J. Child. Lang.

II (1984), 43-79· Printed in Great Britain

The current status of the motherese hypothesis"

LILA R. GLEITMAN,
University of Pennsylvania

ELISSA L. NEWPORT
University of Illinois

AND

HENRY GLEITMAN
University of Pennsylvania

(Received I August 1982)

ABSTRACT

Partially conflicting results from correlational studies of maternal speech

style and its effects on child language learning motivate a comparative
discussion of Newport, Gleitrnan & Gleitman (1977) and Furrow,
Nelson & Benedict (1979), and a reanalysis of the original Newport et al.
data. In the current analysis the data are from two groups of children
equated for age, in response to the methodological questions raised by
Furrow et al.; but, in line with the original Newport et al. analysis,
linguistic differences between these age-equated children are handled- by
partial correlation. Under this new analysis the original results reported
by Newport et al. are reproduced. In addition, however, most effects of
the mother on the child's language growth are found to be restricted to
a very young age group. Moreover, the new analysis suggests that
increased complexity of maternal speech is positively correlated with
child language growth in this age range. The findings are discussed in
terms of a theoretical analysis of the Motherese Hypothesis; the

[.] This research was supported in part by a grant to L. R. Gleitman and B. Landau from
the National Foundation for the March of Dimes; by grants NS 16878 and BNS-80-13509
to E. L. Newport and T. Sup alia from the National Institutes of Health and the National
Science Foundation, respectively; and by grant HD 05951 to J. Campione and A. Brown
from the National Institutes of Health. We thank Adele Abrahamsen, Jon Baron,
D. Crystal, Barbara Landau, Richard Meier, Steve Pinker, Marilyn Shatz, Scott
Weinstein and an anonymous reviewer for helpful comments on an earlier draft of this
paper, and Jim Morgan, for computer expertise in analyses of the data. Address for
correspondence: Lila Gleitman, Department of Psychology, University of Pennsylvania,
3815 Walnut Street, Philadelphia, Pa, 19104, USA.

43
 CHILD LANGUAGE
THE MOTHERESE HYPOTHESIS

conditions of both learner and environment in which' simplified' data be sure, the style is often informal and hence some utterances consist for
could aid a learner. Finally, the results of our past work, those of Furrow example, of an isolated noun-phrase, rather than a whole subject' and
et al., and those of the present analysis, are discussed as they fit into, and ~redicate); and, in content, they generally concern things and events that are
add to, current theorizing about the language acquisition process. III View, as opposed to, say, talk about absent rabbits or past birthday parties.
The Motherese Hypothesis is the hypothesis that these SPECIAL properties
of caretaker speech playa causal role in acquisition. (See Snow (1972b), Cross
INTRODUCTION

It has almost never been disputed that language is acquired under direct
(1977), and Pinker (1979), for explication, and Newport (1977), NGG, and
Wexler ~ ~ulicover (1980) for criticisms of this general position.) The
!
I
influence of the learner's environment, since all children learn just the hypothesl~ IS.not that motherese is different in kind from ordinary talk among
language they hear. For the same reason, it has never been in dispute that adults. It IS Important to keep in mind that the sentences of motherese are
the way caretakers talk, and the circumstances under which they talk, affect most often regulation English sentences - in fact more often than they are in
learning. Since there is virtually no disagreement on these points, outsiders the hurried and elliptical speech among adults. What differs in motherese
may find it odd that there is fierce contention in the literature on the is t~e restriction on the choices among the allowable structures and contents.
effectiveness of maternal speech style - what we have called MOTHERESE (cf. j
In Its strongest form, the Motherese Hypothesis is that these restrictive
Newport 1977, Newport, Gleitman & Gleitman 1977) (henceforth, NGG) - in ~hoices are the requirement for learning. In a weaker form, the hypothesis ,I
guiding and organizing the learning. We have argued that the effects of IS that the more the caretaker restricts the sentence types and contents of the "~I
II.'
maternal speech are significantly modulated by biases of the child learner language in this fashion, the faster and less error-laden the learning will be.
about how to store and manipulate incoming information, and about the
allowable structures and contents of a language. To this extent, the character Measuring maternal speech and its influence on language learning
of the learning is not a straightforward function of the linguistic environment.
The most obvious way to study maternal speech and its influence on the
In contrast, many others, notably Furrow, Nelson & Benedict (1979) learner would appear to be correlational. This is because of the difficulties
(henceforth, FNB), have argued that the effects of the caretaker are consid- in manipulating the input speech, and aiso because of difficulties in gaining
erably broader and more straightforward than we have found. the co-operation of very young language learners in any situation where
In the present paper, after a brief introduction to the Motherese Hypothesis linguistically novel materials are presented." The idea is to measure the rate
as currently conceived, we will compare the methods and findings of NGG and character of learning by exploiting the normal range of variation among
and FNB, and present a new analysis of our past findings. This new analysis the ~aretakers. If there is some ideal input to a learner, those mothers who
was suggested by certain difficulties with our past work pointed out by FNB, use It the most consistently should have children who acquire the language
with which we agree. The new analytic procedures therefore mirror in many the more quickly and the least errorfully, on average. This method was used I
respects those performed by FNB with their own subjects. As we will show, to study syntactic aspects of maternal speech by both NGG and FNB. In
however, with these new procedures the original results reported by NGG detail, NGG measured various characteristics of maternal speech at a
re-emerge largely as before; in contrast, the findings and interpretations from particular time; measured a variety of the structures and contents produced :1
FNB are shown to be more difficult to use as a basis for theorizing about the
Motherese Hypothesis. Finally, after these methodological issues and the data
by their children, at two points in time; computed a growth score (language
'I
themselves have been presented, we will examine the logical problems for a [IJ This is n~t to say these latter methods cannot be attempted at all. Our group of investigators
Motherese Hypothesis and review the current status of our. knowledge of has examined language learning as it occurs in experiments performed by nature, that is,
in the pre.sence of pathologies that render the child's environment deviant in various ways.
language learning, in light of them. (For partial reviews, see L. R. Gleitman 198 I, Newport 1981, 1982.) In addition, we and
others have performed some traditionally experimental studies, which include examining
The Motherese Hypothesis the effects of maternal speech style on responsiveness (Shipley, Smith & Gleitman 1969)
and. comprehenSIOn of the child (Sachs & Truswell 1978, Snow 1972a, Newport &
It is easy to notice that mothers do not talk in the same way to very young
Glelt~an 1977), as well as examining the effects of experimentally controlled input in
children that they talk to, say, learned professors of linguistics (Newport adults learnt~g of artificial languages in the laboratory (Morgan & Newport 1981).
(1977), and see Snow & Ferguson (1977) for a number of descriptions of Training studies from K. E. Nelson (1976), Nelson, Carskaddon, & Bonvillian (1973),
~azden (1965), Feldman (1971), and Shatz & Hoff-Ginsburg (in prep.) have also been
motherese, and Hoff-Ginsburg & Shatz (1982) for a recent review). Maternal
mformatlve. None the less, there is some difficulty in achieving stable experimental results
utterances are very short, usually consisting of a single clause; they are clearly from young language learners owing to their difficulties with artificial materials and
enunciated; they almost never contain a true grammatical error (though, to perhaps their general disinclination to do as they are told. '

44 45
 THE MOTHERESE HYPOTHESIS
CHILD LANGUAGE

at Time, -language at Tirne.) for each measure of the child's speech; a.nd the smaller you are, the faster you will grow in the interval between
then correlated mother's speech with child language growth to see which measurements), the following outcome is sure to result: the MORE the mother
properties of maternal speech predicted which properties of the children's looked down at her child at the first measurement, the GREATER the growth
during the interval between measurements. We would surely not want to
language growth. ..
Such correlational studies face a number of problems. First, they typically conclude that this' Motherese looking' was' simple regard' and thus caused
measure maternal speech style by asking about the FREQUENCY with which t~e the child to grow. The same possibility, that the maternal speech style is an
various structures and contents appear in the maternal corpora. But there IS effect, rather than a cause, of language growth, plagues interpretation of
some reason to wonder whether sheer statistical preponderances in the d~ta correlational studies of mother/child language.
base are important factors in learning (notice, for example, that formal studies NGG handled this problem by statistically equating their child subjects
of learnability, e.g. Wexler & Culicover (1980), take appearance of some on each of the measures of child language, at the first interview. This was done
structure under interpretable conditions, not frequency of appearance of that by computing a correlation between mother at T'ime. and child change from
structure to be the sufficient condition for its acquisition). None the less, Time, to T'ime., but then partialling out the effects of the child's initial age
there is sorne plausibility to the idea that the child will seize upon those forms and language score on each of the measures taken (rXY. ab). This in effect
and contents that are the most characteristic of the data base. removes correlations between mother and child growth that are effects of the\
A more serious problem with the correlational method arises from the child on the mother (mothers use more motherese to younger children) and'
probable nonlinearity of the child's language learning curves. For language of the child on the child (language growth is faster in younger than in older
(as for many other domains of learning), learning rates may decelerate; that children)." Luckily, some variation in maternal usage remains, above and
is, the less you know the faster you change, and the more you know the slower beyond that attributable to the child's current age and language. And also
you change, as documented for certain measures of language growth by NG.G. luckily, some variation in child growth rate, on various measures, remains,
(It is important to note that ANY nonlinearity in growth cur-:es - ~cceleratlOn above and beyond that variation attributable to where the child fell at first
as well as deceleration - would yield the problem under discussIOn, .f~r the measurement on the language learning curves.
same or related reasons; the example we now work out in more detail IS t~e
case of deceleration.) If learning rates decelerate, a child measured early 10 [2] To be more precise, the partialling procedure removes from the correlation between
mother and child growth that portion due to the effects of the child on the mother and of
his development will show rapid growth, while a child first measured later the child on the child, IN so FAR AS THE LATTER ARE TAPPED BY OUR MEASURES, AND IN SO
in development will show less growth, all quite independent of what the FAR AS THE LATTER ARE THEMSELVES LINEAR RELATIONS. That is, most importantly, the
caretaker is doing. In addition, the caretaker may adjust her speech style to Pearson product moment partial correlation procedure assumes that all relations measured
are linear, and therefore in particular that the relations between the child's initial age or
what the learner knows at that particular developmental moment. As a
language scores and the child's growth from T'irne, to T'irnej , and between the mother's
consequence of these two facts, spurious correlations between mother and speech at T'ime, and the child's initial age and language scores, are linear. These
child may result, because of effects by the child on the mother, and effects assumptions regarding our data are not unreasonable. (Note that we previously suggested
that growth curves are NONLINEAR, but this translates into a LINEAR relation between initial
by the child on the child, rather than because of effects by the mother on the
state and rate of subsequent growth). However, although the assumptions are not
child. . unreasonable, they could be untrue. We have dealt with this possibility as follows:
The difficulty of disentangling these problems is easily shown by taking an (I) One may.in principle avoid assumptions of linearity by using non parametric, rather
analogy from physical growth curves, where no one suspe~ts that the than parametnc, correlational procedures. Unfortunately, at the time at which these
i~vestigations were conducted (although this may be changed in the future), no probability
caretaker's behaviour has much of an effect. For instance, we might measure distribution had ever been computed for Spearman rho partial correlations (or any other
the angle of regard from the caretaker when looking at her child. The smal~er r~nk.-order partial correlations), and there was therefore no way of evaluating their
the child, the smaller the angle of regard from caretaker to child: As the. child Significance. We therefore considered this an unsatisfactory alternative.
(2) w.e visually inspected scattergrams of the actual data points over which the partial
grows, the angle of regard increases. We would clearly be ~akmg a mistake
correlations were conducted in NGG (that is, scattergrams of the residuals). These
if we assumed that it was the mother's looking down behaviour that cau.sed appeared to be linear enough to justify some confidence in the assumptions.
the subsequent physical growth of the child, even though w.e sha~l obta1O.a . (3) Most important, if the assumptions are untrue, they largely work against our own
massive correlation between maternal angle of regard and child height. 'This interpretations and in favour of FNB's, rather than the reverse. That is, if the relations
are either partly or wholly nonlinear, the partialling procedure will remove less of the
particular problem can be resolved by the computation of a g~o~th ~core, contaminating variance than would be desirable. This in turn would lead to LARGER
correlating mother's angle of regard with the CHANGE 10 the child s height. ~partial) correlations between maternal speech and child language growth, a result more
However, there is a worse problem in interpreting the causal role of mime With the Motherese Hypothesis (and, as we understand them, the hypotheses
entertained by FNB) than with our own.
maternal regard. Owing to the nonlinearity of physical growth (on average,
47
46
 CHILD LANGUAGE
THE MOTHERESE HYPOTHESIS
I h t NGG created, by statistical manipulation, a set of child~en ~ho
n ~ or " measurement on the measures taken, T ey t en As should now be clear, the problem in the NGG study was not with partial
were Identical at the first f h h hl'ld on each such measure, correlations, The problem was that these were conducted over a relatively
f owth rate or eac sue c
computed measures 0 gr I t the first measurement time, broad age span, within which specific factors of motherese that affect the
d h ith the materna usages a
and correl~te t ese WI b k d : how did the differing maternal usages at learner might have changed, If this span is narrowed, the partialling does not
Th question could now e as e , T' T'
, have this possibly obscuring effect, On the contrary, the failure to partial out
T' e l affect the child growth rates during the interval from
Ime
(in NGG's study, a six month mterva , ,
, I) ,3 ,
Imel to ime,
h d d to the
(somehow) learning-curve differences among the children (i.e. under certain
circumstances, doing simple correlations) may obscure the chances of dis-
However FNB raised some plausible objections to, th~ met ~thawnhl'ch we
' , f d objections WI entangling cause and effect in the relationship between maternal speech and
child language development,
data on which thedco;~~t~~~s:e~~eo~e:n~r::I;sis implicitly assumed that
agree, They argue t, a I ntents and structures on the child were The Furrow, Nelson & Benedict study
CO
the effects of the varrous materna I t This is because the partialling
, d elopmenta momen s. Subjects and procedures of the FNB study
the same at varymg ev, "II th r than actually equates children
d f NGG which statrstica y ra e , ,
FNB accepted our initial arguments, that simple correlations between
proce ure 0 f , ,,'
nd linguistic a bili
I rties, m y
a be insensitive to effects of
over a range 0 ages a I ithin a particular developmental unequal children and their (therefore possibly also unequal) mothers might
mothers on children that occur bon y wlbll I'f the children studied were yield spurious correlations, But their remedy was different from ours, Rather
hi Id not e a pro em
period, Of course t IS wou d li isti abilities within which effects of than relying on partial correlations, they equated learners in a new way, They
within a narrow range of ages an mbgUls IC d to' be constant But the chose the reasonable stratagem of selecting children who were apparently
' h bly e presume ,
maternal speech rrug t reason a rs of a e and from identical at the first measurement, which seemed to preclude the need to
children NGG in fact studied ranged from one to two ~~a, g
partial out differences among them, They then argued that simple correlations
, , id bl advanced language abilities.
beginning t~ CO?SI era y,morle 'bl There is no reason to believe that the between the mothers' speech at 'Time, and the children's language abilities
FNB's objections are quite p ausi e. this
at Time, could now be interpreted just as our Own (partial) correlations with
aternal s eech on language learning must be the s,ame over
growth scores were interpreted: as effects of maternal usage on child progress,
e~ects of m p d lin uistic abilities, It is entirely possible that, s~y,
wide range of ages an g h I hen he knows the declarative FNB chose as their subjects children whose age was the same at the first
' good for t e earner w
imperatl~ebsen~e~cf~sr ~i~ if he does not, Correlations over such a ,range: even measurement (I; 6 years) and most of whose MLUs were also the same,
structure, ut a '11 ' k u that vanance m the namely I '0, There are two reasons why this choice of the youngest and least
if the children are statistically equated, WI , not ~IC p th of only sophisticated learners is reasonable, First, it is easier to make the claim that

