Cognirion.

40 (1991) 21-81

Syntactic subjects in the early speech of

American and Italian children*
Virginia Valian
Departmerct of Psychology, Hunter College. 695 Park Avenue, New York. #NY 10021. U.S. A.

Received January 10. 1990. final revision accepted January IS. 1991

Abstract
Valian. V.. 1991. Syntactic subjects in the early speech of American and Italian children. Cognition.
40: 21-81

Why do young children leave out sentential subjects? Two competence-deficit

hypotheses and a performance-limitation account are evaluated in the present set
of studies. American children appear to understand that English requires subjects
before mean length of utterance (MLU) 2.0. On balance, performance factors
account for the data best. Natural conversations between 21 American children
(ranging in age from 1;lO to 2:s and in MLU from I .53 to 4.38) and their mothers
were taped, transcribed, and analyzed to determine when American children under-
stand that English requires subjects. We measured the frequency of subjects (Study
I); types of pronominal subjects, including expletives (Study 2); frequency of mod-
als and semi-auxiliaries (Study 3); frequency of infinitival to, past tense, third
person singular, and subordinate clauses (Study 4); length of verb phrase, fre-
quency of different types of verbs, and frequency of direct objects (Study 5). For
Studies I and 3 we also used, for comparative purposes, transcripts of 5 Italian
children, taped monthly for a year. Even our lowest-MLU American group (5
children between 1.5 and 1.99) used subjects and pronominal subjects more than
twice as often as the Italian children, and correctly case-marked their subjects. The
American children also produced examples of all the sentence elements measured.

*This research was supported in part by a grant from the National Institute of Child Health and
Human Development (HD-24369). and in part by a grant from The City University of New York
PSC-CUNY Research Award Program. My thanks for helpful comments and criticisms to P. Bloom.
R. Fiengo, L. Gerken. J. Katz, W. O’Grady. M. Potter, A. Radford. and the anonymous readers.
Rarely can an author have been so fortunate in her reviewers! I am grateful to S. Utakis for her
excellent assistance on this project, and to L. Brighenti for her skillful translations and data analysis
of the Italian transcripts. My continued thanks to the parents and children whose conversations form
the basis of this study.

OOlO-0277/91/$18.80 0 1991 - Elsevier Science Publishers B.V. All rights reserved.

Introduction

Children’s very early speech has long been known for its “telegraphic” character
(Brown R: Fraser, 1963). Young 2-year-olds leave out. or only inconsistently
include, a variety of constituents in their utterances, among them subject noun
phrases (NPs). determiners (Dets - “a”, “the”). modals (-‘can”, “will”, “may”.
-*must”), the copula “be”, and so on. Such omissions have received great atten-
tion and analysis over the past thirty years, especially in the 1970s (Antinucci 8:
Parisi, 1973: Bloom, 1970: Bloom, Miller. & Hood. 1975: Bowerman. 1973;
Braine, 1974; Brown. 1973: Greenfield & Smith. 1976; Valian, 1956 - for a review
see Bowerman. 1978). Nevertheless, our understanding of why children’s early
utterances are short - why constituents are missing - is still rudimentary. The
present set of five studies is intended to shed light on this question by focussing
on sentence subjects. which are often missing in young children’s speech.
There are two basic sorts of explanations of missing constituents: competence
explanations and performance explanations. A competence explanation is that
children’s early grammars do not contain the (adult) rules or structures that would
produce utterances containing the necessary elements. A performance esplana-
tion is that children operate under processing limitations that restrict the lengths
of utterances that they can produce. In the former case, children do not X-now
that subject noun phrases, say, are required. In the latter case, children know
that they are required, but do not successfully apply their knowledge in every
utterance.
Competence and performance explanations need not be mutually exclusive.
Both the child’s competence and her performance could be deficient. Further,
competence and performance factors might operate to different degrees for differ-
ent constituents. A child could leave out some constituents because she does not
know they are required, and leave out others to reduce processing demands. Our
principal aim here is to present a fuller picture of children’s usage of subjects
than has been available from previous work: another is to determine when Ameri-
can children understand that subjects are obligatory and to evaluate competing
explanations for children’s inconsistent subject use; a third is to examine the
interaction of competence and performance factors in acquisition.
It is necessary to assume some linguistic description of the adult grammar that
the child is acquiring. Most of the competence explanations that will be examined
have assumed a version of modern transformational grammar, called government-
binding (GB) theory (Chomsky, 1981, 1982), and that is therefore the formalism
adopted here. The formalism is used here both as an approximate description of
the adult grammar, and as a constraint on possible hypotheses about the child’s
early grammar. Thus, descriptions of the child’s grammar that would ‘violate
assumptions of transformational grammar are not entertained. At the same time,
however, we explicitly do not adopt the language acquisition mechanism that is
often associated with the formalism (parameter-setting).
 In this introduction we first review facts about the nature of subjects in differ-
ent languages. and then consider different language acquisition accounts. The
reader who wishes to have an overview of the findings can read the summary
provided at the end of each study.

