidence that there is a decline over age
in the ability to acquire a first language. It also tells us, however, that
Lenneberg’s portrayal is at least partially incorrect in two regards. First,
the results show a continuous linear decline in ability, instead of a sudden
drop-off at puberty as his hypothesis implies. (This study does not tell us
whether the linear function asymptotes or continues to decline after pu-
berty, since separate groups of later learners, before vs. after puberty,
were not tested.) Second, it should be noted that, while the postpubescent
learners did not reach as high a level of proficiency as the native or early
learners, language had not become totally unlearnable for them. This
rules out any extreme interpretation of the critical period hypothesis.
In sum, current evidence supports the notion of a maturationally de-
limited critical period for first language acquisition, with some modifica-
tions from Lenneberg’s original formulation. However, this evidence is
compatible with a number of quite different accounts of the nature of the
underlying maturational change. Evidence concerning age effects on sec-
ond language learning can contribute to a further delineation of critical
period accounts.
Second Language Acquisition
What it can and cannot tell us about the critical period. Given the early
difficulties of performing a direct test of the critical period hypothesis on
first language acquisition, many researchers undertook studies of second
language acquisition over age as a test of the hypothesis. Some investi-
gators have suggested that a critical period theory must predict that chil-
dren are better than adults at learning second languages, as well as first
languages. Consequently, they have viewed any evidence to the contrary
as evidence against the critical period hypothesis (cf. Snow, 1983, for
discussion).
In our opinion, data on this issue do have an important consequence for
a critical period theory of language acquisition. However, it is not that the
critical period hypothesis could be rejected on such evidence, but rather
that it can be refined or clarified by such evidence. A critical period
theory for language acquisition would have quite a different character
depending upon whether second language acquisition were included in its
effects.
To capture this distinction there are two different ways we can state the
critical period hypothesis, one which does not include second language
acquisition in its effects and one that does:
Version One: The exercise hypothesis. Early in life, humans have a
superior capacity for acquiring languages. If the capacity is not exercised
during this time, it will disappear or decline with maturation. If the ca-
pacity is exercised, however, further language learning abilities will re-
main intact throughout life.
Version Two: The maturational state hypothesis. Early in life, humans
have a superior capacity for acquiring languages. This capacity disap-
pears or declines with maturation.
Notice that, although very different in character, the two versions make
the same predictions with regard to first language acquisition. They differ,
however, in their predictions for second language acquisition.
The exercise hypothesis predicts that children will be superior to adults
in acquiring a first language. By this account, if learners are not exposed
to a first language during childhood, they will be unable to acquire any
language fully at a later date. However, as long as they have acquired a
first language during childhood, the ability to acquire language will remain
intact and can be utilized at any age. On such a hypothesis, second lan-
guage learning should be equivalent in children and adults, or perhaps
even superior in adults due to their greater skills in their first language as
well as in many related domains.
This hypothesis is not unlike the conception of the visual critical period
described for cats (Hubel & Wiesel, 1963), where early visual experience
is required to maintain and refine the structure of the visual cortex, or the
conception of the critical period described for attachment in dogs (Scott,
1980), where early attachment to one dog is required for subsequently
normal socialization and permits unlimited later attachments to other
members of the same species. Indeed, as will be discussed below, some
of the current evidence on second language learning could be interpreted
to support an exercise hypothesis.
In contrast, the maturational state hypothesis claims that there is some-
thing special about the maturational state of the child’s brain which makes
children particularly adept at acquiring any language, first as well as
second. This hypothesis predicts that language learning abilities decline
with maturation, regardless of early linguistic experience: acquiring a first
language early in life will not guarantee the ability to acquire a second
language later in life. In this version, then, children will be better in
second language learning as well as first.
With certain qualifications, the critical period hypothesis that Lenne-
berg put forth can be subsumed under either version. In fact, it is not
absolutely clear which version he would have favored. Some comments
CRITICAL PERIOD 65
he made suggest that he thinks the young learner has a superior capacity
for acquiring second languages, and therefore that he would favor the
maturational state hypothesis:
. . . the incidence of “language learning blocks” rapidly increases after puberty.
Also automatic acquisition from mere exposure to a given language seems to dis-
appear after this age and foreign languages have to be taught and learned through
a conscious and labored effort. Foreign accents cannot be overcome easily after
puberty. (Lenneberg, 1%7, p. 176)
However, other comments within the same paragraph sound as if he
would have favored the exercise hypothesis:
. . . our ability to learn foreign languages tends to confuse the picture. Most indi-
viduals of average intelligence are able to learn a second language after the begin-
ning of their second decade . . . a person can learn to communicate in a foreign
language at the age of forty. This does not trouble our basic hypothesis on age
limitation because we may assume that the cerebral organization for language
learning as such has taken place during childhood, and since natural languages tend
to resemble one another in many fundamental aspects the matrix for language skills
is present. (Lenneberg, 1967, p. 176)
Since Lenneberg’s was one of the first proposals in this area, it is not
surprising that he did not take a definitive stand on this issue, particularly
since there were at that time few data to support either view. Neverthe-
less, it is a crucial distinction that should be made in any subsequent
account of a critical period.
Research on Age Effects on Second Language Acquisition

Is there an age-related limitation on the learning of a second language?
A number of studies have investigated this question since the time of
Lenneberg’s book, focusing particularly on the acquisition of phonology
and grammar., Superficially, these studies appear to contradict one an-
other; some have been said to demonstrate an adult advantage, some a
child advantage.
This apparent contradiction is resolved when one separates perfor-
mance in the early stages of learning from eventual attainment in the
language. (For a review of these studies, with a conclusion similar to the
one presented here, see Krashen, Long, & Scarcella, 1982.) Most of the
studies of second language learning have examined just the early stages of
learning; these studies tend to show an adult advantage in both phonology
(Asher & Price, 1967; Olson & Samuels, 1973; Snow & Hoefnagel-Hohle,
1977) and syntax (Snow & Hoefnagel-Hohle, 1978). Adults thus seem to
begin moving toward second language proficiency more quickly. How-
ever, this advantage appears to be short-lived.
In contrast, studies of eventual attainment in the language show a su-
periority for subjects who began learning in childhood, both in phonology

(Asher & Garcia, 1969; Seliger, Krashen, & Ladefoged, 1975; Oyama,
1976) and in syntax (Oyama, 1978; Patkowski, 1980). However, most of
the studies of child-adult differences in ultimate attainment have focused
on pronunciation. With anecdotal evidence that late learners do carry an
accent and experimental findings that support it, most investigators will
concede a child advantage for acquiring phonology (though not necessar-
ily a maturational one; see, for example, Snow & Hoefnagel-Hohle, 1977,
Olson & Samuels, 1973).
There is much less available evidence on child-adult differences in the
ultimate attainment of grammar. To our knowledge, only two studies have
been done. In both, the subjects were U.S. immigrants who were exposed
to English upon moving to the United States and who had lived in the
United States for at least five years at time of the test.
In one study, subjects’ syntactic ability was assessed by trained judges
who assigned syntactic ratings to written transcripts of the subjects’
speech from tape recorded interviews (Patkowski, 1980). For purposes of
analysis, subjects’ scores were divided along two variables: age of arrival
in the United States (before vs. after age fifteen), and years in the United
States (under vs. over 18 years). Additionally, measures of the subjects’
exposure to English in both natural and classroom settings were taken.
Using either the results from the analysis of variance test or correlations,
age of arrival was the only significant predictor of syntactic proficiency,
with the prepubescent learners outperforming the postpubescent learners.
The correlation of age of arrival with score was - .74, which indicates a
linear trend; however, the exact shape of the relationship cannot be de-
termined from the reported results.
In the second study mentioned, subjects were measured on their ability
to repeat spoken English sentences which had been masked with white
noise (Oyama, 1978). This task was meant to tap the ability to integrate
different sources of linguistic knowledge including phonology, syntax,
intonation, and redundancy patterns. Admittedly this is not a pure mea-
sure of syntactic ability; however, it presumably involves syntactic
knowledge (along with other factors). This study found the same pattern
of results just reported: age of arrival was the only significant predictor of
test performance.
In addition, the Oyama study addressed important claims regarding
whether children’s superiority over adults in final attainment is due to
factors other than maturation, which happen to be correlated with age.
For example, it has been argued that the adult is less motivated than the
child to learn the language fully, is more self-conscious about speaking
(i.e., practicing and making errors), does not have the cultural identi$ka-
tion with the host country necessary to become fluent, and in general is
CRITICAL PERIOD 67
less able to achieve the open attitudinal and affective state required for
language acquisition to take place (for reviews of this view, see Schu-
mann, 1975; Krashen, 1982). To test these claims, Oyama measured each
of these variables, plus other candidate predictors, using interview and
questionnaire material. Simple correlations showed a good association
between these variables and test score; however, partial correlations re-
moving the effects of age of arrival became essentially zero. In contrast,
when the reverse procedure was performed, removing each of these vari-
ables from the relationship between age of arrival and test score, the
partial correlation remained large and significant. In short, age of arrival,
rather than the attitudinal variables, predicted language performance.
These are important findings, for they support the view that age effects
are not simply an artifact of child-adult differences in affective conditions
of learning. However, a more rigorous test of this question could be
performed. Nonmaturational hypotheses do not typically propose that
one attitudinal variable, for example, self-consciousness, will alone pre-
dict performance; rather, they propose that the combination of all of these
variables favors children over adults. Thus a more stringent test would
involve partialling out all of the attitudinal variables together from age of
arrival, and then determining whether there is any predictive power left.
The study we present in the present paper is an attempt to supplement
the findings of these earlier studies. It is similar to the two studies dis-
cussed above, in that the focus is on ultimate command of the grammar of
the second language as a function of age of exposure to that language. It
differs from previous studies, however, in the way subjects’ proficiency
in the language is assessed and in the types of analyses performed. First,
a detailed evaluation of subjects’ knowledge of numerous aspects of En-
glish morphology and syntax is performed. This allows us to examine the
relationship between age of exposure and an overall measure of English
proficiency, as well as the possible differential effects of age of exposure
on various aspects of grammatical structure. Second, a wide range of ages
of exposure is examined, so that the precise shape of the function relating
age to proficiency can be determined. Third, multivariate analyses are
used to evaluate the relative contributions to proficiency of age as well as
a number of affective, sociological, and environmental conditions of
learning.
In detail, the primary questions that we address are as follows:
(1) Is there an age-related effect on learning the grammar of a second