::~~::;':r:~:hc:'~~:~ni~~::~~,
moment
t;h~::";~~"':::
to exploit
~~;:g~;~~
~:::lo;;:en"l
some particular characteristic of the environmen ,
. the children are' the same' by using Time Zero, This is because the various
linguistic structures and contents thereafter develop at quite differing rates
across children, making it virtually impossible to find subjects one can be at
all confident are' the same', The second, perhaps weaker, reason is that if
, '" nter reting such correlations between mother any learning group would seem to be the most central for the Motherese
b] Of course there are further difficulties ~n I h'ldP is the case with any interpretation of
and child as effects of the mother on t e c I ,as I' that we have eliminated one Hypothesis, it is that initial age group,
I' I data We only mean to c aim I' ,
causality from corre attO~a " F ther hindrances we ignore only by c arming FNB claimed to find a large number of effects of maternal simplification
known hindrance t~ such mte~pretattOns'l u: but by no means certain, argument for
that partial correlations contnbute to a p aus ble, ' inted t by FNB, on child language acquisition that NGG did not find, FNB argued that this
maternal effects on child growth, Similar provisos are pomte ou was because of NGG's failure to examine effects within narrow developmental
ranges, However, a number of problems of sample size and constitution make
[4] I is not in statistically equating children who are n~t
We must make clear that the prob em I t hildren to the extent that their the FNB findings less than definitive, One problem is the small sample of
' h ti Iling procedure on yequa es c d
actually equal, smce t ~par ia ith (i e influence) their mothers' speech styles an
ages and language ab ilities correlate w, ", b tween children that is uncorrelated utterances (100 per mother, and 100 per child) that formed the basis for the
h t Whatever vanance exists e "
their own growt ra es . , , ' 'lady whatever vanance exists analysis (in contrast, the data in NGG are based on an average of 5 I 3
h I d growth rate remams,
with maternal speec stye an , I t d ith differences between c hild
sirm , I ren utterances for each mother and on as many child utterances as were produced
I h tyles that IS uncorre a e wI'
between materna speec s 'II' procedure is done a corre atton
' h bl 'that after the partra mg , , b in an hour's session, or up to about 300), In addition, the number of dyads
remains. T e pro em IS .' " Id not pick up relationships etween
b
computed etween t h ese r emammg vanances d wou th by just one sub range 0 f c hild I ren, studied by FNB was small. There were 6 dyads for which the child's MLU
maternal speech and accelerated or decelerate grow
was I '0 and the age of the child I; 6 (as opposed to 15 dyads studied by NGG,
48
49
i:'
 THE MOTHERESE HYPOTHESIS
CHILD LANGUAGE

b I ) FNB also included in their analyses data provided by a seventh comprehension-score differences among FNB's subjects suggest that MLU
:~~'e:to:f 'the same age, but who had an MLU of roughly I'~ at the first isnot a sufficiently sensitive measure of the current language status of one-word
me~surement introducing a serious contaminant into the obtamed correla- speakers. Therefore, it is an inadequate measure on which to equalize one-word
speakers."
tions Since ~his outlier contributes a large proportion of thed.~tall~~d
In sum, various difficulties of FNB's study provide some impetus for
data ~ 1_ and since the sample size was very small, there ~re some Ih cu lel~
. 7. these findings. Another difficulty stemmmg from t. e sma further replication. But more than these limitations, it is the very great
in evalu~tm~ h fi din from FNB that most of the vanance on plausibility of the FNB objections to our prior study, and their sophisticated
sample SIZe IS the fu~t er n g was due almost entirely to two of ,work with seven new children, that cries out for further replication. As we
certain of their most Important measures will now show, certain differences between the results achieved by these
the seven subjects (ibid.: 431). f havin small samples and performing authors and our own, and differences in the interpretation of these findings,
The most senous con~equ~nce
large numbers of correlatlOn~s
of measurement error. NG.
t: 0