Lnnguage facts and theories

Children’s use of subject noun phrases provides a special opportunity to investi-

gate both inconsistent usage of obligatory elements, and acquisition of a funda-
mental piece of information about one’s language - whether or not overt subjects
are required. English requires an overt subject in tensed clauses. (The subjects
of untensed clauses. or infinitivals, need not be overt, as in “I want to go”, where
no overt subject appears for “to go”. Our discussion will be concerned solely with
tensed clauses.)
In English, (1) is grammatical while (2) is not allowed:

(1) I am a good kid

(2) * Am a good kid
In contrast. in Italian, both (3) and (4) are allowed (abstracting away from the
error of the missing determiner):
(3) 10 sono bravo tato
I am good kid
(4) Sono bravo tato
Am good kid
There can be an empty or “null” subject in Italian. English and French are among
the languages that require overt subjects; Italian and Spanish are among the
languages that allow null subjects.
Parameters of language represent linguistically significant dimensions of lan-
guage, within which there is very narrow variation (Chomsky, 1981, 1982). The
null subject parameter refers to the dichotomy between those languages which
require overt surface subjects and those which allow null subjects. The null sub-
ject parameter is now thought to involve two different parameters: a licensing
parameter and an identification parameter (Jaeggli & Safir, 1989; Rizzi, 1986).
The licensing parameter refers to the presence or absence of elements within
the inflectional system. called INFL. short for INFLection. INFL contains tense
and agreement features, and, in English, modals; it is the new-style version of
Aux (see Radford. 1988 for background). If the necessary element in INFL is
present, null subjects are allowed: if absent, they are not. Exactly what the INFL
element is which licenses null subjects is unknown (see. for discussion, Jaeggli &
Safir, 1989: Rizzi, 1986). It is the licensing aspect of null subjects which will be
our principal concern here.
 If a language has the element which licenses null subjects. there are then
different possibilities for how the subject will be identified. that is. how the person
and number of the subject will be determined. The particular reference of the
null subject is thought to reflect some sort of agreement process. though the type
of agreement will vary amon, 0 languages (Jaeggli & Safir. 19SY: Rizzi. 1956).
The null subject parameter in principle also controls other language features
in addition to whether an overt subject is required (for a list of candidates, see
Riemsdijk & Williams, 1986, pp. 298-303). For example, languages that allow
null subjects typically do not have “expletive’*. or referentially empty. pronouns
(though there may be exceptions). An example, in English, of a referentially
empty pronoun is the “it” in (5). In Italian, such a non-referential form of “it”
does not exist. The only equivalent of (5) in Italian is (6):

(5) It seems that Jane loves Mary

(6) Sembra the Jane ama Mary
Seems that Jane loves Mary

Expletive ‘*it” also occurs in “weather” expressions: “It’s raining”. “It’s snowing”,
and so on.
The null subject itself is a form which is abstractly present, but not pro-
nounced. That is, the subject position is not empty: it is filled, but with a form
lacking phonetic content. That form has the character of a referential pronoun.
Null subjects are usually held to occur only in tensed clauses: the absent subjects
of infinitives are considered to be a different empty form - one which is typically
anaphoric in character (e.g., Jaeggli & Safir, 1989, but see also Borer. 1989, and
Huang. 1989, for different proposals). The null subject of a tensed clause is called
pro (pronounced “small pro” or “little pro”), while the null subject of an infinitive
is called PRO (pronounced “big PRO”); see Chomsky (1982) for first mention
of pro.