language?
(2) If so, what is the nature of this relationship? What is the shape of
the function relating age to learning and ultimate performance, and where
(if anywhere) does the relationship plateau or decline?
(3) Can experimental or attitudinal variables, separately or together,

explain the effects obtained for age of learning?
(4) What areas of the grammar are the most and least problematic for
learners of different age groups?
In answering these questions we hope to gain a better understanding of
the nature of the critical period and, most particularly, to be able to decide
between the two versions of the critical period outlined above.
METHOD
Subjects
Subjects were 46 native Chinese or Korean speakers who learned En-
glish as a second language. Chinese and Korean were chosen as the native
languages because of their typological dissimilarity to English. (For con-
sideration of the effects of the first language on the second, see Discus-
sion.) No differences were found in the results for the two language
groups, so they will be presented together throughout the paper.
The primary criterion for selecting subjects was that they vary in the
age at which they moved to the United States and thereby first became
immersed in English. All subjects were exposed to English by native
speakers in the United States. In addition, to be sure that subjects had
sufftcient experience with English to be considered at their ultimate at-
tainment in the language, every attempt was made to obtain subjects who
had lived in the United States for many years. Minimum criteria were as
follows: all subjects had to have at least five years of exposure to English
and had to have lived in the United States for an unbroken stay of at least
three years prior to the time of test. Finally, to ensure ample exposure to
English and to ensure some homogeneity of social background, all sub-
jects were selected from the student and faculty population at an Amer-
ican university (University of Illinois). Subjects were recruited through
posted sign-up sheets, letters, and by word of mouth.
The resulting 46 subjects varied in age of arrival in the United States
from ages 3 to 39; throughout that range there was a fairly even distribu-
tion of ages of arrival. Age of arrival was considered the age of first
exposure to English. Three additional subjects were tested but eliminated
from data analysis when our posttest interview revealed that they did not
meet the above criteria: One did not have an unbroken stay in the United
States for three years prior to test; the second did not arrive in the United
States until adulthood but was immersed in English through attending an
all-English-speaking school in a foreign country. For both of these sub-
jects, then, age of immersion could not be determined unambiguously.
The third subject was eliminated because her early exposure to English
CRITICAL PERIOD 69
was from her Chinese parents, who had no prior experience with English
but nevertheless decided to speak only English in the home upon their
arrival in the United States. Most of her early exposure to English was
therefore not to standard English.
Additional experiential characteristics of the subjects varied for sub-
jects arriving in the United States early vs. late in life, and will be dis-
cussed separately for these two groups. In all cases, these experiential
characteristics, as well as age of arrival, will be evaluated for their rela-
tionship to performance in English.
Early arrivals. There were 23 subjects, 12 males and 11 females, who
had arrived in the United States before age 15. These early arrivals were,
at the time of test, for the most part freshman or sophomore undergrad-
uates who received money or class credit for their participation. All of
these subjects, from the time of arrival until college, lived in an environ-
ment where their native language was spoken in the home and English
spoken outside of the home. Once they entered college, all lived predom-
inantly in an English-speaking environment.
Late arrivals. The remaining 23 subjects were 17 males and 6 females
who had arrived in the United States after age 17. Prior to coming to the
United States, all of these subjects had had between 2 and 12 years of
mandatory formal English instruction in their native country. This raised
two possible concerns: One, the classroom experience might reduce the
effect of age of arrival on learning, since age of first exposure to English
for these subjects is earlier than age of arrival. Two, “age of learning”
may turn out to be better defined by age of starting classes rather than age
of arrival, which would result in a narrower range of ages than desired.
Whether point 2 is true is an interesting question itself and will be exam-
ined empirically in the results section.
At the time of test, these subjects were primarily professors, research
associates, and graduate students. All subjects, in both the early and late
arrivals groups, had at least some years of schooling while in the United
States. Within the late arrivals, the smallest number of years of school in
the United States was 3 years, the largest 10, with an average of 6 years
for the group.
For some of the subjects, the language environment was analogous to
that of the early arrivals, in which the native language was spoken in the
home and English spoken at school and work; for others, particularly
those that were unmarried, the language environment was almost all En-
glish. Thus in terms of exposure on a day to day basis, it does not appear
that the early arrivals have any advantage over the late arrivals.
In terms of years of exposure in the United States the late and early
arrivals also are fairly even. See Table 1. The average number of years in
the United States for early and late arrivals is 9.8 and 9.9, respectively.
TABLE 1
The Distribution of Early and Late Arrivals in Terms of the Number of Years They
Lived in the U.S.
Age of arrival
Years in the U.S. 3-15 17-39
3-6 4 7
7-10 10 11
11-15 9 3
23-26 0 2
The main difference between the two groups is that the late arrivals have
a larger range of years in the United States.
To provide a baseline performance on tests of English, 23 native speak-
ers of English were run. Two additional native subjects participated but
were not included in the analysis, one because the posttest interview
revealed that he acquired English outside of the United States, and one
because she spoke a nonstandard dialect of English.
Procedure
The subjects were tested on their knowledge of English syntax and
morphology by being asked to judge the grammaticality of spoken English
sentences of varying types (see Materials). While such a task, of course,
in principle requires metalinguistic skills in addition to knowledge of the
language, virtually perfect performance is shown on the same task by 6-
and 7-year old native speakers in subsequent studies (Johnson, Newport,
& Strauss, in press). This suggests that the metalinguistic skills necessary
for our task can only be minimally demanding for an adult and that any
variation obtained in performance on the task among adults must be due
to variation in knowledge of the language.
The test sentences were recorded on tape by a native American female
voice (E.N.). Each sentence was read twice, with a l-2 second pause
separating the repetitions. They were said clearly, with normal intonation
at a slow to moderate speed. The ungrammatical sentences were spoken
with the intonation pattern of the grammatical counterpart. There was a
3-4 second delay between the different sentences.
Subjects were tested individually in the laboratory. They were in-
structed to make a grammaticality judgment for each sentence, guessing
if they were not sure. It was made clear to the subject that if the sentence
was incomplete or otherwise wrong for any reason, they should regard it
as ungrammatical. The subject recorded yes/no responses on an answer
sheet by circling Y or N. To avoid giving cues to the subject, the exper-
CRITICAL PERIOD 71
imenter did not face the subject during the testing session while the tape
was going. Subjects were given a break halfway through the test, but were
told prior to starting that they should tell the experimenter to stop the tape
at any time if they need to break sooner, either if the tape was too fast for
them or if they were simply getting tired.
Following the grammaticality judgment test, subjects were interviewed
for approximately half an hour about their language background. Infor-
mation was gathered about the type and amount of exposure to English
they had, from when they were first learning the language until the time
of test. Motivational and additudinal measures were also taken, by having
the subjects rate themselves on a scale of 1 to 5 with regard to those
measures.
None of the subjects were blind as to the nature of the experiment.
They were told prior to participating that we were interested in determin-
ing whether children or adults are better at learning second languages;
they were not told, however, what type of results were expected.
Materials
The judgments of grammaticality test was modeled loosely after one
used by Linebarger, Schwartz, and Saffran (1983) in a study unrelated to
the present one. Our test, however, has a different set of English con-
structions and corresponding test sentences than those of Linebarger et
al., with the exception of two rule types which are noted.2
Our test was composed of 276 sentences.3 Of these, 140 were ungram-
matical. The other 136 formed the grammatical counterparts of these
sentences.4 The pairs that were formed, between the ungrammatical and
grammatical counterparts, were sentences that were exactly the same
except for one rule violation contained in the ungrammatical sentence.
The pairs of sentences were constructed to test 12 types of rules of En-
glish, listed in Table 2. The test contained between 6 and 16 pairs of
sentences testing each rule type. The members of a pair were, however,
not adjacent to each other, but rather were placed in opposite halves of
* We thank Marcia Linebarger for making these and other tests available to us.
3 An additional six sentences, three ungrammatical and three the grammatical counter-
parts of these, were included in the test but were eliminated from scoring because native
speakers of English made large numbers of errors in judging their grammaticality, due to
either auditory problems or dialect variations.
4 The numbers of ungrammatical and grammatical sentences are unequal because some
rule types have more than one grammatical sentence, or more than one ungrammatical
sentence, within each set of counterparts (see, for example, the section on particle move-
ment). For the most part, however, the grammatical and ungrammatical sentences form
pairs, and for ease of presentation they will be referred to as “pairs” throughout the paper.
TABLE 2
12 Rule Types Tested in Grammaticality Judgment Task
1. Past tense 7. Particle movement
2. Plural 8. Subcategorization
3. Third person singular 9. Auxiliaries
4. Present progressive 10. Yes/no questions
5. Determiners 11. Wh-questions
6. Pronomhlization 12. Word order
the test.’ Within each half, sentences were presented in random order (see
Design for further details.)
To ensure as much as possible that the sentences tested the rules under
study and not extraneous factors, sentences were constructed to contain
only relatively high frequency words, most of which were only one or two
syllables in length. The location of the grammatical error (at the begin-
ning, middle, or end of the sentence), the basic phrase structure of the
sentence, and the sentence length (ranging from 5 to 11 words per sen-
tence) were balanced across pairs of sentences testing each rule type, so
that each rule type was tested by a set of sentences comparable in all of
these regards.
The 12 rule types we tested were chosen to represent a wide variety of
the most basic aspects of English sentence structure. (Indeed, according
to our expectations, native speakers of English found the test very easy,
with ungrammatical sentences producing strong feelings of ungrammati-
cality.) Within the 12 rules types, there were four rule types which dealt
specifically with English morphology: past tense, plural, third person
singular, and present progressive. They will be discussed together since
many of the violations were constructed along similar lines. The other
eight types involved various rules of English syntax. Within each rule
type, the violations were formed on the basis of a few basic formats, with
several pairs of sentences (typically 4) using each format. These are dis-
cussed in more detail, with examples of the structure of the pairs, below.
Morphology: Past tense, plural, third person singular, and present pro-
gressive. For morphology, the grammatical sentence always contained
the target morpheme in a required context, while the grammatical viola-
tion was created using one of four formats:
(1) by omitting the required morpheme;
(2) by replacing the required morpheme with an inappropriate mor-
pheme from a different class;
(3) by making an irregular item regular;
(4) by attaching a regular marking to an already irregularly marked
item.
CRITICAL PERIOD 73
The first format was used to make ungrammatical sentences for all four
types of morphology. The sentence pairs were constructed so that the
grammatical context required the target morpheme, making it a grammat-
ical violation when the morpheme was omitted in one of the sentences of
the pair. For example, in sentences (la) and (lb), a plural marker is
required on the noun “pig,” and is present in (la) but is omitted in (lb).
In sentences (2a) and (2b), the present progressive ending is required on
the verb “speak”; it is present in (2a) but omitted in (2b).
(la) The farmer bought two pigs at the market.
*( 1b) The farmer bought two pig at the market.
(2a) The little boy is speaking to a policeman.
*(2b) The little boy is speak to a policeman.
Sentences were structured similarly for the other classes of morphemes.
The second format applied only to the verb morphology. One sentence
of the pair was correct; the other had an inappropriate tense marking for
the context. Consider, for example, sentences (3a) and (3b).
(3a) Yesterday the hunter shot a deer.
*(3b) Yesterday the hunter shoots a deer.
In (3a), the verb is in the past tense form as required, while in (3b) the verb
“shoot” occurs in present tense form in a past tense context.
The last two formats for creating the ill-formed sentences could be used
only for past tense and plural forms. An ill-formed sentence created by
making an irregular item regular is exemplified in sentence (4b), with its
grammatical counterpart in (4a). Similarly, the ungrammatical sentence
(5b) has a regular marking added on an already marked irregular.
(4a) A shoe salesman sees many feet throughout the day.
*(4b) A shoe salesman sees many foots throughout the day.
(5a) A bat flew into our attic last night.
*(5b) A bat flewed into our attic last night.
The test was constructed so that there was an equal number of sentence
pairs (4) in each format used for each type of morphology. However, due
to the nature of the morphemes, it was impossible for all of the formats to
be applied to all of the four rule types. Therefore the past tense and plural
are tested by more sentence pairs than are the third person or the present
progressive.
Determiners. To test subjects’ knowledge of determiners, the grammat-
ical member of the sentence pairs was constructed so that a determiner in
a particular position was either necessary or not allowed. The ungram-
matical counterparts were then formed by one of three methods: (1) by
omitting them in required contexts, as in sentence (6b); (2) by substituting
the indefinite for the definite, as in (7b); and (3) by inserting them where
neither article is allowed, see(8b). These examplescan be contrasted with
their grammatical counterparts (6a), (7a), and @a), respectively:
(6a) Tom is reading a book in the bathtub.
*(6b) Tom is reading book in the bathtub.
(7a) The boys are going to the zoo this Saturday.
*(7b) A boys are going to the zoo this Saturday.
@a) Larry went home after the party.
*(8b) Larry went the home after the party.
In many cases, there are other ways of construing the errors; for ex-
ample, (6b) may be construed as a plural error, instead of a determiner
error, for not having the plural marking on the noun “book.” In caseslike
these, where the error classification was ambiguous, the semantic con-
texts were created to try to bias the listener into the preferred reading.
For example, in (6) the reason Tom is in the bathtub is to sway the subject
into expecting that he is reading only one book rather than many.
Pronominalization. The sentence pairs for this rule type contain some
type of pronominal. The ungrammatical sentenceswere formed to include
one of the following violations: (1) the wrong case marking on the pro-
noun; (2) an error in gender or number agreement for the pronoun; or (3)
an erroneous form of the possessiveadjective.
The violations of case involved using nominative pronouns in objective
positions (see (9a) and (9b)), and objective pronouns in nominative posi-
tions:
(9a) Susan is making some cookies for us.
*(9b) Susan is making some cookies for we.
Gender and number were tested by capitalizing on the fact that reflex-
ive pronouns have to agree with the noun they are coindexed with. Sen-
tence (1Oa)is an example of correct gender agreement, while (lob) shows
a gender agreement violation:
(1Oa) The girl cut herself on a piece of glass.
*(lob) The girl cut himself on a piece of glass.
For possessiveadjectives, the error is in the form the word takes. So,
for example, some ungrammatical items have a possessiveadjective with
the possessive marker added, as in (llb). Compare this to the correct
form in (lla):
(1la) Carol is cooking dinner for her family.
*(l lb) Carol is cooking dinner for hers family.
Particle movement. With some minor changes, all of the items in this
CRITICAL PERIOD 75
rule type are from Linebarger et al. (1983). Here the sentences take ad-
vantage of the differences between particles and prepositions. The ill-
formed sentences were created by treating prepositions as particles, that
is, by moving the preposition to the right of the object NP as in (12b), as
compared to the correct form in (12a). These were contrasted with gram-
matical sentences with particles in their moved and unmoved positions, as
in (13a) and (13b). Additionally, other sentences were ill-formed by mov-
ing the particle outside its own clause as in (13~). Notice that, for this rule
type, the sets of counterpart sentences are not pairs but triples:
(12a) The man climbed up the ladder carefully.