I
bt:ined correlations may be a result
~~~:ed worried about the reliabil.ity of
f nd performed a spht-half
provide yet another impetus for further replication and analysis.

FNB's findings and interpretations \

. I fi di for their samp e 0 IS, a . h
correlatlOna n ings . . ( b I ) Those correlations whjc Table 1 presents the simple correlations obtained by FNB (ibid.: 433). Table \
I · th r findmgs see e ow .
correlational ana YSlSon el. f h d t as well as on the overall 2 presents the original findings from NGG. It should be obvious that the
. . h I n spht-halves 0 tea a, .
maintain t emse ves 0 I' bl the subJ'ects studied. (There IS tabulated findings are not directly comparable. As can be seen, not all the
. I t be taken as re ia e on .
analysis, can at eas . . Tt f the findings on yet further subject measures are the same, and as we have stated the NGG findings are from
of course still a question of the reliabi I Ylvsi f the internal reliability of their learners who range over a much larger developmental period (but whose age
groups.) In contrast, FNB present no ana YSlS0 and initial language level have been statistically equated by the double
findings. . f hat FNB's subjects were not really' the partialling procedure). But some results are much the same even so, as a
Most important of all is the act t d I t To be sure their subjects comparative inspection of the tables shows.
. . . . II I f language eve opmen . ,
same' in their mrtra eve 0 b FNB themselves point out On the other hand, FNB obtained many significant correlations that we did
were mostly equal in pr~ducti~e languageec:t~i:: language development, on not, as is also clear from a comparison of these tables. Many of their
tage f
(ibid.: 435), they vaned in their sh °h r (PBenedict 1976). The trouble is correlations are quite puzzling. For example (Table I), they found that
d b one of t e aut ors
measures deve Iope ~ . Ian uage were predictive of the mothers who used more copulas and more contractions to their offspring had
g
that these differences m level of brecehPtlve d measurement (a nine-month children who came to say fewer noun-phrases per utterance than the offspring
., f d t ve growth y t e secon
child s rate 0 pro uc I . ) As the authors again note, of mothers who used fewer copulas or contractions. If such an effect is real,
· h d measurement was at 2,3 .
'1 interva I , i.e. t e secon h h bt ined simple correlations between what could be its explanation? Contractions have to do with auxiliary verb
ibility t att eo ar .
this leaves open t h e pOSSI I I h be due to the relationships
maternal speech and .child langu;gehgro~~d,:n::mprehension abilities. Thus [6] More precisely, the investigators measured comprehension when the subjects were aged
of both to a third vanable, name y t e c e ual at first measurement, so that I; 5, and then again when they were I; 7. The body of the measures used for the correlations
were collected at I; 6. The children were alike in comprehension (as well as MLU) at I; 5,
FNB's attempt to find subJec~s who weI r~ q that are interpretable without but differed from each other at I;7. Therefore a claim that comprehension differences did
btai ther/chlld corre atlOns
one could 0 tam mo h ssful.! In particular, the not affect the results at I; 6 is plausible; but just as plausible is the possibility that they
statistical manipulation, was not altoget er succe . did affect them. We ourselves used production MLU (and its various subcomponents) as
measures of the language status of our subjects. However, since there was indeed variation
relevant new study of mothers' speech and child among our subjects in MLU, it is likely that MLU correlated with other indices (e.g.
[5] After the time this paper went to press, a G f d Satterly & Wells 1983)' Because of comprehension), and thus that when we partialled out MLU or its subcomponents, we
reun
language growth was published (Barnehs,. ut d d'eta'II However one brief comment were also partialling out other aspects of language status, In contrast, when MLU does
bl dd ess t IS stu Y 1O· , T'
the timing, we are una e to a r I ted to equate their child subjects at ime, NOT vary among subjects (as was the case for FNB's one-word speakers), one should be
is in order here. Like FNB, Bar~es et ~d' attem P. levant ways. (In their case, they chose . less confident that it alone is a sufficient measure of language status, and therefore that
hild here 10 fact I entica
l 10 re ild' MLU
by selecting c I ren wow ibl t the time when the chi s its equality across subjects reflects relevant equality for correlational purposes. It has been
speech samples which were all as ~ose as ;o~s'di: n~t succeed in this aim. They report shown (Braine 1976, Bloom 1973, and other sources) that there are major differences in
equalled 1'5.) However, like FNB, arnes ea. . and moreover varied in age' current language among children, during the relatively lengthy one-word period, The
in f ' din MLU from 1'0 to 221, I
that their subjects 10 act vane 'I h FNB's results then the Barnes et a, differential comprehension progress of the FNB subjects during the period I; 5-1; 7 may
from I; 6 to 2; 9, Perhaps even more senolus y t ~n nd child langu~ge g:owth may in fact. very well be a reflex of this underlying difference - that is, where the subjects were, within
hit' s between materna speec a the one-word stage, at I; 5 and I; 6.
results on t e re awn, f b h t the child's initial language and age.
be due to the relatIOnships 0 ot 0
51
SO
 THE MOTHERESE HYPOTHESIS
CHILD LANGUAGE

structure; why should they affect the growth of noun-phrases but not of
;mple correlations between maternal speech at I ;6 and child auxiliaries? Similarly, why should copulas affect the growth of noun-phrases?
TABLE 1. S• a
speech at 2 ; 3 As another example of the same point, they found that the greater the
Child language proportion of interjections (e.g. Mm-hmm) from mothers, the more verbs
their offspring used in their utterances.
Vb/utt NP/utt
Aux/VP MLU FNB did make an attempt to interpret their findings. This is in line with
Maternal speech
-0"22 a serious attitude towards explaining the learning effects. To say merely that
-0'25 -0'28
-0'03 0'58 whatever the' child seems to be influenced by was the' simple input' would
Declaratives 0'72" 0'64
0'85" 0'34
I Yes/no-questions 0'06 0'02 be to beg the questions that are at issue. Rather, the authors proceeded by
I -0'47
I Imperatives
Wh-questions
-0'30 -0'37 -0'33
-0'70"
-0'48
-0·68"
the sensible means of seeking external support (apart from the correlational
findings) for the view that these effective properties of maternal speech are
I Words(MLU)
-0'38 -0·69"
-0'53 -0·60 -0'46
the linguistically or experientially simple ones. But in our view their attempts
-0'55
S-nodes/utt 0'43
I Interjections
0·64 0'57 0'67"
-0·62
were not always successfuL We turn now to the specific effects reported by \,
-0'75" -0·81"" FNB (Table I), and their interpretations.
\ -0'58 0'55
Pronouns 0'72" 0'74" Sentence type. To begin with, FNB found that a preponderance of maternal
0·60 -0'55
Noun/pronoun ratio -0'71" -0'78"
\ -0·66 -0'77"" Y1s/no-questions clearly speeds the acquisition of verbal auxiliaries by the
Verbs -0'85"" -0'90""
I Copulas
-0'58
-0"46 -0'47 -0·63 learner. Nee had reported the same effect. Thus both Tables 1 and 2 reveal
I Tense
-0'09
-0·65 -0'58 -0'84"
highly significant correlations between this feature of the mother's speech and
I Contractions
-0"21

subsequent child growth. As we will discuss later, this massive and stable
a Adapted from FNB (ibid. : 433)· correlational effect is predictable on theoretical grounds, and dovetails well
I,
" P < 0'05· with a variety of further findings about language acquisition. However, the
"I "" P < 0'025. remaining findings were less convincingly related to any linguistic or
I
acquisitional theory, as we will now try to show.
\
i
Syntactic simplicity. FNB claimed that a number of the correlations In
. rrelations between maternal speech and child
" Table 1 are predictable from' formal grammatical theory'. Their source was
\
TAB L E 2. Double-partwl c~. initial child age and language> .
!! language growth, partwllmg out an account of the so-called Standard Theory (Chomsky 1965), written by
I
I
\
Child language growth (Time, - Time,)
Jacobs & Rosenbaum (1968). Based on their reading of this work, FNB claim
I that English grammar represents only nouns and verbals in 'deep structure',
Vb/utt NP/utt
I
\
Maternal speech Aux/VP Infl/NP MLU

0'16 0'02
all further content being inserted
hypothesis about the grammar,
by transformation.
FNB next conjectured
Based on such a
that those elements
0'10
I Declaratives
Yes/ no-questions
0'25
0·88"""
0'01
-0'05 0'50+
-0'38
0'35
-0'29
0'16
0'19
introduced in deep structure
transformations are more complex.
are the simplest, while those introduced by

-0'55" -0'52+ -0'24

Imperatives -0'29 -0'02
-0'36 -0'07 One difficulty with this line of argument is that the Jacobs & Rosenbaum
Wh-questions -0'12 -0'08
0'58" 0'13 analysis differs in several major and relevant ways from the Standard Theory
-0'09 0'03 -0'16
Deixis 0'14 0'23 as it was usually described: for example, most versions of Standard Theory
0'51+ -0'05 -0'27
Expansion -0'51+ -0'50+
Repetition -0'58" 0'22 never in fact introduced surface lexical elements by transformation. Moreover,
0'14 0'38
0'34 0'10 0'31 linguistic findings of the last 15 years have overwhelmed and defeated
MLU 0'37 0'05
0'21 -0'05 transformational grammars of this general sort, which have yielded to far
S_nodes/utt -0'08 0'11
0'42
0'53+ 0"22 more persuasive and richer recent linguistic descriptions. (For discussion see,
Interjections
Taken from NGG (ibid.: 132).
for example, Chomsky (1981), Bresnan (1978, 1982), and eazd~r (1981).)
a
+ P < 0'10. Interpretation of empirical findings in learning against grammatical theories
P < 0'05· must be in terms of current knowledge of the latter, not early attempts that
"",," P < 0'001. by now have been rejected.
53
52
 THE MOTHERESE HYPOTHESIS
CHILD LANGUAGE
the few properties that correlates positivel and . .
growth in our own studies (NGG T y significantly with language
Most important, however, even accepting FNB's view of the grammar, the I . ' ,see able 2)
findings are not easily understood on this basis. After all, as Table I shows, n pursuit of a similar semantic claim F' .
the mother's speech are' less c ' lu NB offered the Idea that verbs in
the mother's use of verbs (on FNB's supposition, deep-structure elements) oncrete t an noun d h f
or, the young learners Thi . FN " s an t ere ore make trouble
correlates NEGATIVELY with acquisition of MLU and verbs. In contrast, the f . . IS IS B s inter tati f ,.'
negative correlation they obtai d b pre ation or the significant
mother's use of auxiliaries in yes/no-questions (on their supposition, NOT ame etween mat I
rate of using verbs in his h erna use of verbs and child's
deep-structure elements) correlates POSITIVELY with child learning of the h or er own speech (that'
t e mother uses verbs the less th hild I IS, apparently, the more
auxiliaries. At the same time, other elements alleged not to be in deep · ' ec I earns ab t h '
structures (the pronouns, copulas and contractions) correlate NEGATIVELY Interpretation seems rather odd Af 11 ?u t em; see Table I). This
ter
conjecture, it seems to be th I" h a , looking at the other side of this
with child learning. These facts taken together show that there is no consistent e calm t at to teach verb ' .
t em. In that case the moth' s It ISgood not to present
way of predicting the learning rate from the mother's use of the deep-structure h , erese interpretati h
verbs would finally be learned M' IOn as no explanation for how
elements _ if these are, as alleged, the nouns and the verbals only.
concrete than nouns seem . ore Ihmportant, the claim that verbs are less
In summary, FNB's results are not predicted by the grammar to which they s no more t an an li
to be sure, as the authors note the " unexp icated assertion (though
subscribe. In some cases, their correlations make no obvious sense (e.g. the
'literature).8 ' assertion ISoften made in the psychological
mother's use of copulas is negatively correlated with the child's acquisition
of noun-phrases). In other cases, the sign of the correlations is inexplicable Most generally, the semantic conjectures '
(e.g. the mother's use of verbs correlates NEGATIVELY with the child's possibility that motherese i 'h J d from FNB come down to the
. s a ere an now' Ian
acquisition of verbs). These puzzling findings have no external support from
present dogs and ducks rath h Chri guage, one that traffics in
· . er t an nstmas d
IS quite possibly the fact of the matter b es p~st an absent cats. This
linguistic theories in any of their various renditions, nor to our knowledge,
or any other study we kno f ' u~ no finding from the FNB study
from features of language behaviour. w 0 , suggests either T h '
maternal speech aids the I' ,HAT sue a property of the
Semantic simplicity. FNB argued that their findings may be in tune with , earrung, or HOW It w ld d
forthcoming results from L d & ,ou 0 so. (Not incidentally
semantic simplicity. They remarked that' abstract' language may be harder an au Gleitman b ' '
successful acquisition of h si h a out a blind learner's
than' concrete' language for a cognitively immature learner. A topic they took suc sig t-related terms a I k '
suggest that the notion of 'h d s 00 , picture and green
up from this perspective is the apparent finding that the use of pronouns by ere an now' be gs th e questions '
~q answer. it is designed
the mother is negatively correlated with certain growth measures (see again
Table I V They believe that the difficulty is caused by the deictic property ·Brevity. Next, FNB pointed to a set of corr .
(variable reference) of the pronouns - that these have no physical distinc- the mother's sentences and the hild' I elatlO~s between brevity of
relationship between simpl c Ih s anguage learning as instances of the
tiveness and are low in imageability. But this argument is weak. For one thing, h e speec style and h ild I '
such non-imageable words as fun and bad appear in earliest maternal speech t e maternal variables they interp t i h' c I earrung. In particular
f re m t IS regard '
and are not notoriously hard to learn. In addition, deictic terms are not requency correlates positively with hild I are INTERJECTIONS, whose
. OF, WORDS PER UTTERANCE hi h c I anguage growth, and MEAN NUMBER
generally hard to learn (deictic this and here are among the earliest vocabulary
growth (see Table I). Ho~e w IC correlates ne~atively with child language
items, for many English speakers), and deictic usage by the mother is one of' · . . ver, there are some diffi I' h
mterjections are utterances lik cu ties ere. For example