Acquisition theories: Two types of competence accounts

Any acquisition theory must specify both a mechanism of grammatical change

and the hypothesized nature of the child’s grammar before and after the change.
Grammar-based approaches to acquisition are especially useful because they pro-
vide details about the child’s formal representations. Such details amount to op-
erational definitions of theoretical claims about a child’s knowledge. For example,
the statement that a child has an optional subject cannot be evaluated unless we
know what grammatical form that optionality takes: we need to know what men-
tal representation is proposed.
Grammar-based approaches tend to issue in competence-deficit accounts. The
empirial analyses we present will bear on two broad classes of competence-deficit
accounts. The two differ considerably in their theoretical explanations but overlap
to a large extent in their predictions about acquisition.
 The pro hypothesis
In one class there is an empty category, pro, underlying the superficial absence
of subjects (Hyams, 1986, 1987). We will refer to this as the pro hypothesis. On
the pro hypothesis, subjects are not optional: all sentences have subjects. and
some of those subjects consist of the unpronounced pronominal pro. The child’s
productions look as if the subject is optional. but there is an underlying abstract
subject.
Hyams’s initial pro hypothesis (1986) provided both a linguistic analysis of the
null subject parameter and an account of several features of the acquisition of
English and Italian. The core of the account was that all children begin acquisition
with the null subject value of the parameter. Children in non-null subject lan-
guages like English have to reset the parameter: before they reset the parameter
their speech will grammatically resemble the speech of null subject language
children, such as Italians.
Hyams (1986) linked the often-noted inconsistent usage of subject noun
phrases in American children’s early speech to a concurrent absence of expletive
subjects, modal verbs, and certain forms of be. Consistent subject usage, and use
of expletive subjects, modals, and be, were predicted to appear at roughly the
same time in the child’s productions. The mechanism whereby the parameter was
reset was the child’s coming to notice expletives, which would then force a switch
to the other value. Previously collected acquisition data appeared to demonstrate
that subjects, expletives, and modals were indeed linked. In a different version
of the pro hypothesis, Hyams (1987) likens the American child to the Chinese
child rather than the Italian child. Since that version is less well developed than
the initial pro hypothesis, we concentrate on the first version.

The VP hypothesis
In another class of explanations the child has an immature, or incomplete,
grammar (Guilfoyle, 1984; Guilfoyle & Noonan, 1989; Kazman, 1988). We will
refer to this as the verb phrase (VP) hypothesis, because the grammar lacks an
INFL phrase (and a COMP - short for complementizer - phrase), and consists
only of a VP, and the lexical phrases that can appear within a VP (NP, prepos-
itional phrase (PP), and adjective phrase (AP)). Figure 1 shows a tree diagram
of the presumed adult representation, with the hypothesized initial child represen-
tation circled.’
For both child and adult, the VP has an optional SPEC (short for specifier)
node plus a V’ (read as V-bar; see Radford. 1988, for an introduction to X-bar

‘The VP hypothesis has been proposed within the transformational framework. Although it might
appear to countenance violations of universal grammar. the spirit of the proposal is that not all parts
of universal grammar are immediately available to the child. The grammar only commits sins of
omission, not commission.
26 I’. Vulian

Figure 1. Tree diagram .shobtYng adult English wderlying .structare ami h_vpothesi:ed child
substtwttlre (circled). CP = complement phrase und corrrspond.r to old-style
s” (read as S-double bar). Each X phrase has the structwe of optional SPEC
(specifier) and x’ (read as X-bur). SPEC is not itself a category, bat a grammat-
ical frrnction, and hence is not sh0rc.n in the tree. The boxed positions in the
tree are SPEC positions. Each X’ has the structure of an obligator_v head and
optional complements. CP thus consists of an optional SPEC and C’: C’ con-
sists of the head, COMP, and its complement, IP. IP = INFL linflection)
phrase. and corresponds roughly to old-style S; IP consists of an optional SPEC
and I’. I’ consists of the head, INFL. and its complement, VP. I,VFL includes
tense and agreement markers, and, optionally, medals: it corresponcls rough!\
to old-style Aux. VP = i*erb phrase and consists of an optional SPEC. here
filled with a “subject” NP, and v’. If’ consists of V and optional now? phrase
(NP). (The NP notation is used rather than the determiner phrase notation fbr
ease of exposition.) The subject originates under SPEC of VP and moves. ill
the adttlt grammar. to SPEC of IP. as shoHw by the arrobt’.

(NP) /
syntax) node. VP subjects are inserted under the SPEC node. Since the child’s
grammar consists of only a VP, and since SPEC is optional, a VP subject may
or may not be present. The V’ node in turn has an obligatory V node and an
optional direct object NP node. On the VP hypothesis there are no genuine
sentential subjects in early child speech - only subjects of VPs: those VP subjects
are optional. When the remainder of the grammar develops, and INFL is present.
genuine subjects which are appropriately cased as nominative can appear.
In the adult grammar for English. the VP subjects will be raised to become
subjects of the INFL phrase (as shown by the arrow in Figure 1). and will receive
nominative case through a tense or agreement element in INFL. In English. other
features of the grammar will force an NP into SPEC of VP. In other languages.
raising into IP (INFL phrase) may not be obligatory, because there are other
ways in those languages that features in INFL can assign case to subjects. The
VP hypothesis predicts absence of tense. infinitival to, and medals, and lack of
nominative case-marking.
O-Grady, Peters, and Masterson (19S9) have proposed that the child initially
does not have tense, and therefore cannot distinguish between situations when a
subject is required (before tensed verbs) and situations when it is not (before
some infinitives). Their linguistic framework, unlike the one we are assuming.
allows for sentences without explicit subjects (O-Grady, personal communica-
tion). They report that the three children whose corpora they analyzed began
using subjects consistently when they began using the past tense productively.
Like the VP hypothesis, then, O’Grady et al. predict a relationship between
subject use and tense. Unlike the VP hypothesis, however, no relationship is
predicted between subject use and medals.