*(12b) The man climbed the ladder up carefully.
(13a) Kevin called up Nancy for a date.
(12b) Kevin called Nancy up for a date.
*(13c) Kevin called Nancy for a date up.
Subcategorization. The items in this rule type are also from Linebarger
et al. (1983). These items test subjects’ knowledge of the subcategoriza-
tion frames of various verbs. In English, individual verbs determine the
type of syntactic frames that may follow them. For example, some verbs
require a direct object, while others require prepositional phrases. Be-
cause the details of these frames are lexically determined, ill-formed sen-
tences could be created by changing the structure of the required frame
for a particular verb while keeping the meaning intact. Thus, the change
in these sentences involved using the subcategorization frame of a seman-
tically similar verb. See, for example, the contrasts below.
(14a) The man allows his son to watch T.V.

*(14b) The man allows his son watch T.V.
(1Sa) The man lets his son watch T.V.
*(15b) The man lets his son to watch T.V.
The ungrammatical sentences were formed by exchanging the different

subcategorization frames of the two semantically similar verbs “allow”
and “let.”
Auxiliaries. In this rule type, the atlix requirements for different aux-
iliary verbs were tested. In particular, the ungrammatical sentences were
formed by violating three rules of auxiliaries. Each rule, with an example
of the correct and incorrect forms, is given below:
“Have” requires a past participle.

(16a) The baby bird has fallen from the oak tree.
*(16b) The baby bird has fall from the oak tree.
Following any form of “be,” the main verb must take the progressive.
(17a) Fred will be getting a raise next month.
*(17b) Fred will be get a raise next month.
Only the first element of Aux is tensed.
(Ma) Leonard should have written a letter to his mother.
*(18b) Leonard should has written a letter to his mother.
Yes/no questions. For this rule type, the ungrammatical sentencescon-
tain primarily errors in subject-aux inversion. The errors are of three
types. In one, two auxiliaries are moved in front of the subject, as in
(19b). In another, both the auxiliary and the verb are fronted (20b); and in
the third, the verb is fronted in a sentence where do-insertion would
normally occur, as in (21b). The grammatical counterparts are (19a),
(2Oa),and (21a), respectively.
(19a) Has the king been served his dinner?
*(19b) Has been the king served his dinner?
(20a) Can the little girl ride a bicycle?
*(2Ob) Can ride the little girl a bicycle?
(21a) Did Bill dance at the party last night?
*(21b) Danced Bill at the party last night?
Additionally, there were some ungrammatical sentencesformed by copy-
ing, instead of moving, the auxiliary verb, the difference being shown in
(22a) and (22b):
(22a) Can the boy drive a tractor?
*(22b) Can the boy can drive a tractor?
M-questions. The ungrammatical wh-questions have three forms, two
of them also dealing with aux. In one form, no subject-aux inversion
occurs, as in (23b) as compared with (23a); in the other, do-insertion is
omitted, as in (24b) compared to (24a):
(23a) When will Sam fix his car?
*(23b) When Sam will fix his car?
(24a) What do they sell at the comer store?
*(24b) What they sell at the comer store?
The third form of the ungrammatical wh-questions was lexical. A ques-
tion was ill-formed by substituting an incorrect wh-word for a correct one.
In sentence (25b), for example, “why” cannot be used unless the sub-
categorization frame of the verb “put” is satisfied by supplying a loca-
tive. Sentence (25a) satisfies this restriction by replacing the locative with
a locative wh-word.
CRITICAL PERIOD 77
(25a) Where did she put the book?