hIe yes, mm-hmm and th lik '
p rases. Moreover, interjection I " e I e, as well as isolated
s corre ate WIth growth in verbs. But how
(7] The authors also contend that pronouns are syntactically more complex than nouns,
'because they need more transformations before translation into surface structure','
However, the idea that pronouns are inserted by transformation has not been seriously' 8] In contrast, such investigators as Gentner '
6 " ~d some supporting empirical data for W~1982) have, provided a possible explanation
entertained by linguists for quite some time (cf. Bach 1970, Dougherty 19 9, Lasnik 1976);
Another argument for the syntactic complexity of the pronouns offered by FNB is that entner argues that verbs conflate s~ma ti y
world, .while nouns in all Ian ua
t ouns
might be easier to learn than verbs'
n ICe .ements variably, over the languages of the
pronouns mark case, gender, and number 'in fairly regular fashion'. But why should o
language would likely divide t~e d~~s categon,ze the world about equivalently (i e
pronouns' regular marking of these properties present special difficulties, as opposed to.
, .dogs smaller, with a word for each of tChoncept into say, dogs bigger than a breadb~x 'annd
nouns, whose case, gender and number is variably (often covertly) marked in Englishl'
'th
be, CIShild' pre-existing categorizations eseof two
obi categorines. ) H ence, according to Gentner
Should there be more trouble learning nouns in languages which mark case, gender an4'
, P asIS for early lexical learning , B u t notice
' jeers,G but not
that ' actions, could form as' ecure
number' in fairly regular fashion'? The developmental findings, in fact, are that regularly:
, ROPERTlES OF THE CHILD LEARNER (hi's ' ,entner S explanation rests ON INNATE
inflected nouns (as in Turkish) as well as the inflections themselves are easy to learn; as,
intoo Just
iust certain
certai conceptual cate ies) pre-exrstmg d isposrtrons
' , to carve up the w ld
opposed to irregularly inflected nouns and their inflections (as in Serbo-Croatian). FO,r , gories , NOT ON PROPERTIES OF THE DATA PRESENTED, or
82
the evidence and discussion see Slob in (19 ),
55
54
 CHILD LANGUAGE
THE MOTHERESE HYPOTHESIS
could mm-hmm help one learn verbs, since it contains no verb ?Some
reanalysis, separating verb-phrase fragments from interjections, might clarify A'replication of Furrow Nelson & B di (
this issue one way or the other, but in the present form of the data the
I 'ene tct I979)
n sum, Our overall reaction to FNB' di ,
conjecture from FNB is quite puzzling. As for maternal utterance length, we explanations as there are I' s ISCUSSIOnis that there are as many
corre atlOns few of th izh
will return to the issue of relations between brevity and language learning in before going further, it is essential to ~sk wh em trg ,t or compelling. But
a later discussion. For now, suffice it to say that there are reasons to question, We turn immediately then t I" ether FNB s findmgs are stable.
the stability of the relationship FNB have obtained, and theoretical reasons , ,0 a rep rcation of FNB I'
a reanalysis of our Own d t ' accomp ished through
, a a.
to claim that a restriction of sentential complexity might make the learning
HARDER rather than easier.

Intelligibility. FNB offer a final argument for the role of' input simplicity'. METHOD

This has to do with the phonological clarity of the mother's speech. However,
In response to the problems inherent in the .
FNB have not directly measured phonological clarity of the maternal speech. now divided our original sub'ect . N?G analysi., (see above), we
A finding that they attempted to relate to this issue is a negative correlation chronological age was very I ~ dPopu.latlOn into groups within which
c ose an re-did th I'
between language growth and maternal use of contractions. But we believe each of these age-equated gro 'Th e corre ational analyses within
this issue is a very complex one. ups. ese reanalyses a FNB'
met h odological complaint and t nswer to s general
. 0 our Own anal . f h '
Strictly speaking, there is nothing unintelligible about a contraction, from In. what we did previously d ib d YSIS0 t e problems inherent
,as escn e above.
the physical point of view. What is complex about it is that it conflates, within
a single lexical item, a pair of formatives which are in other parts of the Subjects
language two separate lexical items. For example, can't is not hard to hear;
Originally, NGG had three subject rou G' ,
the problem is that it is not formationally simple, but rather must be analysed ranging in age from 12 to gh ps. roup I consIsted of 3 individuals
15 mont s at the fi t .
as can+mot, As Gleitman & Wanner (1982) have discussed (and named the both very small and younger than the FNB rs measureme?t. ThIS group,
Three Bears Hypothesis), learners seem to have strong biases about the current analysis. Grou II' , ~roup, has been discarder] for the
semantic elements that can and cannot be conflated in a single word, and biases p consIsted of SIX iridivid I '.
18 to 21 months. (Note that G II ' VI ua s, rangmg in age from
in general towards representing formatives as separate words. (For discussion, . h roup subjects ar .
In C ronological age as the FNB subiecr . e appr~xImately the same
see the many citations in Gleitman & Wanner, and, particularly, Slobin (1973, subjects in MLU' see b I )G J s, however, they dIffer from the FNB
, e ow. roup III con' t d f ' .
in press), Newport & Supalla (1980), and Peters (1981).) The explanation to range from 24 to 27 mo th b ( SIS e 0 SIXmdividuals, chosen
of this correlation, then, if it is stable, depends on properties of the CHILD n s ut as the sample t d ) .
24 to 25 months at the first urne out ranging from
(who abjures certain conflations) rather than on surface, physical properties measurement For II
first-born and half later-b . II . groups and III, half were
of the incoming stimulation from the mother. .
f u II-tirne caretakers om, a were female All h d h
Th bi . a mot ers who were
. e su jects were upper iddl I
Summary of the FNB explanations. In our view, FNB have not succeeded and professional families Not' h h rm e c ass, from academic
, . Ice, t en t at each ' .
in the search for independent justifications for why their significant corre- '" same SIze as FNB's sin I ' group IS apprOXImately the
g e group.
lations were just the ones they were. Thus there is no reason, independent of
the correlations themselves, to think that their effective inputs are those that + Procedure
are simple AS STIMULI. If the effects are genuine, further explanation of the
The subjects were visited in their homes Th
initial state and (representational and/or inductive) biases in the learner will' liInguage development of thei hild . e mothers were told that the
be necessary to explain why these particular kinds of data were' the simple, " eIr c I ren was der i '.
method was simply for the h un er mvestIgatlOn, and that the
ones for learning'. In contrast, as we remarked and will discuss further, the' , mot er to chat ith d I '
dunng this hour. The invest' t ' ' WI, an p ay WIth, her child
more restricted correlations found by NGG fit in nicely with what is currently iga or s mvolveme t
note-taking. In practice co . b n was generally restricted to
known of a psycholinguistically crucial distinction: the open class/closed class' , nversatlOn et ' .
well as child and mother took I dtrr i ween mvestIgator and mother, as
subdivision of the basic linguistic vocabulary, and the syntactic functions, , , pace uring thes . B
aim was to detail the d'LI b e sessrons, ecause half our
Iuerences etwe I
these two classes subserve. to adults, the investigator m d e~ ~aterna speech to the child and
a e sure to initi t ' f
the mother, when it did ia e m ormal conversation with
I not occur spontaneousl The i .
to each home six months lat d . y. e mvestIgator returned
56 . er, an mstituted the same procedure, FOllowing
57
 CHILD LANGUAGE
THE MOTHERESE HYPOTHESIS
the second sessions, the mothers were informed that their own speech, as well
as that of the child, had been under investigation, and that the actual purpose TABLE 4, Means for measures oj maternal speech
of the study had to do with investigating effects of maternal speech on child
Maternal speech Age group II
language growth, We then solicited permission to use all the data that had Age group III
been collected, with knowledge of these purposes, In each case, permission Declaratives" 0'27
0'34
was granted, Yes/no-
questions"
Measurement oj the child and mother speech, Maternal utterances were
Imperatives"
0'21 0'20
0"20
separated into those addressed to the experimenter and those addressed to Wh-questions"
0'16
0'14 0'13
the child, Each set was coded for intelligibility, well-forrnedness, sentence Deixis"
0'17
0'17
length, structural complexity (indexed as number of sentence-nodes per Expansion
0'07 0'08
utterance and derivational length), psycho linguistic complexity (explicitness Repetition 0'28
0'15
with which the surface form preserves the underlying structure), sentence MLU
5"3 5"57
type (declarative, imperative, etc), frequency of self-repetition, and frequency S-nodes/utterance "16 1'21
of expansion, Only utterances addressed to the child are of present relevance, Interjections
0'14
Unintelligible
Child speech was coded for syntactic complexity, estimated through mean 0'03

length of utterance (MLU), mean noun-phrase frequency and length, mean ["J Two passes were made in coding the maternal s h ' '
verb-phrase frequency and length, inflection of noun-phrases (plural and into full sentences vs other (i t " peec ,F,rst, the utterances were divided
' 10 erjections ungra ti I
possessive marking), and auxiliary structure (modals and aspect marking) for and unintelligible and incomplete sentenc:s) Eac:::;; rea sentences, fragments, idioms,
of PERCENTAGE OF ALL UTTERANCES whi h f II' , hese categones was scored in terms
both the first session and the succeeding one six months later. Finally, 'r, IC e into the categor H '
categones ro r which there was no ' (' y, ere, we orrut mention of
, vanance e g idio ) d h '
, growth scores' were obtained by computing the difference between the first no obtamed data (i e there were VI't II " ms an t ose for which there were
' , r ua y no UI'lgramm ti I
and second interviews on each of these measures, (See NGG for the complete If were then submitted to further analyses which a Ica, sentences), The full sentences
SENTENCES which fell into the categor T'h Iwere scored 10 terms of PERCENTAGE OF FULL
description and examples of the coding scherne.) " y, ese atter categones a t d(
of t h ese within each age group add 0 re s arre and proportions
Some of these measures showed virtually no variance over the groups (e.g, also computed on full sentences Iup to 100 Yo, except for rounding errors), MLU was
on y ,
mothers' ungrammaticality); others were largely redundant with each other
on these groups (e.g. sentence type and psycholinguistic complexity), These
ranges, and standard deviations for our sub' ect
nonvarying and redundant measures are not reported below, The means, measures are presented in Tabl d ~ groups on all the remaining
, es 3 an 4,

TABLE 3, Means, ranges and standard deviationsjor measures oj child speech Analysis
We have already described our prior methods f '
Time, Time, minus Time,
reader is referred again to NGG W di 0 analysis: for full details, the
Child speech Mean Range S,D, Mean Range S,D, which involved performing sim I~an: IS~USS here o,nly the' new analyses,
speech and child I p partial correlatIOns between maternal
Age group II (18'5 to 2!'3 months) II and III. anguage growth, now computed separately for age groups
0'00-0'06 0'02 0'15 0'03-0'36 0'16
Auxiliaries / verb- phrase 0'01
Ii Inflections/noun-phrase 0'04 O'OO-O'I1 0'05 0'09 -0'02-0'19 0'09 Simple correlations, To compare our result '
MLU 1'52 1'05-3'32 0,88 0,86 0'25-2',8 0'73 performed simple correlatl'ons
b t h WIth those of FNB, we first
0'05 e ween eac of th
Verbs/utterance 0'21 0'00-0'47 0'17 0'35 0'30-0'45
speech and each of the measures f hild I e measures of maternal
II Noun -phrases / utterance 0'94 0'80-1'14 0'13 0'22 -0'23-0'55 0'20 ~'
shown in Table 5, As in the FNB st:d
c
~c anguage growt~, The results are
Age group III (23'9 to 24'8 months) are many significant correlations y tt om~are Table 5 WIth Table I), there
Auxiliaries/verb-phrase 0'09 0'00-0'31 0'14 0'27 -0'12-0'49
measures, However there ar ,sca ere across the child and mother
Inflections/noun-phrase 0'04-0'27 0'08 0'05 -0'10-0'19
0'15 , e many reasons to b f h
MLU 1'98 1'16-3'46 0,82 1'40 0'32-2'15 One problem has to do with their reliabilit ' e wary 0 t ese outcomes,
Verbs/utterance 0'31 0'06-0'65 0'21 0'35 0'07-0'55 on which each analysis was perfo d dY g ' iven the small speech samples
Noun-phrases/utterance 1°11 0'84-1'47 0'23 0'37 -0'07-0'70 , , rme ,an t he small nu b f bi
contnbuting to each cell Ad' , m er 0 su jeers now
, s oppose to our orrgi I t d .h
there are in the present analysis' bi , na s u y WIt IS subjects,
, SIXsu jeers 1D each of the two groups, Note
58
,~ 59 '
--"
 THE MOTHERESE HYPOTHESIS
CHILD LANGUAGE
'study this was performed separately for each of the two age groups. We began
5. Simple correlations between maternal speech and child by dividing the data into two halves by separating the odd pages of the coding
TABLE
language growth sheets from the even pages. We then computed each measure on each half
of the data, and then computed the (simple) correlations between maternal
Child growth (Time, - T'ime.)
speech measures and child growth measures on each of these two halves
Vb/utt NP/utt
Aux/VP Infl/NP MLU separately. The next step was to compare the correlations obtained on these
Maternal speech
two halves with those obtained on the overall analysis (that is, with both odd
Age group II
0'84+ -0'10 and even pages of the coding sheets combined). Our criterion was conservative.
0'12 0'17
0'23 0'28
Declaratives
0'08 0'24 -0'09 Only if the correlations obtained on each of the two halves were at
Yes/no-questions 0'73" -0'24
-0'27 -0'42 approximately the same level of statistical significance as those of the overall
-0·60+ -0'50
Imperatives -0'33 -0'43
-0·63+
, Wh-questions
-0'71+ -0'17 analysis did we consider them reliable; otherwise, we viewed the overall
;' 0'72+ 0·80" 0'51 -0'59 0·82"
0'44
correlations as at best unreliable, at worst artifacts of measurement error.
Deixis 0'51 0'15
0·81" 0'27 Because we will argue below that there are further problems with the simple
Expansion -0·62+ -0'39 -0'51
-0'74" -0"49
Repetition -0·01 0'22 correlational procedure in any case, and because the presentation of the full \
0'14 0'09
0·82" 0·66+ outcomes of the split-half analysis would be cumbersome, we present '
MLU 0'52 -0'34
0'89" 0'37