Variable rules
An early account of inconsistent subject use, proposed by Bloom et al. (1975)
considerably before the recent work on the null subject parameter, was a model
in which the probability of producing a constituent was a function of factors like
the familiarity of the verb being used, the inclusion of a negative marker in the
utterance, and certain discourse features. Since Bloom et al. say little about the
formal representation underlying the children’s productions, it is not clear
whether the probability model proposes a competence deficit. The formal rep-
resentation they sketch shows an obligatory subject, thereby seeming to rule out
a competence deficit. In addition, recent comments by Bloom (1991) suggest that
a performance-oriented model is intended: “children will omit the subject when
their cognitive processing abilities are exceeded, for example, when they use new
verbs, nouns or pronouns; or add negation or attribution to the sentence”.
On the other hand, Bloom, Miller, and Hood (1975) present the model as one
which incorporates a variable rule, suggesting a competence deficit. On such a
model, the child knows that there are such things as subjects. but does not know
that they are obligatory. The proposal would differ from the pro hypothesis
because the abstract element pro is not part of the child’s grammar. It would
differ from the VP hypothesis because the child is not limited to a VP represen-
tation. Instead. the child has a full “English” grammar, including genuine sub-
jects. but the child thinks that those subject NPs are optional.
Within the formalism of present-day transformational grammar which we are
assuming as a constraint on the child’s grammar, a variable rule for subjects is
impossible: subjects are mandatory. The only option within our current
framework is for the child to believe that subjects are required, but that they can
receive a null pronunciation (as may sometimes actually be the case with deter-
miners). A subject NP node is always present, as in Bloom et al.‘s (1975) sketch.
but the child believes that a subject NP can be morphophonemically empty. The
error is not the same as the pro hypothesis, because the abstract element pro -
with its pronominal properties - is not part of the child’s grammar. According to
the pro hypothesis the child has an abstract pronominal which lacks phonetic
content: according to the no pronunciation interpretation of the variable rules
approach, the child inserts no lexical item under the subject NP node. Radford
(1990) specifically considers a very similar proposal.
The factors Bloom et al. (1975) propose as determinants for subject use are
then factors nhich determine when the subject NP receives a lexical spell-out:
the child’s formal representation would or would not lexically spell out the subject
NP, depending on the interaction of the performance factors. The factors Bloom
et al. investigated, ranging from word familiarity to discourse features, were
related in a complicated way to the likelihood of producing an utterance of a
given length, as measured by number of major constituents. Most analyses did
not specifically concern subjects, but the likelihood of producing two- versus
three-constituent sentences.
One important analysis concerned how “complexity” elsewhere in the sentence
affected utterance length. Some types of complexity, such as use of a negative.
verb + particle construction, or the possessive, were more common in two- than
three-constituent utterances. On the other hand, other types. such as the use of
verb and noun inflections, medals. semi-Aux’s. and some determiners, were as
common in three- as two-constituent utterances. Interestingly, verb inflections
were more common in subject-verb combinations than in verb-object combina-
tions.
There are no obvious empirical predictions which would distinguish Bloom et
al.‘s (1975) model, which we may term the “null spell-out” alternative, from a
pure performance deficit model. Any findings which support a performance de-
ficit model will also support a null spell-out model. Therefore, even if the other
competence-deficit hypotheses can be ruled out, we will not be able to choose
between the null spell-out competence deficit and a pure performance deficit.
 Swtuctic .wbject.s in enrly speech

Others
There are other competence-deficit hypotheses which do not fit so neatly into
our two classes, such as Lebeaux (1957). Roeper and Weissenborn (1990). and
Radford (1990). Since the major empirical predictions have been made by the
pro hypothesis or the VP hypothesis, we concentrate on them here.

Relation between data and theory: Limitations of competence-deficit accounts

Weak evidential basis

Competence-deficit theories have been limited by the weakness of the data
they have had recourse to. As is proper, acquisition theorists tailor their accounts
to fit the reported facts of acquisition. But. surprisingly. given the abundance of
theoretical accounts of children’s early use of subjects. we lack normative quan-
titative data on the development of American children’s use of subjects, and of
related constituents. Existing accounts use either summary data reported in other
contexts, or data from very small samples of children (and the Brown corpora
have been used repeatedly). It is often difficult to know how common a reported
pattern is.’
Before theory development can progress. and before current theories can be
winnowed, we need data from a large number of children in a variety of languages
(and replications by other investigators with new samples of children) on the
quantitative development of subjects and other syntactic elements.’ We need to
know exactly what the phenomenon is that we are trying to account for.
As a first step in that process, we measure American children’s use of subjects
across a broad mean length of utterance range (MLUs 1.53-4.38). plus their use
of the sentential elements which the two principal competence-deficit hypotheses
have implicated as linked to knowledge of subjects. We also compare American
and Italian children’s production of subjects.