*(25b) Why did she put the book?
Word Order. In this last rule type, basic word order rules are tested.
Sentences of three types were used: intransitive (NP-V), transitive (NP-
V-NP), and dative (NP-V-NP-NP). Within each type, the ungrammatical
sentences were formed by systematically rearranging the verbs and noun
phrases so that all of the possible orders of constituents occurred. Thus
the simplest ill-formed sentence involves the reversal of an NP and in-
transitive verb, as in (26a) versus (26b); the most complex involves the
rearrangement of NPs and V in double-object structures, as in (27a) ver-
sus (27b).
(26a) The woman paints.
*(26b) Paints the woman.
(27a) Martha asked the policeman a question.
*(27b) Martha a question asked the policeman.
Design
The test was divided into two halves. An equal number of exemplars of
each rule type and subrule type were represented in each half. The gram-
matical and ungrammatical members of a pair were in opposite halves of
the test. Within each half, sentences were randomized in such a way that
no rule type was concentrated in one section of the test, and no run of
grammatical or ungrammatical sentences was longer than four.
Results
Age of Acquisition
Age of acquisition and ultimate performance. The primary question of
this study involved examining the relationship between age of learning
English as a second language and performance on the test of English
grammar. The results show a clear and strong relationship between age of
arrival in the United States and performance. Subjects who began acquir-
ing English in the United States at an earlier age obtained higher scores on
the test than those that began later, r = - .77, p < .Ol.
A more detailed understanding of this relationship can be gained from
Table 3 and Fig. 1. Subjects were grouped by age of arrival into categories
similar to those used in past research (e.g., Snow & Hoefnagel-Hohle,
1978). Table 3 presents the mean score, standard deviation, and the
ranges of the number of correct responses and the number of errors for
each group and for the native English comparison group. The means are
also presented graphically in Fig. 1. The adjacent age groups were com-
TABLE 3
Mean Scores of Nonnative and Native Speakers of English
Age of arrival
Natives 3-7 8-10 11-15 17-39
(n = 23) (n = 7) (n = 8) (n = 8) (n = 23)
Means 268.8 269.3 256.0 235.9 210.3
SD 2.9 2.8 6.0 13.6 22.8
Raw 275-265 272-264 263-247 251-212 254-163
(Errors) (l-11) (4-U (13-29) w-64) (22-l 13)
Note. Maximum score = 276.
pared, two at a time, by a set of two-sample t tests using separate variance

estimates.5
The first comparison involved determining whether there was any dif-
ference between the age 3-7 group and the native group in their perfor-
mance in English. The two groups were not significantly different from
each other, t(10.4) = 1.28,p > .05; indeed, the two groups were entirely
overlapping in performance. In contrast, all of the other age groups per-
formed significantly below the natives (for natives vs. the next closest
group @-lo), t(8.1) = 6.67, p < .Ol). This suggests that, if one is im-
mersed in a second language before the age of 7, one is able to achieve
native fluency in the language;6however, immersion even soon after that
age results in a decrement in ultimate performance.
Given that the 3-7 group is the only group that reached native perfor-
mance, it is perhaps not surprising that the difference between the means
of the 3-7 and 8-10 age groups is significant, t(l0) = 5.59, p > .Ol. As can
be seen in Table 3, while the absolute difference between the means of
these two groups is small, both groups have very small SDS,and the range
of scores for the 3-7 group is entirely nonoverlapping with the 8-10 group.
All of the later adjacent age groups are also significantly different from
each other. The age 8-10 group obtained higher scores than the 11-15
group, t(9.7) = 3.83, p < .Ol, with almost nonoverlapping distributions
between the two groups, and the age 11-15 group obtained higher scores
than the 17-39 (adult) group, t(21) = 3.78, p C .Ol.
In sum, there appears to be a strong linear relationship between age of
exposure to the languageand ultimate performance in that language, up to
5 Using a two-sample t statistic where the variance of each group is estimated separately
is appropriate whenever the population variances are not assumed to be equal, as is the case
here.
6 It is always possible,,however, that the equivalence in performance between natives and
the 3-7 group is due to a ceiling effect on our test, and that tests of more complex aspects
of English syntax would show differences even between these groups.
CRITICAL PERIOD 79
260 -
o 250-
b
ii 240 -
:
I” 230 -
220 -
210 -
/
I I I I I
Native 3-7 0-10 11-15 17-39
Age of Arrival
FIG. 1. The relationship between age of arrival in the United States and total score correct
on the test of English grammar.
adulthood. In the next section we examine the shape of this function in

more detail.
The effects of age of acquisition before vs. after puberty. An important
question to answer is whether, throughout adulthood, performance con-
tinues to decline as a function of age of exposure or whether it plateaus at
some point (H. Gleitman, personal communication). If the explanation for
late learners’ poorer performance relates to maturation, performance
should not continue to decline over age, for presumably there are not
many important maturational differences between, for example, the brain
of a 17-year old and the brain of a 27-year old. Instead, there should be a
consistent decline in performance over age for those exposed to the lan-
guage before puberty, but no systematic relationship to age of exposure,
and a leveling off of ultimate performance, among those exposed to the
language after puberty. This is precisely what was found.
Subjects were divided into two groups in terms of age of exposure, from
age 3-15 versus 17-39, with an equal number of subjects (N = 23) in each
group. The correlations between age of exposure and performance for
these two groups were strikingly different. For the group first exposed to
English between the ages of 3 and 15, the correlation was - .87, p < .Ol.
Note that this correlation is even more substantial than that for the sub-
jects as a whole. In contrast, for the group first exposed to English be-
tween the ages of 17 and 39, there is no significant correlation, r = - .16,
p > .OS.Scatterplots demonstrating this effect are presented in Figs. 2a
and b.
Age of acquisition and variance in ultimate performance. Another age-
related result, which is obvious from inspecting the scatterplots of Fig. 2
e. Subjects Arriving Ages 3-15 ,r= -.S7
2351 .
225 . l
.
215 I,, , , , ,
2 4 6 8 10 12 14 16
Age of Arrival
b. Subjects Arriving Ages 17-39, r=-.16

27tSk , , , , , , , , I , , p
.
250 -
240 - .
230 - l 8 . .
.
220 - .
B
210 - .
.
200-m ’ . -
l .
190 -
.
180 - . .
170-
.
160;
I I I I I I I I I I I ’
16 18 20 22 24 26 28 30 32 34 38 38 40
Age of Arrival
1 Note: The Y-axes are on different Scales.
FIG. 2. Scatterplots of test score in relation to age of arrival for subjects arriving in United
States before vs. after puberty.
and the SDS in Table 3, is the heterogeneous variance. For groups who
acquired English at early ages, the variance is very small; with increasing
age of exposure, variance gets larger, creating a megaphone shape, so that
for subjects exposed to English after 15 the variance is very large. Note
that it would have been quite possible to find that the means of these
groups increased but the variance stayed constant over the age groups.
The heterogeneity of variance obtained, and the relation between age of
acquisition and variance, is an independent result.
This heterogeneity of variance underscores two simple but important
points:
(1) Before age 15, and most particularly before age 10, there are very
CRITICAL PERIOD 81
few individual differences in ultimate ability to learn language within any

particular age group; success in learning is almost entirely predicted by
the age at which it begins.
(2) For adults, later age of acquisition determines that one will not
become native or near-native in a language; however, there are large
individual variations in ultimate ability in the language, within the lowered
range of performance.
Age of exposure to formal instruction. It has been assumed thus far that
age of arrival in the United States is the best measure of age of exposure
to the language. For early arrivals it is the only measure available, since
these subjects had no prior experience with English at all. However, for
the late arrivals there are two measures possible: age of arrival in the
United States, or age of beginning English instruction in school within the
native country. There is already a high correlation between age of arrival
and test score for the subjects as a whole, r = - .77; if age of classes is
a better measure of first exposure for the late arrivals, then the correlation
should be even higher when using that as a measure of time of exposure.
This is not what was found. The correlation for the subjects as a whole
between age of exposure, defined as classes or immersion (whichever
came first), and test score is - .67. These correlations, however, are not
statistically different from each other, t(43) = 1.26, p > .05. This is not
surprising since, due to the early arrivals, half of the measurements are
exactly the same; moreover, most of the late arrivals are defined as later
learners (pubescent or postpubescent) either way they are measured. Be-
cause of this overlap in measurement, the best way to evaluate the effect
of age of classes is to do so using only the subjects who had classroom
instruction. For these subjects alone (N = 23), the correlation between
age of classes and test score is - .33, which is not significant, p > .05.
This result has two implications. First, it means that we are using the
right measure for “age of exposure”; age of arrival in the United States,
with its resulting immersion in English, is more strongly related to ulti-
mate performance in English than is age of beginning formal English
instruction. More profoundly, it means that the learning which occurs in
the formal language classroom may be unlike the learning which occurs
during immersion, such that early instruction does not necessarily have
the advantage for ultimate performance that is held by early immersion. It
should be noted, however, that this last conclusion may be limited by the
relatively narrow age range for formal instruction found in our subjects:
our subjects all began their English classes between the ages of 7 and 16,
with most subjects beginning at ages 12-15. This conclusion may also be
restricted to the type of formal instruction received in Chinese and Ko-
rean schools (and, of course, any other schools in which the instruction is
similarly forma& and should be less true the more formal instruction
approximates immersion in the United States. In any event, age of arrival
in the United States appears to be the better measure of age of acquisition
for the population we studied.
Experiential and Attitudinal Variables
Experiential variables. Years of exposure in the United States was also
a variable of interest in this study. First, careful attention was paid to
balance the years of exposure between early and late learners. This was
done in order to avoid the possibility that obtained age effects would be
due to differences in years of exposure, rather than to true differences in
age of exposure. That we were successful in controlling for years of
exposure between the early and late learners is apparent from the lack
of correlation between age of arrival and years in the U.S., r = - .09,
p > .os.
Beyond controlling for this potential confound, it is also important to
ask what effect years of exposure has on learning, independent of the age
effects. It is known that number of years has some effect on subjects’
competence during the initial stages of learning a second language (see,
for example, Snow & Hoefnagel-Hohle, 1978). At an extreme, people
who have been in a host country for 1% years must perform better than
those who have only been there half a year. The question here is, how-
ever, do people continue to improve over time through continued expo-
sure to the language, or do they reach an asymptote after a certain number
of years? To answer this question, a correlation coefftcient was computed
between years of exposure in the United States and test performance. The
resulting correlation, r = .16, is not significant, p > .05 (see also Table 4).
This is in agreement with other studies (Oyama, 1978; Patkowski, 1980),
also showing no significant effect of the number of years of exposure on
languageperformance for learners beyond the first few years of exposure.
In addition to years of exposure, Table 4 also presents other variables
which we considered possible experiential correlates with ultimate per-
TABLE 4
Correlation Coeffkients of Experiential Variables with Score
Interview variable Correlation w/score
Length of exposure (years in the U.S.) .16
Amount of initial exposure (fEst year or two in U.S.) .03
Age of English classes” - .33
Years of English classes’ .25
Motivation to learn in classes” .05
LICorrelations for late learners only; measure not applicable to other subjects.
CRITICAL PERIOD 83
formance, such as amount of initial exposure to English, classroom ex-