S_nodes/utt 0·61+ here only a crucial subset of the outcomes for illustration. Table 6 presents
0'77" 0'18
0'75" 0'30 -0·63+
Interjections -0'12 0'83+ a subset of the overall simple correlations taken from Table 5, namely the
-0'69+ -0'67
Unintelligible correlations between maternal complexity and child growth in MLU, verbs
Age group I II
\ 0'08 per utterance, and noun-phrases per utterance, for age group III. These are
r, 0'19 -0'70" -0'33 -0'34
-0'45
Declaratives -0'22 -0'28
the items on which FNB based the claim that simple input enhanced learning
0·62+ 0'01
Yes/no-questions 0'13 0'51
-0'3i 0'07
Imperatives -0'57+
0'38 -0'42 rate. As the table shows, we like FNB obtained numerous significant negative
0'37 0'03
-0'11
Wh_questions 0'24 correlations on the overall analysis - that is, the simpler the mothers' speech,
0'36 0'12
-0'18 0'48 -0'26 the more rapidly their children appeared to acquire the language. However,
Deixis -0'12 -0'47
0'59 -0'06 0'02
Expansion
0'51 0"44 0'74" in Table 6 we also present the outcome of the split-half analysis for just these
0'02
Repetition -0'68+ correlations.
-0'72+ -0'31
-0'46 -0'45
MLU -0·84" -0'67+ -0'98" As can be seen in Table 6, none of these correlations is reliable on the two
-0'23 -0'24
S-nodes/utt -0'32
0'26 -0'27 -0'59 split-halves. In all cases, a significant overall correlation shows up on the
0'15 -0'08 0'54
Interjections 0'33 . split-half as at best a significant correlation on one half but a nonsignificant
-0'04 0'27
Unintelligible
(often approximately zero) correlation on the other half. This outcome
+ P< 0'10.
P < 0'05· 6. Overall correlations and their split-half correlations for maternal
"•••• P<O·OI .
TAB LE
speech complexity and child language growth

. . he same number of subjects tested by FNB, and Child language growth (Time, - Tirne.)
however, that this IS about t . from about 100 to 300 per
that we had a larger sample of utteranceSs - rangmgther In short our sample Maternal speech . MLU Vb/utt NP/utt
. f b t 250 to 50 per mo. ,
I' child and ranging rom a ou . I f r the children and more than five
MLU Overall -0'72+ -0'31 -0·68+
is on the average more than twrce as arge 0
Odd -0'55 -0'15 -0'72+
\ \ times as large for th~ir mothers. f lations reported - 55 for each of Even -0'76" -0'41 -0'54
J
Considering the sizeable number 0 corre mber of substantial S-nodes/utt Overall -0'84" -0'67+ -0·98 ••••
one might well expect a nu
our two age groupS - . h hal data base (small number of Odd 0'09 -0'15 -0'08
. b h lone given t e s axy . Even -0'83" -0'75" -0'92""
correlatIOns y c ance a '. f h) f m which they were derived.
d t ornts or eac ro
\, subjects, althoug h many a a p f' ting the reliability of these + P<O·OI.

l\
. some way 0 estrrna I
Under the Clfcumstances, . d loyed the same procedure. P < 0'05·
. d tial To this en ,we ernp "
correlatIOns seeme essen I. . 'r -half ' analysis .. In the present "" P < 0'001.
i NGG , namely what we called a sp It
use d In 61
60
 THE MOTHERESE HYPOTHESIS
CHILD LANGUAGE
TAB LE .8. ~artial correlations between maternal s eec .
7. Partial correlations between maternal speech and child language (partzallzng out initial child Ian ) ~ . hand child language growth
, guage , but omitting unreliable correlations»
TABLE
growth (partialling out initial child language)
Child language growth (Time 2- T' rrne, )
Child language growth (Time. - T'ime.)
Maternal speech Aux/VP Infl/NP MLU Vb/utt NP/utt
Vb/utt NP/utt
Infl/NP MLU
Maternal speech Aux/VP
Age group II
Age group II Declaratives 0·66 0'23 0'28 0'98* -0'14
0'98* -0"4
0'23 0'28 " Yes/no-questions 0'72 0'10 0'43 -0'02 0'48
0·66 -0'02 0"48
Declaratives 0"0 0'43 Imperatives -0'7' -0·6, -0'54 -0'27
0'72 -0'89* -0'27
Yes/no-questions -0·6, -0'54 Wh-questions -0'20 -0'76 -0'5' -0'44
-0'7' -0'5' -0"44
Imperatives -0'20 -0'76
-0·82+ Deixis 0'65 -0'73
Wh_questions -0'73 0·85
0'84+ 0·65 .Expansion 0'85+ 0'30 0'73 0'39 0'53
0'8,+ 0'39 0'53
Deixis 0'30 0'73 Repetition -0'94" -0'58 -0'86+ -0'73 -0'5'
0'85+ -0'73 -0'5'
Expansion -0'58 -0·86+
-0'94* MLU 0'85+ 0'2' 0'46 0'1 I 0·61
Repetition 0'11
0·6,
0'2' 0"46 0·8,+ S-nodes/utt 0'90" 0'37 0'72 -0'39
0'85+ -0'39
MLU 0'37 0'72
0'90* Interjections 0·82+ 0'3' 0'87+ 0'32 0·60
S_nodes/utt 0'32 0·60
0·87+ Unintelligible -0,68 -0'3' 0'99"
0'82+ 0'3' -0·85+
Interjections -0'31 0'99*
-0'92* -0·68 Age group III
Unintelligible
Age group III Declaratives 0'48 -0'73 0"2 0'17 0'75
0"7 0'75
-0'73 0'12 /Yes/no-questions 0'9'* -0'26 0'92* 0·66 0'28
0'48 0·66 0'28
Declaratives -0'26 0'92* Imperatives -0'69 -0'03 -0'58 -0'34 0"9
0'9'* -0'34 0"9
Yes / no_questions -0'03 -0'58 ,Wh-questions -0'04 0'73 0'07 0'21 -0'37
-0·69 0'2' -0'37
Imperatives 0'07
-0'04 0'73 -0'65 0'26 -0'34 -0'42 -0'79
Wh_questions -0"42 -0'79
0'26 -0'34 0'59 -0'53 0'21 -0"5 -0'15
-0'65 -0"5 -0"5
Deixis -0'53 0'21 -0"4 0·66 0'30 0'58 -0'08
0'59 0'58 -0'08
Expansion 0·66 0'30
-0"4 0'12 0'50 -0'5' -0'24 -0'23
Repetition -0'24 -0'23
0'50 -0'5' 0'58 0'26 -0·60 -0'49
0·,2 -0'49 -0'95*
MLU 0'26 -0·60
S_nodes/utt 0'58 -0'25 0'32 -0"5 0'15 -0'23 -0'25
0"5 -0'23 -0'28 -0'06 -0'22 -0'08
-0"5 -0'08 -0"4'
Interjections 0'32 -0'22
-0'28 -0'06
Unintelligible -0'4'
, 'P < 0·,0.
+ P < 0"0. P < 0'05·
* P < 0'05· 'Unreliable correlations' are those who h ..
.,,,overall analysis but which were insigni~ca ~ere significant or marginally significant on the
,~'" n on one or both of the split halves.
suggests that all of these overall correlations are unreliable and may thus be
artifacts of measurement error. They therefore can hardly be taken as a proper spe.echand child language growth are the .
basis for theorizing about the Motherese Hypothesis. Note that FN hich to address the hypoth . W h appropnate measures through
. eSls. e t erefore t -
performed no similar statistical test for the reliability of their findings, whi artial correlations. urn to a presentation of the
9
were based on seven subjects, with fewer measurements of each. Partial correlations. Simple correlations b
In any case, we do not believe that simple correlations between mater anguagegrowth may be contami t d etween maternal speech and child
t.~~e(in bo~h age and linguisti:n:b~iti~~ ~ *~d variable - the child's initial
(9) We should note that it was not generally the case, either in the simple correiatiofl.!
IS,CUSS partial correlations in whi h ) erefore, we here present and
presented here or the partial correlations presented later, that the split-half analysis , h .chiild growth scores ,IChrl maternal speech measures are correlated
resulted in such widely discrepant correlations for the two halves of the data or in suc\ It .
a large proportion of overall correlations which must therefore be discarded as unrelia
"l'd' .
I S initial age and langu' w I e removmg the' vanance attnbutable
. to the
Nevertheless, as the present subset illustrates, this does sometimes happen, undersco
age.
the necessity of the procedure for assessing the internal reliability of the 0 In contrast to NGG , we now performed these correlations on two separate
correlations. 63
62
 CHILD LANGUAGE
THE MOTHERESE HYPOTHESIS
age groups, within which age varied only slightly. We therefore partialled out satisfactorily, should be approached '
only the child's initial score on each language measure. (FNB could not data of this sort CAN support th "~fost gmgerly, However, correlational
eonzmg I the foIl' "
perform such a partialling, as their only measure of child's initial stage- are met: (a) they should coh ' h ' owmg mmlmal requirements
I ere wit what IS k
MLU - did not vary. As already mentioned, this measure did not guarantee anguage and its learning' (b) th h nown on other grounds about
d ' ,ey s ould part' I k
initial equality of the subject population since FNB's subjects did in fact vary to 0 with initial variance in the sub' la, out nown contaminants
on comprehension measures, which in turn correlated with the growth tern ally reliable using , ject populatIOn; (c) they should be in-
, ( ,conservatIve cnteria f luari
scores). Table 7 presents the overall partial correlations for each of the two "', d) even so, they should b t k or eva uatmg reliability; but
h e a en as pro tern and '
age groups, without taking into account their reliability as assessed by a ' t e enormous difficulty of' t ' ,suggestIve only, because of
, d m erpretmg them causall W' h
split-half analysis. Many of these correlations are sizeable and reach statistical nun ,we turn now to consideri h " y. It these provisos in
significance. But again, this may be because the small sample size and limited Stage dependence of tb I ng t e remammg results.
e corre atlOnalfindi I '
number of utterances for each mother and child may lead to large but a prediction made by FNB I h ngs. nspecnon of Table 8 supports
, name y t at the effe t f
unreliable findings. Table 8 therefore presents these same correlations, but may vary with the language sta e of th hi c so maternal characteristics
without those for which the two split-halves did not lead to similarly , almost all of the correlations a gl' . de c ild learner. As the table shows
. re unrte to th .'
significant outcomes. In detail, we have omitted overall correlations with Unmterpretable one) Th ' e younger group (mcluding the
. e one major exc ti h
p < o- 10, if these did not survive the split-half procedure at the same level maternal yes/no-questions on a '1' ep IOn as to do with the effect of
UXIiary growth hi h i . . .
of significance; and we have omitted significant overall correlations if their younger group, but reliabl iznifi . ,w IC IS mSlgmficant in the
splithalvesdidnotreachatleastthemarginal- P < 0'10 -levelofsignificance. To the extent that such r~s:I~~~:I~nt m t~e older group.
they may suggest that selected f uP un er further experimental review
. eatures 0f th ' ,
mfluences only relatively ea I . h I .e envIr?nment exert their major
RESULTS ,I ' r y m t e earnmg pe d (d .
earnmg the relevant stru t' no esprte the fact that
hi cures IS not compl t f .
As a preliminary to discussion, note that many of the simple correlations .fall t IS). Such a finding would k ,e e or qUIte some time beyond
. ma e sense If for e I h .
away when partialled to correct for variability in the children at the initial of th e envIronmental input wI' ' xarnp e, t e child's analysis
. ere re atively su fici I (
measurement (compare Table 5 with Table 7). But note further that still more Input-dependent) in the I per cia and therefore unusually
. ear y stages of I . b
correlations fall away when the partial correlations are submitted to the , orgafllzationally deeper (a d h f earmng, ut structurally and
d d' n t ere ore much I '
split-half analysis (compare Table 7 with Table 8). This is the first suggestion epen ent) m later stages of I' ess superfiCIally input-
fi d' earn mg. Alternativ I h
that there is real difficulty - not just theoretical difficulty - in relying too, n mg may be an artifact of differi . e y, t e stage-dependent
heavily on findings from any single analysis of correlation results, from just age groups, or differing relevan I efnng vanances of the measures in the two
. . ce 0 our partIcula I '.
a few individuals. POInts in learning Further h r anguage mdlces at the two
. . researc and 'I I
However, the same point is made most tellingly by looking at the outcomes , 'dIfferent, less superficial m f' partlcu ar y research using quite
b' , easures 0 the envr
themselves. One of these is surely 'garbage'. We achieve, much to our ,e reqUIred to disentangle th' ronment and of learning, will
- 'B . ese ISsues,
chagrin, a POSITIVE, 0'99 correlation between maternal UNINTELLIGIBILITY and . u~ in so far as this general effect can b
child growth in verbs per utterance, in the younger age group (Table 8, Group .most Important implication is worth t . e assumed to be a stable one, its
II). As we stated earlier, when a larger number of correlations are done on ". '(and see particularly Newport (1 ~ fatmg here. As our fi?dings have shown
a very few subjects, spurious correlations are likely to show up here and there, , mother's usage does not h 97;: or ~ fuller analYSIS in this regard) the
Since FNB used less stringent statistical procedures, again on a very small " period from one to three y~a:~gAe t rlalm~tlcallY during the child's lear~ing
.h bi . a tImes for I
sample of subjects and fewer data points, they obtained a larger number of WIt su 1ect/auxiliary inversl'o ,examp e, yes/no-questions
n appear as a rnai
such uninterpretable results (e.g. the mother's use of copulas impairs the , corpora. And at all times th . f jor segment of the maternal
. h ' e vanous unctors a . hei
child's learning about noun-phrases). In t e maternal utterances Y t ppear In t err requisite places
f hi . et, as we now seem to h hi
Since the present analysis yields one such correlation, which does not reflect 0,' t, ISmaterial at one age but n t t h see, t e c ild exploits some
'fin . 0 a anot er Thi . h
what we know of the real world of learners and tutors, some question arises ,e tunmg' hypothesis namely th t h. ISISt e reverse of the so-called
'j-' , a mot ers cha thei
about whether one should interpret the other correlations (Table 8) with great' ,e~rns, in correspondence with th h' nge err usage as the child
seriousness. Our own view is that correlational effects from a small number .dnven learning procedure (see Cros: (~ angmg need~. of an environmentally
of subjects, whose precise initial states cannot really be determined very to explain language learning' a d P' ;77) for a POSItIOn that uses fine tuning
3 ,n in er (1979) for discussion). As we now
65
]CL II
 THE MOTHERESE HYPOTHESIS
CHILD LANGUAGE
subjects . contributing
. to each an a IYSls
. b ecame smaller 0 h .
see, it is most importantly the CHILD who changes (in the material he attends we are inclined to believe that th . . . n t eoretical grounds
'1 ese original results I d
to and exploits), rather than the MOTHER (in how she speaks). As usual, we resent f at ure to replicate th h are rea, espite our
P
must look to properties of the child learner, more than to specific properties m uture imvestigations.
. future em ere ,eve and beli th ey will
. show up again

of his environment, to explain the learning (see Newport (1982) for a general
discussion of the effects of learners on language design).
DISCUSSION
Major correlational effects. Table 8 reveals that there is an effect of maternal
yes/no-questions on the child's growth in auxiliaries and (as an artifact of Explaining the learning effects from the maternal corpu
this), his MLU, for Group III only. As stated above, the remaining significant The
. overall findin g fr om our new analysis is th t hs
and reliable correlations are for Group II only. In that latter group we find in our procedures to the legiti t bi a ,wen we have responded
NGG ma e 0 jections of FNB th '.
a marginally significant effect of maternal expansions on the child's growth, are largely reproduced Th . ' e original results of