‘Many authors have relied on data presented in Bellugi (1967). even though those data represent
summaries from fairly widely separated time periods for three children (and for some measures data
from only one child are presented). A second source is data from Bloom (1970). and Bloom, Light-
bown, and Hood (1975). Bloom (1970) includes examples of many types of utterances of 3 children
studied longitudinally, and Bloom, Lightbown, and Hood (1975) include in their appendix examples
of utterances from Bloom’s 3 children plus another child. Presentation of the full corpus for each
child (a total of over 27,000 utterances) would obviously have been impossible, and unnecessary for
those authors’ purposes. But the presentation of examples rather than a systematic sample makes the
published data inappropriate as a source of data for analyzing subject use. Finally, Radford (1990)
collected 39 cross-sectional samples, and 12 longitudinal samples - a huge corpus. Unfortunately. he
presents almost no quantitative data, so his data cannot be compared with ours.
The present study points up the need for investigators to sample large numbers of children, to
sample children who are not part of existing databases, and to provide numerical and proportional
tabulations of usage of different constituents. Otherwise, the idiosyncrasies of individual children’s
patterns may be enshrined as typical of language development, and infrequent examples may be
accepted as the norm.
 The basic fact from kvhich all the theories have started is children’s earl!
inconsistent use of subjects. Competence theorists have interpreted this inconsis-
tent usage as a competence deficit: American children do not know that English
requires explicit subjects. either because they improperly have pro in their gram-
mar (pro hypothesis) or because they have an immature grammar (VP
hypothesis). O-Grady et al. (19S9) used a criterion of 90% use of subjects to
classify children as having an obligatory surface subject. Hitherto. no criterion
had been proposed, so that their use of one represented an advance of the field.
On the basis of the 90% criterion they concluded that there was a stage of
optional subjecthood. since the three American children they examined (two from
the Brown corpora) did not start out using subjects 90% of the time.
The problem with the 90% criterion. initially developed by Brown (1973) for
other syntactic elements, is that it is probably too strict.j While it is safe to say
that a child who uses subjects in 90% of the required contexts understands that
overt subjects are obligatory, it is not equally safe to say that a child who uses
subjects less than 90% of the time lacks that knowledge. The lower the child’s
MLU, and the lower her level of development generally. the more difficult it will
be for her to express her knowledge fully. and the more likely it is that her usage
will be inconsistent. Inconsistency is not in and of itself a hallmark of a compe-
tence deficit. Therefore, we take it as an open question. and a theoretical ques-
tion, whether the American child ever thinks that overt subjects are not required.

Ancillary assumptions
A second limitation of competence-deficit accounts is that, on some matters,
as Hyams (1987) notes, linguistic theory is mute, and ancillary assumptions must
be brought in to derive predictions. The example Hyams (1987) gives is of the
initial setting of the parameter: linguistic theory is neutral with respect to which
value should be the initial value (if any). In certain cases (as with medals, dis-
cussed below) predictions are derived from the additional assumptions rather than
from the linguistic analysis.

‘Brown (1973) reported on the points at which 3 children produced various morphemes in 90%
of the contexts in which they were obligatory in 3 successive taping sessions. The criterion has often
been used as if it represented knowledge onset. But the children clearly had knowledge of the
morphemes in question considerably before they could supply them in 90% of required contexts. For
2 of the children. 5 of the 14 morphemes examined had not reached criterion by MLU 1. and for
the third child S out of 14 morphemes had not reached criterion by MLU 1. MLU 1 is quite an
advanced point in acquisition. We therefore think Brown’s morpheme data are an inappropriate
source for determining knowledge onset. In addition. given the much larger sample examined by de
Villiers and de Villiers (1973). their morpheme data are preferable.
 Swtactic sttbjects in early speech 31

Acquisition theories: Performance accounts

Under the heading of “performance factors” we will include a large variety of

variables, representing in some cases limitations (as in limited working memory),
and in others merely performance characteristics (as in prosody and discourse
factors).