perience, and attitude. Most of these variables were computed from in-
formation provided by the subjects; amount of initial exposure (measured
as the percentage of time English was used during the first year or two in
the United States) and motivation to learn in English classes (rated 1 to 5)
were estimates provided by the subjects. None of the correlations are
significant.
Regarding amount of initial exposure, the mean percentage for the
group is 51.4%, with a standard deviation of 20.2%. Unless subjects’
estimates are inaccurate, it appears that ultimate performance is not sen-
sitive to fairly large differences in amount of initial exposure to the lan-
guage, at least not after the subjects have been immersed in the language
for a number of years.
The classroom variables include the age at which the subjects began
English classes in their native country (already discussed in the previous
section), the number of years they took English classes, and their ratings
of how motivated they were to learn English in the classroom. Again,
none of these variables correlate significantly with performance. It may
be of interest for future research, however, that age of starting English
classes is the highest of the (nonsignificant) experiential correlations. This
may suggest some benefit of early classroom exposure, if classroom ex-
posure occurred earlier than in the population we studied, and particu-
larly if the classroom were more like immersion.
Attitudinal variables. Some investigators (see Schumann, 1975, and
Krashen, 1982, for reviews) have suggested that age effects are secondary
by-products of changes in people’s level of self-consciousness, in their
cultural identification, and in their motivation to learn a second language
well, rather than maturational changes in learning. To address this claim,
correlation and regression analyses were performed. Table 5 presents
correlations of such attitudinal variables with test score as well as with
age of arrival. These variables were measured by asking subjects to rate
themselves according to the questions presented at the bottom of Table 5.
The correlations show a strong relationship between these attitudinal
variables and both test score and age of arrival. Higher ratings of Amer-
ican identification and increased measures of motivation were associated
with better performance in English and with younger age of arrival, while
higher ratings of self-consciousness were associated with poorer perfor-
mance and with later age of arrival. Both of these sets of results would be
predicted by a theory which attempted to explain age differences in lan-
guage learning as a function of attitudinal variables correlated with age,
rather than a function of maturation. The other possibility is, of course,
the reverse; the attitudinal variables may have obtained their correlations
with test score as a result of the correlation with age of arrival. Thus it
TABLE 5
Correlation Coeftlcients of Attitudinal Variables with Test Score and Age of Arrival
Attitudinal variables Test score Age of arrival
Identitication .63** - .55**
Self-consciousness - .36* .19
Motivation .39** - .48**
* p < .05.
** p < .Ol.
Questions:
1. How strongly would you say you identify with the American culture? (subjects reply)
If 5 means you strongly identify with the American culture, that is, you feel like a complete
American, and 1 means not at all, how would you rate your identification?
2. Did you feel self-conscious while learning English in the United States? (Most often an
explanation was needed here). How would you rate that on a scale from 1 to 5, where 5 is
very self-conscious and 1 is not at all?
3. Motivation is a composite of two questions: (a) Is it important to you to be able to
speak English well? (subject’s reply) On a scale of 1 to 5, where 5 means very important and
1 means not at all, how would you rate it? (b) Do you plan on staying in the United States?
The composite was formed by adding one point to their importance rating if they planned on
staying in the United States, and by subtracting one point if they did not.
becomes a question as to which is the better measure: age of arrival or

attitudinal variables?
It is clear that age of arrival is the better measure over any of the
attitudinal variables considered alone. The correlation between age of
arrival and test score (r = - .77) surpasses the correlation between any of
the attitudinal variables and test score. Furthermore, the attitudinal vari-
ables are more adversely affected when age of arrival is partialled out
than is age of arrival when each of the attitudinal variables is partialled
out, as shown in Table 6. This is in complete agreement with Oyama’s
(1978) results.
As stated earlier, however, the most powerful evidence against this
alternative hypothesis is to show that age of arrival can account for vari-
TABLE 6
Partial Correlations of Age of Arrival and Attitudinal Variables with Test Score
Attitudinal variables Age of arrival

w/age of arrival removed w/attitudinal variables removed
Identification .39* - .65**
Self-consciousness - .34* - .76**
Motivation -.04 - .72**
* p < .05.
** p < .Ol.
CRITICAL PERIOD 85
ante not accounted for by the attitudinal variables combined. To test this,
a regression analysis was performed using the three attitudinal variables
together, which resulted in a regression coefficient of .47. This was com-
pared to the .69 regression coefficient obtained with the three attitudinal
variables plus age of arrival. The contribution made by age of arrival is
statistically significant F(1,41) = 28.1, p < .Ol. This shows that, inde-
pendent of any possible attitudinal effects, age of arrival has an effect on
learning a second language.
Of independent interest is whether the attitudinal variables can account
for any of the variance not accounted for by age of arrival. Even though
it is clear that age of exposure to a language is an important variable for
predicting ultimate performance, other variables may contribute to this as
well. Unlike previous studies (e.g., Oyama, 1978), we did find added
predictive value with two attitudinal variables: self-consciousness and
American identification. Each of the two makes a significant contribution
to a regression model including only age of arrival (F(1,43) = 5.6, p <
.05., for self-consciousness, and F(1,43) = 7.5, p < .05., for identifica-
tion), as well as a significant contribution to a regression model including
age of arrival and the other attitudinal variable (F( 1,42) = 5.0, p < .05, for
the addition of self-consciousness to age plus identification, and F(1,42)
= 6.9, p < .05., for the addition of identification to age plus self-
consciousness). Motivation, whether analyzed separately or in conjunc-
tion with the other two variables, failed to add significantly to the regres-
sion coefficient. Thus it appears at first glance that a model of second
language learning would have to include both age effects and the effects
of self-consciousness and identification, though not the effects of moti-
vation. Such a model might argue, for example, that while age of arrival
affects language learning, so does the self-consciousness and the cultural
identification of the learner.
At this time one might, however, be cautious about inferring a direct
causal link between self-consciousness and cultural identification to lan-
guage learning, until this result is corroborated in future studies. Not only
are the effects of self-consciousness and cultural identification not sup-
ported in other studies, but also possible mediating variables have not
been ruled out. For example, language performance may be correlated
with subjects’ evaluation of their performance, which may in turn affect
how self-conscious they are and how much they identify with the host
country. Thus poorer learners may, as a result of their performance prob-
lems, become more self-conscious and identify less with the United
States. In this account, greater self-consciousness and less identification
would be the result rather than the cause of the performance problems. In
any case, apart from whether attitudinal variables do or do not play a role,
there is a clear independent effect of age of arrival on ultimate perfor-
mance .
Age of Acquisition and Rule Type

The results show a striking effect of age of acquisition on performance
in our test of English syntax and morphology. It is of interest to know
what particular areas of the grammar create the most and least problems
for second language learners. Are the errors random, with an even dis-
persal across rule type, or do late learners err more frequently on a
particular type of rule? To answer this question, an analysis was per-
formed on age of learning in relation to the differing types of rules eval-
uated on the test. This analysis used only the ungrammatical items, since
it is only the ungrammatical items which can be said to be testing any
particular rule type. That is, when a subject marks a grammatical sen-
tence as ungrammatical, it is unclear what part of the sentence, or gram-
mar, (s)he is having problems with. In contrast, when a subject marks an
ungrammatical sentence as grammatical, (s)he must have failed to repre-
sentjust that structure under test as a native speaker would. For purposes
of this analysis, the age groups were the same as those used previously,
except that the late learner group was further divided into two groups,
(17-24) and (2%39), with an approximately equal number of subjects in
each. This was done to reach a more nearly equal number of subjects in
each of the age of learning groups. A two-way analysis of variance was
performed, using the 12rule types (outlined in the methods section above)
and six ages of acquisition.
The results of the anova showed a significant effect of rule type
F(11,693) = 53.2, p < .Ol, a significant effect of age of acquisition,
F(5,63) = 32.3, p < .Ol, and an interaction between rule type and age of
acquisition, F(55,693) = 8.3, p < .Ol. The age effect here is simply a
reproduction of the finding that early learners perform better than the late
learners; apparently there is no reduction of this effect when scoring only
the ungrammatical test items. The effect of rule type shows that subjects
made more errors on certain rule types than on others. Finally, the inter-
action appears mainly to be the result of late learners making proportion-
ately more errors on some rule types, and proportionately fewer on oth-
ers. Thus, many of the late learners’ errors do not appear to be random;
rather, there are particular parts of the grammar that seem more difficult.
The pattern of errors for each age group across the 12 rule types can be
seen in Fig. 3. In Fig. 3, rule types are ordered along the x-axis in de-
creasing order of difficulty for later learners.’ As can be seen, determiners
and plural morphology appear to be the most difficult for the two latest
groups of learners, with scores significantly worse than chance for deter-
miners (t = 3.35, p < .Ol), and no different from chance for plurals (t =
’ This ranking of rule type diiculty remains the same when using other criteria, for
example, ordering rule type according to the number of subjects who score almost perfectly
on that rule (that is, 0 or 1 item wrong, out of 6 to 16 possible, depending on the rule type).
CRITICAL PERIOD 87
70 I 1 I I I I I I I I I
$... - Native
so - ‘\. ‘.... --3-l -
-..- 8-10
----- 11-15
-.- 17-24
. . . . . .. ... 25-39
DET PLU SUB PST PRO PAR AUX 3P WHQ YNO ORD ING
Rule Type
FIG. 3. Mean percentage of errors on 12 types of English rules.
.16, p > .05). While all of the remaining rule types receive scores signif-
icantly better than chance (t ranges from 3.46 to 26.1, p < ,Ol), they vary
widely in level of performance. Most notably, basic word order rules and
the present progressive are giving very few problems, with most subjects
getting virtually all of the items of these types correct.’
Why are subjects performing better on some rule types and worse on
others? One uninteresting possibility is that the items testing some rule
types are inherently easier than those testing other rule types, since in
some cases different rules are tested by quite different sentential varia-
tions, On the other hand, it is clear that this is not the whole account of
our effects. Rule types tested in very similar ways on our test (e.g., the
various rule types involving morphology) did not show similar degrees of
diflkulty for late learners, suggesting that these rule type effects are not
due to the difficulty of the format by which we tested the rules.
A second possibility is that the subjects suffered from phonological
difficulties which made the items for that rule type difficult to process.
Again, although we cannot definitely eliminate this possibility, we do not
believe it is the whole account of the rule type effects. Rule types with
exactly the same phonological form (e.g., plural and third person singular,
both -s) did not show similar degrees of difficulty for late learners. Also,
rule types testing forms which were phonologically more substantial and
s Some other rule type scores also benefited from subjects’ apparent ease with basic word
order rules. For example, those items testing yes/no question formation by presenting ques-
tions in a V-N-N order (e.g., “Learns Jane math from Mr. Thompson?“) were particularly
easy for subjects. This pattern tit in with a general tendency for V-first items to be easily
judged ungrammatical.
therefore easier to hear (e.g., rule types with whole words reversed or
eliminated) were not necessarily easier for late learners than those that
involved smaller phonological units.
A third possibility is that subjects suffered from interference from the
nature of their first language (Chinese or Korean), and so should show
special difficulty with rule types most different from the first language.
Once again this did not appear to be the full account of our effects. Rule
types equally absent from Chinese and Korean (e.g., past tense and
present progressive) did not show similar degrees of difficulty for late
learners.
Most important, our rule type ordering corresponds in certain striking
ways to the order of difftculty obtained in studies of second language
learners from other first languagebackgrounds, as well as in studies of the
isolated girl Genie. In particular, the relative ease of word order and the
present progressive show up in all of these studies. We believe, then, that
the rule type effects we obtained are at least in part reflections of what is
generally difficult or easy for a late learner. We will return to this issue in
the Discussion section below.
One final question involved the relationship between age of arrival and
each of the individual rule types. Given that late learners’ competence
varies over rule types, it is of interest to know whether age of arrival
predicts performance on only certain selected rules of the second lan-
guage. The data show, however, that this is not the case. Table 7 presents
the correlations between age of arrival and the scores on each of the 12
TABLE 7
Correlation Coefftcients between Age of Arrival and Rule Type”
Rule type Correlations