ne
in verbal auxiliaries. Further, there is a significant positive effect of maternal learning effects appear to b . fie °d major proviso is that the bulk of the
e con ne to the f
complexity, measured as S-nodes per utterance, on the child's growth in , n contrast, we did not in th younger 0 the two age groups.
IFNB ese new analyses rep d
auxiliaries; and a marginal effect of maternal complexity on the same auxiliary , except where their find' I' ro uce any of the results of
ddi I mgs rep icate our ow W b I
variable, when complexity is measured in terms of maternal MLU (the same a rtiona outcomes in the FNB n. e e ieve that the
measure that FNB call WORDS in Table 1). Thus the results are the same in correlations between mother and hstldUddY' ~ scattered and puzzling set of
a . c I , enve from th . fail
major respects as in the original analysis of N GG (Table 2): the mother's ppropnate statistical proced d eir ai ure to institute
I h ures an cross-check h
effects are primarily on the child's growth in the FUNCTOR or closed-class engt above. Then how ca s, as we ave argued at some
n we account for the fa t th FN
vocabulary; and primarily the mother's closed-class usage has effects on the repro d uce our own results? It sh ld b bvi c at B sometimes
_ ou e 0 VIOUSthat I
child's growth. Many other features of the mother's usage have no measurable' .more conservative statistical p dures wi ess conservative and
. h roce ures will h h
effect on any measure of the child's growth; and many aspects of child p enomena which are rob t d ave t e same outcome on those
us an stable' th '11'
growth are affected by no measured feature of the mother's usage. A single p enomena arising from ' ey WI differ on unstable
h measurement error in th t h I
new effect in the new analysis that crosscuts this major distinction, again for proce ures may take these t b ' ' ate ess conservative
. d 0 e genume outcom hi l
Group II only, is a significant relation between maternal,declarative sentences vatrve procedures will eli h es, w I e the most conser-
irrunate t em as m
and the child's growth on verbs per utterance. All of these major effects will concordance that does exist h easurement artifacts. Given the
I across t ese three an I
genera explanation for the stabl fi di a yses, we now turn to a
be discussed below. There are two properti f he n mgs that reappear in all of them.
Subsidiary effects. The remaining results in Table 8 again reproduce those es 0 t e maternal corpus th .
of NGG and requirelittie additional discussion. As before, we find a stable and FNB seem to play . I I . at, accordmg to both
NGG do wi crucia ro es in the I .
effect of interjections (e.g. mhmm) on the child's use of auxiliaries and their to 0 With COMPLEXITY and SALIENCE of the data earrung process, These have
artifact, MLU. (Notice that these are the same measures on which we postulated properties (e gv semanti presented. To be sure other
.. semantic transp ,
repeatedly find environmental effects.) NGG related this finding to a general, - forward, but on insecure or arguable inte are~cy; see ab~ve) have been put
if vague, notion of' reinforcement'. But we did not find the original results across the three correlational st di rWPretatlOns, and Without stable effects
M th H u res, e now exami h I
terribly interesting for understanding language learning: the question is how, -, o erese ypothesis by revi . . I' mme t e ogic of the
. ewmg Its c alms about I'
even given reinforcement, the child manages to generalize always and only, .an re atmg these to the obtai d . comp exrty and salience
lit d I . ame correlational d t d '
from old grammatical sentences to new grammatical sentences. Finally, there I erature on language acquisition. a a an to the theoretical
is a negative correlation between maternal repetition (a measure not taken by
FNB) and the child's acquisition of the usual materials (auxiliaries and MLU Complexity
measures); for discussion, and evidence that this correlation is a secondary The crucial component of the Moth
best from the simplest data F erese Hypothesis is that the child learns
effect of the types of sentences that get repeated, see NGG. . or example FNB . t
Two correlations that we suspect are real fail to reach significance in Table' measures of maternal speech ' . m erpret almost all the
h I as measures of It I'
8, though they did in the original NGG analysis. One is the negative effect t e east complex speech best h . s comp exrty, and argue that
it i supports t e learner B t h
of imperatives on auxiliary growth. The other is the positive effect of maternal I isnot really obvious why fo I . u as we ave discussed
. ' r examp e maternal '
deixis on the child's learning of the noun-phrase inflection (plural). It if ranos should be considered ki ' pronouns and noun-to-verb
as ma mg any con tributi
n utron to complexity. In
sufficient here to note that certain effects had to fall away when the pool of
67 3-2
66
 THE MOTHERESE HYPOTHESIS
CHILD LANGUAGE
towardsneither
case, the clause
simplhy po th esis
. (h ' towards structure-dependence);
t at IS, in this
contrast, there is a prima facie case that the longer the mother's sentences in
. ,. e sentences nor complex sentences ld b
words, and the more propositions these sentences contain (S-nodes per informative, since the child does not ar ri wou e more
utterance), the more complex that speech in terms of known language basis of input data at all oes not arn.ve at the .correct conclusion on the
descriptions. Therefore we assessed the relation between maternal complexity theoretical position just' ~~:ct~:daPbPropnhateldesc~IPtlOn of the facts, and the
a out t e earnmg problem see Ch k
and language growth using these two measures of the mother.
( 1975), who proposes the second of these two alt ti ' oms y
NGG found no relation between complexity and learning on these measures Related discussio hat oredi erna ives,
(Table 2). But in our reanalysis there is a significant correlation between
a(19b~:)efiTcihal
.effect konf/i:C:::S:et~:::I::~y~ :::e;:!~~ ~!I:;~n~~~~::vi:;
maternal S-nodes per utterance and child growth, and a marginally significant
, err wor ormally did'
correlation between maternal MLU and child growth, for the younger group ational grammar of roughly th:~h:~:k evice that will.acquire ~ transform-
only (see Table 8). Note that these new correlations are POSITIVE. In contrast, problem the f d Y(1965) vanety m finite time. A major
using simple correlations, FNB obtained NEGATIVE correlations with these constituents~o aaCseto 7::~hat the I:arner would not know how to attach moved
same measures (and so did we, for simple correlations in our older age group, had applied Th' blthe dfenved phr~se structures after a transformation
. IS pro em ormally disa I if hi
as shown in Table 5). Given the obvious caveats about different subjects and sentences are available as data _ namely th p~earsh'o~ y I I~hly comple\x
different statistical techniques, we can say little more than that the empirical transformations that a ' ose in w IC successive movement
relation between input simplicity and language acquisition is far from settled. composition of the I tt pply to already moved constituents establish th~
a er.
But there is a prior theoretical question. Suppose you have a theory that
The main lesson we draw from th h ' . , "
emphasizes the role of learning from the environment. Should you predict to show that the la . I ese t eonsts IS this : It IS relatively easy
that the learner is best served by simple input data (as FNB seem to find), it is awesomely h:;;:rg:~ss~:~able If t~~ input includes complex sentences;
or should you predict that complex data are better for the young learner (as simplest sentences T'hi .. learnabdlty If the input IS restricted to the
y
our reanalysis seems to indicate)? learn a system who~e st~~~~:I::Oa~es~~~~~r:~t .reall com: as a, s~rprise. To
The most explicit language-learning theories we have available (Wexler & sensical to suppose that data who h rni gl~g and vanous, It IS common-
Culicover 198o, Chomsky 1981) do not depend on the ordering of the input Data drawn from only t f he mirror this range are the most helpful.
par 0 t e range rru ht di h .
data at all. However, all such explicit analyses that we know of require that, learner will make about th h I g istort t e conjectures the
I not surprised that our rean rl woe range. On these logical grounds, we are
for success, the learner must hear data of at least moderate complexity early
\1:I in the learning sequence. Basically, this is because simple sentences fail to range of data from the I a YSISshuggests that mothers who produce a greater
anguage ave children who di h .
Ii exhibit all aspects of the syntactic structure.
transformations of the Extended
For example, the movement
Standard Theory, as well as of earlier
rules more speedily. If thi ffect I . iscover t e appropnate
our own and others' corre~:t:o::lt ~:t:table (an Issue certainly in doubt, for

II versions of transformational
ment is from clause-position
theory, are structure-dependent;
to clause-positions,
that is, move-
not from string-position to
in direct opposition to the usual M 'has we have repe~tedly stressed), it is
simplest input speeds language learni~g ~~e~:tHYhPottheslsh; namely, that the
. w a ever t e outcomes of the
string-positions (except as string-positions are artifacts of the clause-
positions). An example is the subject/auxiliary inversion of yes/no-questions,
:1, [IOJ and
More precisely, the Motherese Hypothesis sa s t '
in which the auxiliary of the main clause (not the first auxiliary) moves to the more complex input later 0 ' h Y hat the child receives simple input first
front of that clause (not to the front of the string). This explains why adults mentioned are not genuine dl'ffic' It' ne ~Ig thsuppose, therefore, that the difficulties
u res since tech ld 'II
owever, the problem with gettin 'I I WI eventually get complex input
say In the summer, do you go to camp? rather than Do in the summer you go H
have been formulated incorrectly !yctohmPtex dahta only later is that the rule may alread~
to camp? and why they say Is the man who is here a fool? rather than Is the e ime t e relevant com I d h
M, oreover, some errors which the child could' ~ ex ata ave appeared.
man who here is afool? If the learner is exposed only to the simplest sentences, simple data only (e.g. formulating bi ~ake in f~rmulatmg rules on the basis of
su ject+aux inversion m t f ' ,
he has no way of choosing between the string hypothesis and the clause t han t he main-clause auxiliary onl could ,erms 0 ANY auxiliary, rather
feedback when the child produce y) be .repatred only by NEGATIVE data (i.e.
hypothesis, as both will derive correct simple sentences. On logical grounds, This is because such an incorrecSt afn unglrammatflcal utterance that it is ungrammatical).
". o
rmu atron 0 the ul 'II f
then, the complex sentences should be more informative to the learner than t h e POSitive instances the child '11h (si r e WI un ortunately predict all
. WI ear smce they II '
the simple ones. This is precisely the result our reanalysis achieves, and FOR ,b emg moved), so that positive instances will n are a instances of some auxiliary
incorrectly. However there I'Sn id h ever tell him that his rule is formulated
, 0 eVI ence t at ch ld ' .
PRECISELY THOSE STRUCTURES (NAMELY, AUXILIARIES) ON WHICH THE ARGUMENT or systematically' nor _ much' I ren receive negative data either often
IS BASED (again, see Table 8). The only alternative to this is to assume that, '..' more Importantly - MAR ' h
istinguished from corrections of th KED in sue a way that it could be
d o er matters, such as truth or morality (Brown &
ji although the child may receive only simple sentences, he is innately biased
69
68

\
\ '
 "
THE MOTHERESE HYPOTHESIS
CHILD LANGUAGE

. . t s will have to face the question for the learner is 'something else' (something that will predict the effect of
di all mvestlga or hi d the yes/no-questions).
various correlationa I stu res, .' I .d the learner rather than 10 er
d uld in pnnclp e a! nt This problem seems complicated enough. But one more feature of the
of WHETHER simp Ie ata co d d . detail. The intuitive argume
. h they coul 0 so, m id learning makes it more mysterious yet: though the learner profits FROM the
him and explicate ow . flecf far from self-evi ent.
\ , . b helpful IS, on re ectlOn,
I that simple must e
yes/no-questions, what he first accomplishes (presumably, on the basis of
analysis of the yes/no-questions) is reconstruction of the canonical form-
. lf and the closed class. di which never appears in the questions. That is, the child first utters both
\ Salience, canomca orm d b th various correlational stu res
Another important issue ad~r~sse ~ter~al speech, and not others, ~a~e declaratives and questions with medial auxiliary structure (You will pass the
ns why some charactenstiCS of m h ood deal - of selectivity salt, but also What you will pass ?), though the latter virtually never occurs
\ concer . There is some - per aps a g d) in the input in this way (Bellugi 1967).
effects on learnmg rate. e child learns the functors (e.g. t~e, an~, -e
\ in the learner; for example, th host frequent lexical Items m the These complex findings submit only to an equally complex explanation.
I Our own position is as follows. (I) Only certain items are environmentally
rather late even though these are a: :h;Uchild differs from a tape recorder
i mother's speech. To the extent th. formation offered to it), part of the influenced materials - the so-called CLOSED-CLASS items, and the structures ip
which they participate. (2) These items are learned from input which displays
(wh ich must take in exactly the m b k on the child himself. What
I . is thrown ac . them in ways that match the information storage and manipulation biases of
explanation of language . ~armng I h that he accepts and analyses certam
are his internal dispositIOns, sue h ? Given the various findings under the learner. (3) 'What is learned' depends on the child's bias towards
. ignores ot ers. h . s reconstructing' canonical form' in the language being learned. We detail this
materials, but rejects or b f such selection mec amsm
. t uppose that anum er 0 . d position below. As will be clear from the exposition, these three components
discussIOn, we rn us s ERS) intervene between input an
(what NGG have c~lled LEAR~I:; :~~:ourse of language lt~arning. To t~e closely interlock, and together form a plausible generalization about the
output all along the line throug . I. rent it is they and not their language learning process.
h hildren select what matena IS sa I ,
h
extent t at t e c f the acquisition process. Special status of the closed class. As we have stated, the main stable