Performance limitations
Children, being human, are variable and work under performance constraints.
One production constraint is memory. Young children have a smaller working
memory than do adults, and their memory span is correlated with their MLU
(Blake, Quartaro. Austin, & Vingilis. 1989). Chi (1975) has claimed that chil-
dren’s inferior memory is due to their lesser ability to form chunks (rather than
to capacity limitations): children can chunk fewer items together into a unit than
an adult can, because they have less experience with most domains than adults
do. Thus, Chi links lack of knowledge and experience with smaller memory.
Olson (1973) attributes children’s poorer memory to ‘-their failure to organize,
plan, monitor, and integrate their information processing and remembering as
effectively as older children or as adults . ..” (p. 151). It is clear that a 2-year-old
has much less expertise with the various sequential tasks of talking and listening
than an adult has, as well as less grammatical knowledge, and is therefore likely
to form smaller chunks than an adult. Children’s lovver efficiency will result in
their producing shorter utterances overall and shorter constituents within those
utterances.
In the case of adult speech we know that the length of utterances is due to
extra-syntactic performance factors. Our grammars allow us to produce very long
utterances. But the combination of planning the content of what we have to say,
finding and organizing the syntactic structures to express the content, finding the
words, taking into account the listener’s memory limitations, being a good conver-
sational partner, and so on, all conspire to limit the length of our productions.
Children have the same tasks, and much less practice at integrating them, so they
will have at least as many constraints as adults do. Thus, it is highly likely that
children’s performance system is also a limitation on the length of their utter-
ances, and the younger the child the greater the limitation.
It is clear that length limitations alone cannot predict which constituents the
child will fail to include. It will be necessary to couple length limitations with
other determinants, such as the content of the message the child wants to convey,
syntactic requirements, and discourse requirements, to arrive at the systematic
omission of function words compared to content words.
Bloom (1970) argued that some of children’s omissions represented reductions
of elements that were present in deep structure, not all of which could be ex-
pressed in surface structure, because of performance limitations. Although Bloom
presented reduction as a transformational rule (which would make it part of
competence). it seems more in keeping with her argumentation to regard reduc-
tion as a performance process. Bloom (1990) has presented data from three
children (the Brown corpus) supportin, 0 a performance explanation for children’s
early inconsistent use of subjects. The children’s verb phrases were longer when
a subject was absent than when it was present. as would be expected if children
were operating under performance limitations.
There are also data suggesting that the child’s processing load is higher at the
beginning of an utterance than at the end. NPs consisting of a determiner. adjec-
tive, and noun are infrequent in children between MLUs 2.93 and 4.13, and occur
only as objects for children below MLU 3.5 (Valian. 1986). That asymmetry
suggests that objects are easier to elaborate than subjects, perhaps because plan-
ning the utterance is more effortful at the beginning of the utterance (Pinker,
1984). and perhaps because English is right-branching (Lust & Chien. 1984).
Mazuka, Lust, Wakayama, and Snyder (1986) also invoke processing factors. On
the basis of Japanese children’s data, as well as reports of English-speaking chil-
dren’s data, they analyze children’s omission of subjects as due to sensitivity to
the principal branching direction of their language plus processing constraints.
It is. however, also likely that pragmatic factors play a role in the distribution
of NP types. Any NP containing a noun is more common as an object than as a
subject at every MLU observed, while NPs consisting of a pronoun are more
common as a subject than as an object (Bloom. 1990: Limber, 1976; Valian. 1986.
Bloom, Lightbown &r Hood, 1975, found this pattern to be somewhat variable at
low MLUs). The heavy use of pronominal subject NPs may thus reflect two
factors: first, that the subject tends to be “given”. or old, information, while the
object tends to be new information; second. that processing demands are higher
at the beginning of an utterance than near the end. and pronouns are the least
taxing NPs one can produce.
Unfortunately, it is difficult to derive explicit predictions from performance
accounts. L. Bloom’s and P. Bloom’s predictions are one set. We attempt in
Study 5 (on verbs and direct objects) to develop and test additional performance
predictions.

Acceptability
Adult English speakers omit the subject in a variety of contexts (O-Grady et
al., 1989; Valian, 1990a, 1990b). Expletive “it” is omitted, as in (7). from the
New York Times, as are other subjects in extended discourse:
(7) Seems like she always has something twin-related perking
The discourse and prosodic conditions on subject omission in English usage are
quite complex (Valian, 1990a). Omission is more acceptable before a main verb
than before a modal. as is shown by the greater acceptability of (9) compared to
(10) as a continuation of (8):
 (8) She’ll be a big hit
(9) Sings like a dream
(10) Can sing like a dream

Similarly, omission is more acceptable before a main verb than before auxiliary
be, as is illustrated by the contrast between (12) and (13) as continuations of (11):

(11) She’s going to be a big hit

(12) Sings like a dream
(13) Is singing like a dream

Yet when the modal or be is concatenated with a negative, the string becomes
more acceptable, as shown by the sequence in (14) and (15):