Determiners .64**
Plural .75**
Subcategorization .53**
Past tense .79**
Pronouns .73**
Particles .44**
Auxiliaries .45**
Third person singular .29*
Wh-questions .39**
Yes/no questions .50**
Word order .48**
Present progressive .32*
0 These correlations, unlike others with age of arrival, are positive correlations, since they
relate age of arrival to number of errors.
* p < .os.
** p < .Ol.
CRITICAL PERIOD 89
rule types. Despite late learners’ proficiency on some rule types, all of the
rule types showed significant correlations with age of arrival. This result
shows that age of exposure to the language affects all of the structures we
examined, despite variations across rule types in the absolute level of
performance late learners achieved.
DISCUSSION
This study was designed to answer certain empirical questions about
critical period effects in second language learning, and thereby to clarify
and refine theoretical proposals regarding a critical period for language
acquisition more generally. We will begin our discussion by reviewing the
empirical findings, and then turn to the general theoretical issues.
The Basic Empirical Findings
Age of acquisition and ultimate performance. The first question we
asked was whether there was a relation between age of acquisition and
ultimate performance in the grammar of a second language. The results of
this study clearly show such a relation, and therefore support the notion
that children have an advantage over adults in acquiring a second lan-
guage. The overall correlation between age of arrival in the United States
and performance on our test of English grammar was - .77; and, for those
subjects arriving in the United States before puberty, this correlation was
- .87. Indeed, there was a significant correlation between age of arrival
and performance on every type of syntactic and morphological rule we
tested.
These findings are in accord with the results of the previous studies
which have tested asymptotic performance, despite the fact that these
studies used very different measures of English proficiency. (Oyama,
1978, measured number of words detected through white noise; Pat-
kowski, 1980, measured syntactic ratings of production). The present
study enhances the previous studies’ findings by providing a much more
detailed examination of English syntax and morphology. The three stud-
ies, however, complement each other well, for each emphasizes a differ-
ent aspect of language use. Oyama’s study, for example, taps some aspect
of on-line processing in comprehension, Patkowski’s measures free pro-
duction, and in our study we presume to be measuring underlying gram-
matical competence via sentence judgments. Because these studies com-
plement each other, the compatibility of the results is all the more im-
pressive. Together they provide a strong case for the conclusion that
children are indeed better than adults in their ultimate attainment in a
second language.
The effects of age of acquisition before vs. after puberty. The second
question we asked concerned the shape of the relationship between age of
acquisition and ultimate performance. Due to the large range of ages in

the learners we tested, and our division of the early learners into small age
groups, we are able to make a fairly good generalization about the shape
of this relationship. Subjects who arrived in the United States before the
age of seven reached native performance on the test. For arrivals after
that age, there was a linear decline in performance up through puberty.
Subjects who arrived in the United States after puberty performed on the
average much more poorly than those who arrived earlier. After puberty,
however, performance did not continue to decline with increasing age.
Instead, the late arrival group, while performing on the whole more
poorly than the younger arrivals, distinguished itself by having marked
individual differences in performance, something which was not found in
the earlier arrivals.
The pattern of this relationship supports a maturational account of the
age effects found. It does this by the fact that the age effect is present
during a time of ongoing biological and cognitive maturation and absent
after maturation is complete (i.e., at puberty). Thus it appears as if lan-
guage learning ability slowly declines as the human matures and plateaus
at a low level after puberty. The precise level of this plateau differs
between individuals.
Again, these findings are in line with previous studies, although no
previous study has asked this question in detail. Both Oyama (1978) and
Patkowski (1980) reported only overall correlations and grouped means,
with groupings which were larger and slightly different from our own and
with a more limited range of ages of acquisition than our own. Both
studies found the general linear decline of performance with age of ac-
quisition that we found, but the groupings of their subjects make it diffi-
cult to tell whether the precise ages at which we found changes in the
function are supported by their results as well. In addition, in a study of
age of acquisition in relation to first (rather than second) language acqui-
sition, Newport and Supalla (1987; Newport, 1984)found a linear decline
in ultimate performance over three age groups: subjects exposed to Amer-
ican Sign Language from birth vs. at age 4-6 vs. after age 12.9In short, the
surrounding literature on both first and second language acquisition ap-
pears to be generally consistent with the more detailed results obtained in
the present study.
9 One discrepancy between the Newport and Supalla results for fust language acquisition
and the present results for second language acquisition is in the level of performance at-
tained by subjects who began learning the language between the ages of 3 and 7. In the
Newport and Supalla data, the 4-6 age group performed consistently, although not always
significantly, below natives. In the present study, the 3-7 age group was entirely within
native performance. This difference will be discussed below, in the section entitled “The age
at which a decline in performance is first detected.”
CRITICAL PERIOD 91
Experiential and attitudinal variables. The third question we asked was

whether the effects of age of acquisition could be due to experiential or
attitudinal variables coincidentally related to age, rather than to matura-
tional changes in language learning. Our results suggest that entirely non-
maturational explanations for the age effects would be difficult to support.
Certainly the attitudinal variables (motivation, American identification,
and self-consciousness) were unable to explain away the age effects, in
accord with Oyama’s (1978) study. This held true in the present data even
when all three variables together were pitted against age.
It is also doubtful that the age effects are the result of differences in the
amount of English exposure between the younger and older arrivals. This
is true for several reasons. First, the younger arrivals did not ditfer sig-
nificantly, if at all, from the adult arrivals in the amount of English they
were exposed to during learning (see the Method section for description
of the subjects’ experiential characteristics). Second, the nonsignificant
correlation between amount of initial exposure and performance suggests
that second language learning is not particularly sensitive to differences in
the amount of exposure, at least when that exposure has occurred over a
number of years and is fairly high in the first place.
Some researchers have claimed that there are differences in the quality
of the exposure that adults and children receive, rather than in the mere
quantity, and that this difference may account for the age differences
found in language learning success. According to this view, children re-
ceive the ideal type of input for successful language learning, while adults
do not. Many have said, for example, that children receive “simple,”
reduced input which refers to concrete objects, existing in the here and
now. Adults, on the other hand, are exposed to syntactically more com-
plex input which most often refers to abstract concepts and events that
are displaced in space and time. The simple concrete input of the child is
thought to be helpful for language acquisition, while the complex input of
the adult is thought to interfere with language acquisition (Dulay, Burt, &
Krashen, 1982).
Both the empirical and theoretical assumptions underlying this ap-
proach have been disputed. First, the assumption that language is easier
to learn from limited simple input has been questioned (Wexler 62 Culi-
cover, 1980; Gleitman, Newport, & Gleitman, 1984). Second, the empir-
ical evidence for this claim has also been brought into question. Freed
(1980) performed a study which compared the type of input given to adult
and child second language learners and found that adults and children
actually receive comparable input in terms of syntactic complexity (as
measured by the number of clause boundaries). Interestingly, however,
the adult-directed input contained a more limited range of constructions
than the child-directed input. Adults received input which tended to main-
tain the canonical shape of a sentence, while children received sentences