caretakers who are the prime movers 0 . I sentences The maternal correlational effects are limited to effects of and on the closed-class subcom-
le concerns canonlca .
An important ex amp ical form - do not have ponent of the language. In the original NGG study, this distinction appears
. h f structures closest to canon
declaratlves - t e sur ace h three correlational studies, on the categorically: the only significant effects are for closed-class materials (col-
. ble effects over tel umns 1 and 2, Table 2) and their artifact, MLU (since closed-class items
consistent measura , I . I relatl'on of maternal dec a-
fl' (on y a sing e cor ,
appropriate aspects 0 earmng occurs in the present study, contributed to MLU, an effect for closed-class materials will also often show
hild th in verbs per utterance,
ratives on c I grow . f '1 ar in either of the other two up as an effect for MLU; see column 3, Table 2). There are no effects for
II nd even this one ai s to appe I .
for G roup , a . f 'basic' sentences is hard to exp am open-class materials (columns 4 and 5, Table 2). In the present analysis (Table
. ) Thi lative ineffectiveness 0 . I
8), if we disregard the spurious negative effect of maternal intelligibility on
studies. IS re . h imple declaratives early 10 earn-
on conjectures that the learner reqUires t e Sl . t m (cf Pinker & Lebeaux child language growth, there is only a single exception to this general
k hi h to build the syntactic sys e .
ing, as the roc on w IC . '1 derived structure (the subject/aux inverted distinction. In the light of this contrast, it will be worthwhile here to consider
1982). In contrast, a partlcu ar. lati I ffect in ALL THREE of the the open-class/closed-class distinction more closely.
. ) h ws a massive corre ationa e ., .
yes/no-questiOn so. 1 d stable fact shows that simpliCity In Closed-class items, roughly, are the inflections and functors, those items
· 1 t dies This one arge an . 1
corre 1atiOna s u. h d I ti on almost any grammatlca that can occur unstressed in the languages of the world. These include the
( which should favour t e ec ara ives, . .,
a grammar ib . li ity for the learner' rather, simplicity determiners, certain pronominals, complementizers, certain prepositions
theory) does not directly descn e Simp ICI ' and postpositions, certain time adverbials, case markers, tense and aspect
. Moreover there is no evidence that children markers, and so forth. In English in particular, these items are likely also to
Hanlon 1970, Wex!er & Cuhcover ?:~~:ect-aux in~ersion, negative data or not. These be contractable (for discussion, see Zwicky 1976). Just how this closed class
ever produce such Incorrect forms 0 r that the child receives and uses complex data from
facts taken together thus suggest eitbe '1 . h ed'lsposition towards structure- (and its distinction from the open class) should be formally characterized is
h .' tely bui t Wit a pr .
the beginning; or that e IS mna f lations are not considered; or that he IS still a matter of some debate (see Chomsky & Halle 1968, Kean 1979, Bradley,
dependent rules, so that the incorrect orrnu I t 11 until complex data are available.
Garrett & Zurif 1979). But even in advance of a secure formal
innately built to avoid formulating mov~mentMruehsa a Hypothesis which expects that
. t odds with a ot erese , h characterization, it has become obvious in recent years that the distinction
All of these alternatives are a li and that the environment, rat er
. I d ta will be the most helpful at the ear lest st~ges, . 0 0

between open and closed class is crucial to a variety of linguistic functions

simp e a 0 dO 't ns of the child organizes his learning.
o

than the native pre ISPOSI 10 '

71
70
 CHILD LANGUAGE
THE MOTHERESE HYPOTHESIS
and performances, e.g. speech planning (Garrett 1975), parsing (Wanner & see fn. I above). These results are a ain . .
Maratsos 1978), long-term language forgetting (Dorian 1978), and dissocia- learning effects for the closed clas g b consistent With specific and dramatic
tion in certain pathologies (Kean 1979, Marin, Saffran & Schwartz 1976, class. s, ut few and unstable ones for the open
Bradley, Garrett & Zurif 1979). Our investigations, including the correlational . There is evidentl h .
. y, t en, a consistent dift
ones discussed in the present paper, repeatedly show a distinction in learning open-class Items in childh d I' erence between closed- and
as well, based on this same categorial cut. . ' 00 earning and in .
pat h ological linguistic perf a vanety of adult normal and
For example, we have studied the acquisition of gestural language by deaf fin . lOrmances We b Ii
dings we have obtained (almost solel e leve that the correlational
children not exposed to a full gestural corpus (Feldman, Goldin-Meadow & structures) are thus bolst db' Y ~or the closed-class items and
Gleitman 1978). The relevant finding there was that, under conditions where ·
Psyc h 0 IIngulstIc'. ere y this senes of I t d fi .
literature S h' re a endIngs in the
the exposure to formal language input is radically reduced, many open-class if '. . uc conSistency .th h I'
speer c analYSIS, i.e. the part" II' WI t e Iterature, given a
functions (e.g. developing sentences with appropriate argument structure) th ra ing procedure sup t h
e proc~dure itself. In contrast th' ' par. s t e reasonableness of
emerge at appropriate ages, while the closed-class items and functions do not Over a Wide and often perpl .' e Simple correlatIOns of FNB scattered
seem to appear. With somewhat less-reduced, but still impoverished input, exing spectrum f hild '
measures, do not appear to hav . I 0 C I and maternal language
closed-class items develop, but only when the exposure is during infancy; for . . e eqtrrva ent extern I .
InCOnSistency with other k f . a support; In our view! this
!: nown acts di . . h '.
example, in the acquisition of American Sign Language (Newport 1981, proposed new data-collection d I' mInIS es the plausibility of their
Newport & Sup alIa 1980), learners exposed to ASL only late in life do not Salience. We have d ban ana ytrc procedures,
argue a ove that it is pri '1 h
reliably acquire the closed-class morphology. However, native learners whose t h at are the environmentall . fl man y t e closed-class items
. y In uenced teri I
parents learned ASL late in life, and whose input with regard to the closed . Influenced by the environment' I h ma erra s. But HOW are these items
class is therefore strikingly impoverished, do uniformly succeed in acquiring attention on how caretaker sel~cti: t oe present study,. we are focusing
a set of closed-class items and functions. They apparently do this by , language affects the learner But b c b f .str~ctures avaIlable in a single
reanalyzing the open-class, and some irregular closed-class usage, of their int I . erore egInmng w h Id
erna properties of th' , e s au emphasize that
d'C!"' • e vanous natural lang h
r , parents. Similar phenomena have been widely documented in the reanalysis luenng lImits and opportunl't' c hI' uages t emselves impose
,I of pidgin languages into creole languages, which occurs primarily under ( . res lor t e earmn hild F
1973, In press) has presented'd h g c I . or example, Slob in
i conditions of native acquisition (Sankoff & Laberge 1973, Bickerton 1975). d f eVI ence t at child I'
en s 0 words i.e. they I . . ren isten selectively to the
In contrast, the open-class items and functions appear regardless of input d'l h' earn postposltlOned cI d I
rea I y t an pre-positioned clo d I . ose -c ass materials more
or time of acquisition. For example, they appear without accompanying leamed contains important pr:;o~~~s~l:atenals, But if ~he language being
closed-class functions, or with greater variability in the appearance of the caretakers are not free t ed-class matenals, of course the
closed-class, in the acquisition of ASL later in life and in the devising of pidgin o present these word fi II Th
generate d by the mother c - na y, e environment
languages by adults (see the previous references, and Slobin (1977)), as well o I . hi an respond to the child' I .
n y Wit, In the limits imposed b h . s earnmg requirements
I as in more ordinary second-language
(1978) has suggested that the same distinction
learning late in life. Goldin-Meadow
captures what is learned (the
(1982)has recently described
of' t .
In erestmg differences in I
th y t Ia conventIOns of each language, As Slobin
, ese anguage-specifi
.
.
open class) and what is not learned (the closed class) in the late acquisition
earnIng rate r" c properties are predictive
now return to those select' h ' cross mgUlstIcally. This said we
of English by a girl deprived of all input until after puberty (cf. Curtiss 1977). ki IOns t at may be und h '
spea mg a particular language to thei a: ' er t e control of caretakers
,: In short, the closed class seems to appear only under special conditions of NGG ' en Ousprmg
Interpreted certain of the' lr " .
input and time of exposure, while the open class appears regardless of these .of how the maternal input matched slgmfi~ant ~orre~ational effects in terms
factors (see Goldin-Meadow (1982) for a related discussion). because the commonsense vi bProcessmg biases rn the learner. This was
h iews a out the" I '
j: As for experimental manipulation of the input data, there have been some t e.g,rowth effects, We therefore looked to simp es~ mputs ' did not predict
training studies with child learners (e.g. studies of EXPANSION - the case where' learning, rather than simplicl'ty of somethmg about simplicity for
i the mother repeats, but structurally amplifies, what the child says). 'Though A '
. gain, the major example is th ' a:
grammar
' '
to I' h
exp am t e obtained findings
i results seem to be negative when the experimenter expands whatever he hears "'(' I' e meuectIveness of t I .
f SImp e, acnve, declarative) sent. ma erna Use of canonical
I the child say (e.g. Cazden 1965, Feldman 1971), some intriguing effects are ti h ences In all thre I'
cep Jng t e one relation with th h ild ' e corre atlOnal studies (ex-
found for specific structures (e.g. expansions of the auxiliary structures in effectiveness of maternal y / e c I . S use of verbs in group II) and the
I' questions and tag sentences, in studies by K. E. Nelson and his associates;
,
iary.
N
GG conjectured
es no-questIons
that initi I .'.
with '
mverte
.d ' '
subject and auxil-
L,. 72 ra POSItIOn of the auxiliary favours its
73
 CHILD LANGUAGE

THE MOTHERESE HYPOTHESIS

learning (as it might in any theory in which memory is a factor) but, moreover,
that the initial position had the extra advantages of usually being stressed and con~truct, in combination with d I .
auxIliaries. I ec aratlveS, the canonl'cal
noncon tracted. position of
While the / '
Since that time, we have laid out in detail the learning suppositions that d ' yes no-questIOns themsel d
esrgn features the child s b ves 0 not reproduce the I
would yield the special effects of stress and of noncontraction (more generally, the d eerns to e seekin (" anguage
y 0 seem the correct 0' g canonIcal form in E I' h ')
noncliticization) on language learning (see Gleitman & Wanner 1982, and ne'l nes III terms of whar n-: ,ng IS
Newport, Gleitman & Wanner, forthcoming). In brief, a variety of properties frocess~n y la~guage-specific _ biases for coll at ~Illght be quite general - not
of language learning, many of them cross-linguistic, suggest that the learner . m t e envIronment (that is attendin ~c~l~g and storing information
Items). In contrast, while the d 'I . g to IllltJal position and to stres d
is biased in the initial stage to analyse stressed syllables, and ignore the rest
f~rauxiliaries, they do not see: ~~a~;e.s do .rheproduce the canonical for:s
of the waveform; the stressed syllables leap out at the child just as, in visual b lases (that i th III wit such' C •
perception, the figures leap out from the ground. , s, ey contain aux'I·'. IlllormatlOn-handl'
child a I lanes whIch are di Illg
An apparently related effect, which appeared in NGG and is here replicated b I ppears to be predisposed not onl t me. 131 and unstressed). The
ut a so to use these to help himself b ~ 0 attend to iriitin] position and stress
in Group II, is that of maternal expansions on the child's learning of auxiliary
;cross their Unstressed and non_init:~m to. t~ack and store the relevant item:
verbs. Expansions are those maternal utterances which provide the learner
o reconstruct the canonical '. POSitIOns as well. He thereby b .
with an imitation of his preceding utterances, but with the inclusion of the
features are derived Th' PO~ltlOns from which movement d egrns
closed-class items which his own productions omit (Brown & Bellugi 1964, facts (I' . IS, we COlljecture is an int . an stress
Cazden 1965). As NGG argued, following Brown & Bellugi, expansions thus for puzz mg, taken together) that the 'Child I errrn c~aracterization of the
provide the child with the relevant closed-class information at just the point ms (the yes/no-questions more earns pnmanly FROM certain
ot~er forms (the canonical declarati:~:n t~e declaratives) but learns ABOUT
when the child's attention is likely to be focused on the appropriate
construction and the appropriate meaning. Moreover, expansions are also ''ex ummary description. The facts of chi~;rller than the yes/no-questions).1J
B posed to the varying orders of input anguage suggest that the learner
likely to present this information in a stressed form. n
Summarizing, we believe that the effects of maternal input are ONLY THOSE enVterl
&fAGarrett (1974) have named sl:t t~nces,
) ~nds one that is (as Fodor'
ce ra 10m hi h canoIllcal' ,
THAT MATCH THE PROCESSING BIASES OF THE LEARNER. There is an effect of the a fix d' d .w IC the others can be derived Th' ,or. configurationally
characteristic maternal style, to be sure, but only to the extent that i~ " he or er III terms of which the Ian .. IS canOlllcal form provides
congruent with the initial biases of the learner: how he is preprogrammed ~: tl e format f~r stating predicate-arg~:::t IStorganize~ at early stages, and
east accordmg to some ,s ructure III the langua e a
to represent the sound wave to himself. In the cases we have been discussing, transformations At th grammars, statlllg the dornai f g nd,
it is the stressed and initial material that he is inclined to represent selectively. . e surface, this ca' . am 0 movement
many other languages cl noriical ordenng is in E I' h
No objective machine, performing a straightforward manipulation and " osesr to the declarat' ng IS and
rve sentence Thi bi
analysis of the maternal input data, could be expected to make the same - . IS las towards
_ [IIJ Rather reassunngly from this '
selections.
the learner does attend perspectIve, there is one correlatl
at