(14) She’ll never make the big time

(15) Can’t sing worth a nickel

What is acceptable in the adult community forms part of the child’s input. and
is also part of what children must master. The utterances that I have termed
“acceptable” are not grammatical in English (since English does not have pro
subjects, and also cannot be characterized as a simple VP). They lack subjects
and therefore violate the extended projection principle (Chomsky. 1981). which
we are assuming.
Children are exposed to fully grammatical utterances without subjects. in the
form of imperatives. They are also exposed to acceptable utterances which are
not fully grammatical, such as (9), as well as forms like. “Want lunch now?” The
American child must grow into an adult who not only knows that overt subjects
are grammatically required, but also knows when subjects can acceptably be
omitted. The child must not only acquire the correct grammar, but also master
the discourse conditions that allow relaxation of the grammar.
Utterances without subjects could be difficult input for the child in two ways.
The child could be misled into thinking that subjects are not grammatically re-
quired (O’Grady et al., 1989; Valian, 1990a, 1990b). Even if the child has deter-
mined that subjects are grammatically required she may not have learned where
subjects can acceptably be omitted. Given how complicated the conditions on
acceptable omission are, it will take the child time to zero in on the correct
contexts - how much time is not clear. The result in either case in the child’s
speech would be subjects which were absent too frequently: in the former situa-
tion the child’s competence would be deficient, in the second the child’s perfor-
mance. If children know subjects are obligatory, they should confine omissions
to linguistic contexts where adults might also omit them. Thus, in Study 3 we
contrast children’s use of subjects before modals versus main verbs. and in Study
5 we measure subjects in subordinate clauses (where subjects cannot be accepta-
bly omitted).
 Another way of conceptualizing “acceptable” utterances is to think of them as
grammatical within a different grammar. From that perspective, adults have a
variety of grammars. and use different ones in different discourse conditions. But
there are problems with that conceptualization. One is the need to posit that
adult English speakers have at least two grammars: one the grammar that we
usually think of as holding for English, another that either contains pro or allows
for VP clauses. The second is the possibility that an indefinite number of gram-
mars would then be attributed to any adult. in order to account for all the
non-standard utterances that are produced. In any event, on this conceptualiza-
tion as well. the child’s task is more complicated, since she must acquire more
than one grammar, and sort out the principles underlying each.

Prosod)
Gerken (1991) suggests a prosodic explanation of the child’s tendency to omit
unstressed syllables in utterance-initial position, which can also be applied to
adult speech phenomena. If a bias exists against beginning an utterance with an
unstressed syllable, that unstressed syllable will either be assimilated onto a fol-
lowing stressed syllable, or deleted altogether. The lesser acceptability of (10)
compared to (9) may thus have nothing to do with the distinction between modals
and main verbs, but may only reflect our preference to begin sentences with a
stressed syllable: (10) does not lend itself to initial stress. The “weak” nature of
modals and be may make them less acceptable in initial position; when they are
concatenated with a negative, they become “stronger”, and thus more acceptable.
Gerken (1991) had young 2-year-olds imitate sentences with different types of
subject and object NPs (consisting of proper names, pronouns, or full lexical
NPs). She found that children omitted subject NPs more often than object NPs
- roughly 19% omissions versus less than 1% omissions. Children also omitted
determiners more often in subject than object NPs, roughly 31% omissions versus
18% omissions. Pronoun subjects were omitted about 32% of the time, proper
nouns and common nouns about 12% of the time. Gerken hypothesizes that
initial weak syllables of a metrical foot have a certain tendency to be omitted,
so that the same prosodic process is responsible for the omission of subject pro-
nouns and the omission of subject determiners. In addition, Gerken reports that
when a determiner in object position is the initial syllable of a metrical foot, the
omission rate is the same as when it is in subject position. That would suggest
that omission of initial weak syllables, regardless of sentence position, is respon-
sible for subject omission.
There are, then, ample reasons for thinking that factors independent of the
child’s syntactic competence are partially or wholly responsible for children’s
omission of subjects in particular and of syntactic elements in general. Those
factors may include processing load, characteristics of the input, ignorance of the
exact acceptability conditions for subject omission, and prosodic effects. In all
these cases, subjects are more prone to reduction or omission than objects.
 Teasing apart competence and performance
The present study uses two strategies to tease apart competence and perfor-
mance factors. One is to explore both competence and performance factors within
English. To begin with the American children, we examine properties that the
children’s speech should have, on different theories, if they understand that sub-
jects are obligatory. using as a benchmark the speech of children who are consis-
tently using subjects. Thus, we measure use of: subjects, pronominal subjects,
nominative case-marking of pronominal subjects: expletives; modals: infinitival
to, past tense, and third person singular: subordinate clauses: direct objects.
Simultaneously, we look for evidence of performance limitations.
The use of a benchmark, in the form of the speech of children who are consis-
tently using subjects, is an important control. If children who use subjects consis-
tently use, say, few expletives, the absence of expletives in the speech of children
who use subjects inconsistently is not informative.
The second strategy is to compare English-speaking children with children
learning a null subject language, such as Italian. If Italian children are used as a
benchmark, American children’s similarity to them can be assessed. We compare
production of some of the same elements in the two languages, such as subjects,
pronominal subjects, and modals. If the measurements yield the same results. we
can conclude (a) that American children are not sensitive to the differences be-
tween their language and a null subject language or (b) that the measures reflect
universal performance characteristics. If the measurements yield different results,
we can conclude that American children are sensitive to how subjects are used
in their linguistic community.