with more deformations. Thus in terms of transformational complexity,
adults received the simpler input. From this it would be just as reasonable
to argue that adults learn less well because their input is not as complex
and varied as the child’s. In any case, the role of input in second language
learning needs to be better formulated before we can decide whether
children have any advantage in learning a language due to the type of
input they receive.
Age of acquisition and rule type. The fourth question we asked con-
cerned the nature of the effects of age on the attained grammar of the
second language. Our results suggestedthat, although there was an effect
of age of acquisition on every rule type we examined, some rules of
English grammar were more profoundly affected by age of acquisition
than others. In particular, knowledge of the basic word order of the lan-
guage was acquired by all of our subjects, regardless of their age of
learning. Similarly, knowledge of the present progressive (-ing) was ac-
quired by all of our subjects. These areas of competence likewise appear
in other studies of second language learning (see Krashen, 1982, for a
review of the order of morpheme difficulty in second language learning).
Perhaps even more striking, they are the only two aspects of English
which were successfully acquired by Genie, who was exposed to English
as a first language only after puberty (Curtiss, 1977; Goldin-Meadow,
1978). In contrast, other aspects of English syntax and morphology gave
late learners much more difficulty.
We believe that these rule type results are at least in part reflections of
universal factors in learnability, and not merely the result of item di&
culty or transfer from the first language. Newport, Gleitman, and Gleit-
man (1977; Gleitman, Newport, & Gleitman, 1984) and Goldin-Meadow
(1978) have suggestedthat basic word order is a highly resilient property
of languages, appearing in the acquisition of a first language under widely
varying conditions of both input and age of exposure. The present results
on the acquisition of a second languageunder varying conditions of age of
exposure are in accord with these claims. However, accounting for why
word order and -ing are particularly easy for learners remains for future
research.
Before turning to a more general discussion of critical period hypoth-
eses in language learning, we must consider whether the set of results we
have obtained will be replicable on other second language learning groups
or whether they are confined in any way to the particular second language
learners (Chinese and Koreans) we have studied.
Possible effects of thefirst language on second language learning. We
have thus far presented our results as though the findings were general-
izable to second language learning, regardless of the nature of the first
CRITICAL PERIOD 93
language or the relationship between the grammar of the first language

and that of the second language. Indeed we believe this is the case,
although we also recognize that certain aspects of the structure of one’s
first language are likely to have some effects on the learning of the second
language (see, for example, Zehler, 1982, and Hakuta & Cancino, 1977,
for a review of transfer effects in second language learning). Here we wish
to raise two points of relevance to the question of whether our results are
limited in any way to the Chinese and Korean speakers we studied.
First, we do not believe that the relationship found here between age of
exposure and ultimate performance in the second language is unique to
the circumstances where Chinese or Korean is the first language and
English is the second. We did purposely choose to concentrate on first
and second languages where the grammars were sufficiently different that
a significant second language learning problem would arise. Chinese and
Korean are relatively more isolating languages than English and have
syntaxes which are different in many ways from that of English. How-
ever, studies currently underway, as well as certain details of our present
results, suggest that the basic findings do not depend on these particular
language combinations.
Several studies in progress (Johnson & Newport, in press) examine
performance on our test by subjects with a wide variety of first languages.
It is too early to say from these data whether there is any effect of the
nature of the first language (we expect that there might be); however, it is
already clear that the strong correlation between age of arrival and test
performance replicates with subjects from these other first-language
backgrounds.
In addition, the detailed results of the present study suggest that the
nature of the first language cannot fully explain the difficulties of the
second language learner. The examination of performance on the 12 rule
types reveals relationships to age of arrival on every structure we exam-
ined, regardless of how similar or different these structures were to ones
in the first language. For example, determiners and plural inflection,
which gave late learners their most serious difficulties on our test, are
notably lacking in Chinese and Korean; but so are inflections for the
present progressive, on which late learners performed exceptionally well.
A more detailed understanding of which of our effects, if any, may arise
from first language characteristics should emerge from our studies in
progress.
Second, we do not believe that our results derive in any important way
from the input or cultural circumstances which characterize Chinese and
Korean speakers. The Chinese and Korean speakers we tested were per-
haps unusual, compared with many second language learners of English,
in that they often continued close associations with other speakers of their
first language. One might wonder, therefore, whether their exposure to

English or their maintenance of their first language somehow influenced
their second language learning. Again, this is an empirical question which
is best resolved by the results of our studies in progress, which include
many speakers isolated from their first language group as well as speakers
of first languages with large communities. Within the present study, all of
our subjects (both early and late learners) continued speaking their first
language with their families and others into adulthood, and all were ex-
posed to English from native English speakers. In addition, all had a
significant amount of exposure to English, since they were all active
members of an English-speaking community (that is, American schools
and universities). These factors therefore could not be responsible for the
differences we found between early and late learners of English. Whether
these factors have an additional effect on learning, beyond the effect of
age of exposure, was not the focus of our study, although some of our
results do bear on this question.
In sum, we believe that in other language groups the strong effects of
age of acquisition may be accompanied by effects of input, first language
typology, or other variables that do not appear in our data on Chinese and
Korean learners. Most importantly, however, we have reason to expect,
on the basis of our data, that these effects of age of acquisition will
persist.
Theoretical Conclusions for a Critical Period Hypothesis in
Language Acquisition
The present study was performed primarily for the purpose of under-
standing the nature of the critical period for language acquisition. In par-
ticular, we wanted first to discover whether the critical period occurs at
all in second language acquisition or whether it is exclusively a tirst-
language phenomenon. To delineate this distinction we began by present-
ing two possible versions of a critical period hypothesis. They are re-
peated here for convenience.
Version One: The exercise hypothesis. Early in life, humans have a
superior capacity for acquiring languages. If the capacity is not exercised
during this time, it will disappear or decline with maturation. If the ca-
pacity is exercised, however, further language learning abilities will re-
main intact throughout life.
Version Two: The maturational state hypothesis. Early in life, humans
have a superior capacity for acquiring languages. This capacity disap-
pears or declines with maturation.
To reiterate the differences between these two versions, the exercise
hypothesis only requires that a first language be acquired during child-
CRITICAL PERIOD 95
hood; as long as this occurs, the capacity for successful language learning
will remain intact. Thus it predicts no differences between child and adult
learners, due to maturation itself, in the ability to acquire a second lan-
guage to native proficiency. In contrast, the maturational state hypothesis
says that any language, be it first or second, must be acquired during
childhood in order for that language to develop fully. Our results support
the maturational state hypothesis, and not the exercise hypothesis. Hu-
man beings appear to have a special capacity for acquiring language in
childhood, regardless of whether the language is their first or second.
The maturational state hypothesis is, however, not itself an explanation
of critical period phenomena in language; rather, it merely outlines a class
of explanations which would be compatible with our results (namely,
those which posit maturational changes in general language learning abil-
ities). In order to approach a more precise theoretical account of the
phenomena, our study has also provided additional information which
should aid in understanding the nature of the critical period: namely,
information about the shape of the function relating age of acquisition and
ultimate performance. Our results provide three sets of facts which any
theory regarding critical periods would have to account for: the gradual
decline of performance, the age at which a decline in performance is first
detected, and the nature of adult performance.
The gradual decline of performance. Lenneberg’s original proposal of
a critical period in language acquisition seemed to predict a rectangular
function in the relationship between age of acquisition and ultimate per-
formance. That is, Lenneberg hypothesized that “normal” language
learning was possible during the period from infancy to puberty, with a
loss of abilities after puberty. However, the data on second language
learning in the present study did not have this shape. We did not find a flat
relationship between performance and age of learning throughout child-
hood, with a sudden drop in performance marking the end of the critical
period; instead, performance gradually declined from about age seven on,
until adulthood. Insofar as such data are available from other studies of
first and second language acquisition, the same linear trend seems to
appear (Oyama, 1978; Patkowski, 1980; Newport, 1984; Newport & Su-
palla, 1987).
Although this gradual decline is not in accord with Lenneberg’s implied
function, it is in accord with results from other behavioral domains in
which critical periods have been hypothesized. As research accumulates
on critical periods, whether it be in imprinting in ducks (Hess, 1973),
socialization in dogs (Scott, 1978), or song learning in birds (Kroodsma,
1981), it is becoming apparent that most, if not all, critical periods con-
form to the more gradual function. This point has recently been noted by
several investigators (Tanner, 1970; Immelman & Suomi, 1981).
. . . usually these periods consist of . . . beginning and end parts . . . [during]

which the organism is slightly sensitive to the spetitic influence, with a period of
maximum sensitivity in the middle. It is not as a rule an all-or-none phenomenon.
(Tanner, 1970, p. 131).
Whatever mechanisms underlie a critical period effect in language learn-

ing, then, must be compatible with this gradual decline of performance
over age.
The age at which a decline in performance is first detected. Lenne-
berg’s proposal also seemed to imply that a decline in performance should
first appear at puberty. Instead of puberty, we found a small but signiti-
cant decline in performance in subjects who had arrived in the United
States as early as age 8 to 10. Indeed, the only discrepancy we know of
between our results and other data is that, in first language acquisition,
this decline may occur even earlier (Newport, 1984; Newport & Supalla,
1987); in the Newport and Supalla data, a 4-6 age group scored consis-
tently, although not always significantly, below native performance. It is
possible that a similarly early decline may occur in second language ac-
quisition as well on a test that included more complex aspects of syntax
than our own; on our present test, the age 3-7 group scored at ceiling.
Further research is therefore necessary to determine with certainty the
exact point at which a decline in learning begins for second language
acquisition. It is clear from the present data, however, that this decline
begins well before puberty. It also appears that this early decline is small,
and that another more major change occurs around puberty. Proposed
mechanisms underlying a critical period effect in language learning must
therefore account for the details of timing of these changes and, particu-
larly, for the fact that the decline in learning ability begins earlier than
initially thought by most researchers.
The nature of adult performance. There are two aspects of adult per-
formance with which any theoretical account of the critical period must
be compatible. The first is that language does not become totally unleam-
able during adulthood. This has held true in all of the studies which have
tested age differences in asymptotic performance, including both first and
second language learning. In the present study, late learners scored sig-
nificantly above chance on all of the rule types tested except for deter-
miners and plurals. It appears to be the case, then, that quite a few
aspects of language are learnable to a fair degree at any age, even though
deficiencies in this learning occur.
The second aspect of adult performance with which any theory must be
compatible is the great variability found among individuals. For adult
learners, age does not continue to be a predictor of performance; thus any
proposed mechanism accounting for adult performance likewise cannot
be correlated with age. Moreover, while early learners are uniformly
CRITICAL PERIOD 97
successful in acquiring their language to a high degree of proficiency, later