Reconstruction of canonical form. WHAT is it that the child learns? Many, , 'dec/aratives: more com le some stage (that represented b ~nal effect that suggests
language development studies suggest that centrality of canonical forms is a Use the declarative m~ x ;erb forms are produced earher b~ h7duP II) to canonical
methodological ca re requently (see Table 8) H c I ren of mothers who
property of the child's interim grammars (see Gleitman & Wanner 1982, and nons we have ad d . owevej- in ac d .
must be wary in inter opte ,and stated in earhe ' cor WIth the
particularly Slob in & Bever 1982). The first to observe this property, in the should the mother's pretl~g thIS effect, plausible though it r SectIOns of this paper, We
context of early child speech, was Bellugi (1967). She observed that children rather than, for exam Use 0 declaratives affect only verb-s seem~, For o~e thing, why
we have, in the effect~e, hIS tenden~y to express all the ar ~phlstlcatlOn In the learner
not only produce declaratives in canonical orders, but also come to say
plausibility hypothesIs: ~~:rc~~:~usslOn, is one that is at b;st~~~:;1 of.predIcates ? What
questions (incorrectly)
that mark interrogation
position in the verb-phrase
in canonical order. That is, they produce questions
by intonation, but place the auxiliary in its MEDIAb
(e.g. When we can go ?).
at some POInt attendin
one we would have i: I IS,learnIng about declarative Y Interpretable on a
declaratlves, But the correlational sa so It must be that he is
correlational studies p:e d Icte?, Moreover, this effect does note ect IS ~ot precisely the
'hand, this finding d~es':nes~ IS not bolstered In terms of its reli:~~ear In the other two
At the same time, our own correlational findings suggest that the canonical, -' effects: It is at least t ,some of Our canons for the int I rty, But on the other
forms are not the sole or primary data on the basis of which these forms are, ~' supportive acqUisitlO~~I::t~llY coherent according to manye~::et~btlity of c?rrelatlOnal
g
the split-half procedure. ~ ence (e.g, from Slob in & Bever 1982 )UISt1Ctheones, the~e is
learned. For the example we have been discussing, itis the ,yes/no-questions
problematic We ca ummanzIng, the state of the 'd' and the effect surVIves
that provide the useful input data for noticing auxiliaries and beginning to . nnot evaluate I eVI ence on thi b
suggests that it deserves f h secure y whether the effect is I IS su topic is
urt er expenmental pursuu. rea, but ItS plaUSIbIlity
74
75
 CHILD LANGUAGE
THE MOTHERESE HYPOTHESIS
a canonical ordering, and hence towards declaratives, begins to explain why
elude that restrictive and non ob . .
the child preserves this ordering even for the sentence types (e.g. interrogative, b . - VIOUSpredis "
a, out Information-handlin d b POSItIOns of the learner - both
negative) that are reordered at the surface in the input speech. . g an a out langua . If
parent Inductions from th . ge Itse - rather than trans-
However, the child's quest for the canonical sentence is made difficult by e
Ianguage learning. Input corpu b
s, ear most of the burden for
the fact that certain of the relevant materials (e.g. auxiliary verbs) appear in
most input strings in a form that imposes a burden on language perception:
namely, in unstressed syllables and in medial positions. In early stages of
learning, such unstressed and medial items are therefore absent altogether. In Bach E '. REFERENCES
subsequent stages, the learner makes an apparent detour in his learning B ' . (1970). Problommaltzation LI I
strategies: he focuses his attention on certain noncanonical forms (e.g.
arn~s, S., G~tfreund, M., Satterly: D. & ~~~:. G
which predict children's language d I ,. (1983). Characteristics of adult spee h
yes/no-questions) that present these burdensome materials in ways he can Bellugi, U. (1967) The . , . eve opment.JChLang 10.65-84 c
. . acquIsItion of neg ti U· .
readily perceive, that is, initially and with stress. Eventually, then, by . University. a IOn. npubltshed doctoral dissertation H d
B di , arvar
integrating information from the canonical and noncanonical forms, he is able ene rct, H .. (1976). Language comprehension in 1
doctoral dissertation Yale U· . 0-16 month old infants Unpubli h d
to include these materials as well in his own canonically ordered speech. Bickerton D (I ' . ruverstty. . IS e

Bloom L' (19'739)705).Dynadmlcs of a creole system. New York: CUP

,. . ne Wor at a time, The Ha ue' . . .
Bradley, D. C. Garrett M F & Z . g. Mouton.
D ' '" urif, E. G. (I 7 ) S . '.
GENERAL CONCLUSIONS
.' Caplan (ed.), Biological studies of mentalp 9 9· yntactlcdeficlts in Broca's aphasia. In
Brame, M. D. S. (1976) Ch'ld 'fi rocesses. Cambridge Mass M I T
. I ren s rst W d bi ". . .
The Motherese Hypothesis, as usually stated, is that the way the caretaker No. 164. or com matlOns. Monogr. Soc. Res. Ch. Devel.
talks plays a causal role in acquisition. In a general sense this must be so, for Bresnan, J. (1978). A realistic transformational
Miller (eds), Li1lguistic theory and psycho! . ran;mar. In M. Halle, J. Bresnan & G A
it is the only explanation for the fact that language learning is variable - that M (ed.). (1982). The mental representati:!'~;
.LT.
rea tty, C:ambridge, Mass., M.LT.
grammatIcal relatlom. Cambridge, Mass.:
' .
French children learn French and Turkish children learn Turkish. What is
not clear are any details of such a position, for example, what the Hypothesis Brown, R. & Bellugi, U. (1964) Th .
HarvEdRev 34. 133-51. . ree processes rn the child's acquiSitIOn of syntax
asserts about HOW the environment exerts its effects, and the extent to which Brown, R. & Hanl C ( , .
h
properties of the learner himself modulate or reorganize the information
on, . 1970). Derivational com Ie .
C speec . In J. Hayes (ed.), Cogm'tion and the d t xrty and order of acquisition in child
provided in the environment. For the hypothesis to be anything but question- azden, C. B. (1965). Environmental . eve opment of language. New York: Wiley
U bli h assIstance to th h ild ' .
Ch npu IS ed doctoral dissertation Harvard U. . e CIS acquisition of grammar
begging (i.e. to be anything but the claim that whatever input turns out to ornsky, N. (1965). Aspects of the /heor 0 flIverslty.. .
support learning is the' best' or 'simplest '), one must state in an explicit --
--
(1975). Reflections on language. Ne:
(1981). Lectures on government ad'
!o:{~~x. Cambridge, Mass. M.LT.
. . andom House.
fashion what kinds of linguistic environments aid what kinds of learning g
Chomsky, N. & Halle, M. (1968) Th; so:;;l1l · Dordrecht: Foris Publications.
procedures (for such discussions, see the many interesting articles in Baker
Cross: T. G. (1977). Mothers' sp~ech ad'ustr!:attern of Engli~h. New York: Harper & Row.
and McCarthy 198 I). In the absence of such explicit proposals, the claim that vanables. In C. E. Snow & C A F J ents. the contributIOn of selected child r t
. . . . . erguson (eds) T lk' IS ener
certain properties of maternal speech explain the learning seems suspiciously C acquIsItIOn. Cambridge: C.U.P. ' a l1lg to children,' language input and
like affirming the consequent. urnss, S.. (1977). Genie,' a psycholin uistic s
A~ademlc Press. g tudy of a modern-day 'wild child'. New York'
With these provisos in mind, we can nevertheless make some preliminary Donan, N. D. (1978). Thefate ofmor holoo i .
conjectures from the correlational studies reported in this paper, and sup- Dougherty, R. C. (1969). An interpr:ti:e °tt::! complexity i~ language death. Lg 54.590-609.
portive literature from developmental psycholinguistics. These suggest, as Feldman, C. (1971). The effects of variou t y of pronommal reference. FL 5. 488-519
F °lf two to three year-olds. Unpublishe; :a~:s 0i;d.ult responses in the syntactic acquisiti'on
we have discussed at some length above, that while language is learned, edman, H., Goldin-Meadow S & GI . r, ruversuy of Chicago.
through experience with the environment, its ultimate character is materially of lb" ,. ertrnan, L. R (1978) B d
. anguage y Imgulstically deprived deaf h ild . I . eyon Herodotus: the creation
an effect of the learner's own dispositions as to how to organize and exploit F gdesture,'the emergence of language. New Yor~!A re~. ~ A. Lock (ed.), Action, symbol, and
o or, J. A., Bever T G & G M . ca ermc Press.
linguistic stimulation. The major correlational findings supporting this view . ..' '. arrett,. (197 ) Th
F to psycholl1lgulstlcs and generative grammar ~. ..; P,:chology of language,' an introduction
have to do with the facts that the child is selective in WHAT he uses from the urrow, D., Nelson, K. & Ben di t H ( . ew or: McGraw-HilI.
de I . e IC, . 1979). Mothers' spe h hil
environment provided; he is selective about WHEN in the course of acquisition ve opment: some simple relationships JChL 6 ec to c I dren and syntactic
Garrett, M. F (197 ) Th . . ang. 423-42.
he chooses to use it; and he is selective in what he uses it FOR (i.e. what . . 5 . e analysIs of sentence d .
G leadrmnG'l! and motivation, Vol. 9. New York' A~:~ u~tl~n. In G. Bower (ed.), Psychology of
grammatical hypotheses he constructs from the data presented). We con- az ar, . (1981). Unbounded de . . ermc ress.
pendencies and coordinate structure LI I~ 8
76 . ~. 155- 4.
, 77
 CHILD LANGUAGE
THE MOTHERESE HYPOTHESIS

Gentner, D. (1982). Why nouns are learned before verbs: linguistic relativity vs. natural. Shipley, E. F., Smith, C. S & Gleitman L R ( 6) A .
partitioning. In S. Kuczaj (ed.), Language development: language, culture, and cognition. free responses to co .d L ,.. 19 9· study In the acquisition of language:
Hillsdale, N.J.: Erlbaum. Slobin D I mman s; g 45· 322-42.
Gleitman, L. R. (1981). Maturational determinants of language growth. Cognition 10. 103-14. ' F' . (1973)· Cognitive prerequisites for the development of I
C . A . erguson & D I Siobin (ed ) St di if hild grammar. n
Gleitman, L. R. & Wanner, E. (1982). Language acquisition: the state of the art. In E. Wanner Rinehart & Winston·. . s, u tes 0 c I language development. New York: Holt,
& L. R. Gleitman (eds), Language acquisition: the state of the art. New York: C.U.P.
-- (1977). Language change in childhood and in hist I
Goldin-Meadow, S. (1978). A study in human capacities. Science 200. 649-S1. learning and thought New York. Ac d . P ory. n]. Macnamara (ed.), Language
-- (1982). The resilience of recursion: a study of a communication system developed without .' . a ernic ress.
-- (1982): Universal and particular in the acquisition of Ian ua
a conventional language model. In E. Wanner & L. R. Gleitman (eds), Language acquisition:
L. R. Gieltman (eds), Language acquisition: the state of the art. Jew g;. ~n (~\J~nner &
the state of the art. New York: C.U.P.
Hoff-Ginsburg, E. & Shatz, M. (1982). Linguistic input and the child's acquisition of SIObi~InDprIes&s)·BCross-TlingGuis(tiC
, .. ever
8study01
'9 2) Ch Id
child language acquisition. Hillsda~; N.]·.: E;lbaum
. , .
language. PsychBuli. study of word ord:r a~d inflecti~ns 'c ren. use canonical sentence schemas: a crosslinguistic
Jacobs, R. A. & Rosenbaum, P. S. (1968). English transformational grammar. Waltham, Mass.: Snow C E ( . ognttton 12, 229-6S.
, . . 1972a). Young children's responses to adult .
Ginn. Paper presented to the International Congress of A r ;~tences of vCarYlngcomplexity.
Kean, M-L. (1979). Agrammatism: a phonological deficit. Cognition 7.69-84. -- (1972b) M thers' h trr ch i pp ie IngUIstICS, openhagen
Landau, B. & Gleitman, L. R. (forthcoming). Language and the vocabulary of perception in S . 0 ers speec to chIldren learning language. ChDev 6 .
now, C. E. & Ferguson, C. A. (eds) (1 ) T lk· . 43· S49- S·
a blind child. acquisition. New York: C.U.P. . 977· a l1lg to chz/dren: language input and
Lasnik, H. (1976). Remarks on coreference. Ling Anal a, 1-22. Wanner, E. & Maratsos M (1978) A ATN
Marin, 0., Saffran, E. & Schwartz, M. (1976). Dissociations of language in aphasia: impli- '...
] . Bresnan & G A MIller n approach to comprehension In M H II
(ed) L· .. h . . a e,
cations for normal function. AnnNY AcSci 280. 868-84. Mass: M.LT. . . s, tnguistu; t eory and psychological reality. Cambridge,
Morgan, J. & Newport, E. L. (1981). The role of constituent structure in the induction of an Wexler K. & Culicove P ( 8) Formal ori .
artificial language. JVLVB 20. 67-8S. M.LT. r, . 19 o. orma prl1lClples of language acquisition, Cambridge, Mass.
Nelson, K. E. (1976). Facilitating children's syntax. DevPsychol 13. 101-7. Zwicky A M ( 76) 0 I··
Nelson, K. E., Carskaddon, G. & Bonvillian, }. (1973). Syntax acquisition: impact of ex- ~ . : '9 . n c ItICS. Paper presented at the T'hi d Inter nari I
Meeting, VIenna. ir n er natrorra Phonology
perimental variation in adult verbal interaction with the child. ChDev 44· 497-S04.
Newport, E. L. (1977). Motherese: the speech of mothers to young children. In N. J. Castellan, I

D. B. Pisoni & G. Potts (eds), Cognitive theory, Vol. 2. Hillsdale, N.}.: Erlbaum.
-- (1981). Constraints on structure: evidence from American Sign Language and language
learning. In W. A. Collins (ed.), Minnesota symposia on child psychology (Vol. 14). Hillsdale,
N.J.: Erlbaum.
-- (1982). Task specificity in language learning? Evidence from speech perception and
American Sign Language. In E. Wanner & L. R. Gleitman (eds), Language acquisition: the
state of the art. New York: C. U .P.
Newport, E. L. & Gleitman, H. (1977). Maternal self-repetition and the child's acquisition
of language. PRCLD 13· 46-SS.
Newport, E. L., Gleitman, H. & Gleitman, L. R. (1977). Mother, I'd rather do it myself:
some effects and non-effects of maternal speech style. In C. E. Snow & C. A. Ferguson (eds),
Talking to children: language input and acquisition. New York: C. U.P.
Newport, E. L., Gleitman, L. R. & Wanner, E. (forthcoming). The prosodic input to language
learning.
Newport, E. L. & Supalla, T. (1980). The structuring of language: clues from the acquisition
of signed and spoken language. In U. Bellugi & M. Studdert-Kennedy (eds), Signed and
spoken language: biological constraints on linguistic form. Dahlem Konferenzen.
Weinheirri/Deerfield Beach, Fla./Basle: Verlag Chemie.
Peters, A. (1981). Language typology and the segmentation problem in early child language
acquisition. Proceedings of the Berkeley Linguistic Society, 7.
Pinker, S. (1979). Formal models of language learning. Cognition 7. 217-83.
Pinker, S. & Lebeaux, D. (1982). A learnability-theoretic approach to children's language.
Unpublished manuscript, Stanford University.
Sachs, J. & Truswell, L. (1978). Comprehension of two-word instructions by children in the
one-word stage. JChLang 5· '7-24.
Sankoff, G. & Laberge, S. (1973). On the acquisition of native speakers by a language. Kivung
6. 32-47.
Shatz, M. & Hoff-Ginsburg, E. (in prep.) Effects of differentially-enriched input on the
acquisition of auxiliary structure.

79