learners show much greater individual variation (see also Patkowski,
1980, for related comments). A theoretical account of critical period ef-
fects in language learning must therefore consider whether the skills un-
derlying children’s uniformly superior performance are similar to those
used by adult learners, or rather whether adult language learning skill is
controlled by a different set of variables.
Final remarks on a critical period theory of language acquisition. In
sum, we now have a number of findings which should be accounted for in
any explanation of a critical period. There is the nature of the relationship
between age of arrival and performance: a linear decline in performance
up through puberty and a subsequent lack of linearity and great variability
after puberty. There is also the pattern of errors found for the wide range
of aspects of syntax and morphology of English studied: age effects were
found for every rule type, with low levels of performance on every rule
type except word order and present progressive. The primary and most
general finding to accommodate for any critical period theory, of course,
is that the critical period is not just a first language phenomenon, but
extends to a second language as well.
These findings rule out certain types of accounts of a critical period for
language acquisition and make other types of accounts more plausible.
We have suggested that our results are most naturally accommodated by
some type of maturational account, in which there is a gradual decline in
language learning skills over the period of on-going maturational growth
and a stabilization of language learning skills at a low but variable level of
performance at the final mature state. This leaves open, however, the
precise explanation of such a phenomenon. The traditional view of critical
period effects in language learning has been that there is maturational
change in a specific language acquisition device (Lenneberg, 1967; Chom-
sky, 1981). Such a view, with some modifications to incorporate the de-
tailed points of maturational change, is consistent with our results. Also
consistent with our results are views which hypothesize more general
cognitive changes over maturation (see, for example, Newport, 1984).
From this view, an increase in certain cognitive abilities may, paradoxi-
cally, make language learning more difficult. We are hopeful that future
research will provide more detailed results which may differentiate these
views from one another. In any event, the present study makes clear that
some type of critical period account for language acquisition is necessary
and that the proper account of a critical period will include both first and
second language in its effects.
REFERENCES
Asher, J., & Garcia, R. (1%9). The optimal age to learn a foreign language. Modern Lan-
guage Journal, 53, 334-341.
Asher, J., & Price, B. (1%7). The learning strategy of total physical response: Some age
differences. Child Development, 38, 1219-1227.
Chomsky, N. (1981). Lectures on government and binding. Dordrecht, Netherlands: Foris.
Curtiss, S. (1977) Genie: A psycholinguistic study of a modern day “wild child.” New York:
Academic Press.
Dulay, H., Burt, hf., Bi Krashen, S. (1982) Language two. New York: Oxford University
Press.
Freed, B. (1980). Talking to foreigners versus talking to children: Similarities and differ-
ences. In R. Scarcella and S. Krashen (Eds.), Research in second language acquisition.
Rowley, MA: Newbury House.
Goldin-Meadow, S. (1978). A study in human capacities. Science, 280, 649-651.
Gleitman, L. R., Newport, E. L., & Gleitman,H. (1984). The current status of the moth-
erese hypothesis. Journal of Child Language, 11.43-79.
Hakuta, K., & Cancino, H. (1977). Trends in second language acquisition research. Harvard
Educational Review, 47, 294-316.
Hess, E. H. (1973). Imprinting. New York: Van Nostrand.
Hubel, D., & Weisel, T. (1963). Receptive fields of cells in striate cortex of very young,
visually inexperienced kittens. Journal of Neurophysiology, 26,994-1002.
Immelmann, K., & Suomi, S. J. (1981). Sensitive phases in development. In K. Immelmann,
G. W. Barlow, L. Petrinovich, & M. Main (Eds.), Behavioral development: The
Bielefeld Znterdisciplinary Project. Cambridge: Cambridge University Press.
Krashen, S. (1975). The development of cerebral dominance and language learning: More
new evidence. In D. Dato (Ed.), Developmental psycholinguistics: Theory and appli-
cations: Georgetown Round Table on Language and Linguistics. Washington, DC:
Georgetown University.
Krashen, S. (1982). Accounting for child-adult differences in second language rate and
attainment. In S. Krashen, R. Scarcella, & M. Long (Eds.), Child-adult diSferences in
second language acquisition. Rowley, MA: Newbury House.
Krashen, S., Long, M., & Scarcella, R. (1982). Age, rate, and eventual attainment in second
language acquisition. In S. Krashen, R. Scarcella, & M. Long (Eds.), Child-adult dif-
ferences in second language acquisition. Rowley, MA: Newbury House.
Kroodsma, D. E. (1981). Ontogeny of bird song. In K. Immelmann, G. W. Barlow, L.
Petrinovich, & M. Main (Eds.), Behavioral development: The Bielefefd Znterdiscipli-
nary Project. Cambridge: Cambridge University Press.
Lenneberg, E. (1%7). Biological foundations of language. New York: Wiley.
Linebarger, M. C., Schwartz, M. F., & Salfran, E. M. (1983). Sensitivity to grammatical
structure in so-called a grammatic aphasics. Cognition, 13, 361-392.
Newport, E. L. (1984). Constraints on learning: Studies in the acquisition of American Sign
Language. Papers and Reports on Child Language Development, 23, l-22.
Newport, E. L., Gieitman, H., C Gleitman, L. R. (1977). Mother, I’d rather do it myselfz
Some effects and non-effects of maternal speech style. In C. E. Snow & C. A. Fergu-
son (Eds.), Talking to children: Language input and acquisition. Cambridge: Cam-
bridge University Press.
Newport, E. L., & Supalla, T. (1987)A criticalperiod effect in the acquisition of a primary
language. University of Illinois, manuscript under review.
Olson, L., & Samuels,S. (1973). The relationship between age and accuracy of foreign
language pronunciation. Journal of Educational Research, 66, 263-267.
Oyama, S. (1976). A sensitive period for the acquisition of a nonnative phonological system.
Journal of Psycholinguistic Research, 5, 261-285.
Oyama, S. (1978). The sensitive period and comprehension of speech. Working Papers on
Bilingualism, 16, 1-17.
CRITICAL PERIOD 99
Patkowski, M. (1980). The sensitive period for the acquisition of syntax in a second lan-
guage. Language Learning, 30, 449-472.
Schumann, J. (1975). Affective factors and the problem of age in second language acquisi-
tion. Language Learning, 2, 209-235.
Scott, J. P. (1978). Critical periods for the development of social behavior in dogs. In J. P.
Scott (Ed.), Critical periods. Stroudsburg, PA: Dowden, Hutchinson, & Ross.
Scott, J. P. (1980). The domestic dog: A case of multiple identities. In M. A. Roy (Ed.),
Species identity and attachment: A phylogentic evaluation. New York: Garland STPM
Press.
Seliger, H., Krashen, S., & Ladefoged, P. (1975). Maturational constraints in the acquisition
of a native-like accent in second language learning. Language Sciences, 36, 20-22.
Snow, C. (1983). Age differences in second language acquisition: Research findings and folk
psychology. In K. Bailey, M. Long, and S. Peck (Eds.), Second language acquisition
studies. Rowley MA: Newbury House.
Snow, C., & Hoefnagel-Hohle, M. (1977). Age differences in pronunciation of foreign
sounds. Language and Speech, 20, 357-365.
Snow, C., & Hoefnagel-Hohle, M. (1978). The critical period for language acquisition:
Evidence from second language learning. Child Development, 49, 1114-l 128.
Tanner, J. M. (1970). Physical growth. In P. H. Mussen (Ed.), Carmichael’s manual ofchild
psychology. New York: Wiley.
Wexler, K., & Culicover, P. (1980). Formal principles oflanguage acquisition. Cambridge,
MA: The MIT Press.
Zehler, A. M. (1982). The reflection offirst language-derived experience in second language
acquisition. Unpublished doctoral dissertation, University of Illinois.
(Accepted April 22, 1988)

Johnson Newport 1989 Cog Psy

Uploaded by

Copyright:

Available Formats

Johnson Newport 1989 Cog Psy

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Johnson Newport 1989 Cog Psy

Uploaded by

Copyright:

Available Formats

COGNITIVE PSYCHOLOGY 21, 60%’ (1989)

Critical Period Effects in Second Language Learning:

In most behavioral domains, competence is expected to increase over

Evidence for a Critical Period Effect in First Language Acquisition

various types of indirect evidence, for example, differences in recovery

performance with increasing age of exposure, on virtually every mor-

Second Language Acquisition

Research on Age Effects on Second Language Acquisition

periority for subjects who began learning in childhood, both in phonology

(1) Is there an age-related effect on learning the grammar of a second

(3) Can experimental or attitudinal variables, separately or together,

(12a) The man climbed up the ladder carefully.

(14a) The man allows his son to watch T.V.

The ungrammatical sentences were formed by exchanging the different

“Have” requires a past participle.

(25a) Where did she put the book?

Note. Maximum score = 276.

pared, two at a time, by a set of two-sample t tests using separate variance

adulthood. In the next section we examine the shape of this function in

e. Subjects Arriving Ages 3-15 ,r= -.S7

b. Subjects Arriving Ages 17-39, r=-.16

1 Note: The Y-axes are on different Scales.

few individual differences in ultimate ability to learn language within any

formance, such as amount of initial exposure to English, classroom ex-

becomes a question as to which is the better measure: age of arrival or

Attitudinal variables Age of arrival

Age of Acquisition and Rule Type

FIG. 3. Mean percentage of errors on 12 types of English rules.

Rule type Correlations

acquisition and ultimate performance. Due to the large range of ages in

Experiential and attitudinal variables. The third question we asked was

tain the canonical shape of a sentence, while children received sentences

language or the relationship between the grammar of the first language

first language. One might wonder, therefore, whether their exposure to

. . . usually these periods consist of . . . beginning and end parts . . . [during]

Whatever mechanisms underlie a critical period effect in language learn-

successful in acquiring their language to a high degree of proficiency, later

You might also like