Study 1: Subject use

The first and most basic question is how frequently children use subjects. Fre-
quency figures alone will not tell us whether children understand that their lan-
guage does, or does not, require subjects, but the facts about usage will tell us
what the phenomena are that need explaining. Data from American and Italian
children will allow us to contrast usage between speakers of a non-null subject
language and a null subject language, to see whether differences between those
two canonical language types are reflected in the speech of children, or whether
children treat the two languages as if they were the same.

Competence-deficit predictions

On the pro hypothesis (Hyams, 1986, 1987), the American child, like the Italian
child, has available three possible types of subjects: full lexical noun phrases
(NPs), pronoun NPs, and pro NPs. If we ignore temporarily the possible effects
of different input from adults of the two languages, the children of both languages
should produce null subjects. and produce them equally often. since they both
think their language is a null subject language.
Given that a pro NP is considered to be an alternative version of a pronoun
NP. we can also derive the prediction that there will be fewer pronoun NPs in
the speech of a null subject language speaker than in a non-null subject language
speaker. This is especially so if, as Hyams (1956) hypothesizes. there is a principle
like avoid pronoun (Chomsky. 1981). which states that the speaker will not use
a pronoun NP, b u t a pro NP. as the default in a null subject language. Thus, if
children believe their language is a null subject language, and if their performance
is constrained by the avoid pronoun principle, there should be few pronominal
subjects in their speech. The subjects that are expressed should be primarily
lexical rather than pronominal. That prediction can be verified by looking at
pronominal NPs in American and Italian children’s speech.
On the VP hypothesis (e.g.. Guilfoyle & Noonan. 19S9). children of all lan-
guages should look the same at an early stage of development: their grammars
consist of VPs. with optional subjects. If American children at an early point in
development (say, under MLU 2.0) have only VP grammars. then they should
be indistinguishable from children of other languages at a comparable develop-
mental point, again ignoring possible effects of input.
Neither the pro nor the VP hypothesis specifies any particular role for input.
Both make clear claims about the form of the child’s grammar, but not about
the child’s output. The child’s grammar is one determinant of her output; her
performance system is another. If those were the only two determinants, then we
could confidently predict equal use of subjects by all children regardless of target
language. But, since there is evidence that children’s frequency profiles match
their parents’ (Brown, Cazden, & Bellugi, 1973), we might expect parental use
of subjects to influence child usage. American and Italian adults do use subjects
to different degrees, and we might thus expect differences in children’s speech
on that basis alone. We will return to this problem in the discussion.

Performance predictions

Like the competence-deficit accounts, performance accounts also make no firm

predictions about amount of subject usage, but a performance account will predict
certain differences between American and Italian speech. If the American child
understands that her language requires subjects, and omits subjects for perfor-
mance considerations only, there should be more subjects in her speech than in
the speech of an Italian child who understands that overt subjects are syntactically
optional in her language. Similarly, there will be no syntactic reason for American
children to omit pronouns if pro is not part of their grammar. Therefore, a
performance account should also predict a higher degree of pronoun usage among
Americans than Italians.
 MethOd

American corpora

Children and taping procedures. Twenty-one children (12 girls. 9 boys) were
audiotaped, using a cassette tape recorder. in natural conversation and play with
their mothers. Two-year-olds were recruited among acquaintances of the author
and her assistants. and via bulletin boards and local newspaper advertisements.
The children ranged in age from 1:lO to 2% All the children were white. and
the socio-economic status of their parents ranged from working class to upper
middle class. Six households had one or two Ph.D.s or graduate students.
Typically there were two tapin g sessions per child-parent pair. no more than
2 weeks apart. Session 1 was usually half an hour, and Session 2 one hour. Taping
was performed at the child’s home or day care center. or in a college play room.

Transcription and ML U calculation. Each tape was transcribed by one lis-

tener and checked by another, uith the exception of a child of MLU 3.72. whose
transcript came from a previous study. Conventional English orthography was
used. Doubtful portions were placed in parentheses, while completely unintellig-
ible portions were so indicated.
The children’s MLUs in morphemes were calculated according to Brown’s
(1973) procedures for each taping session. Each child’s average MLU was then
used as the most reliable estimate of the child’s overall linguistic level during the
taping period. The MLUs ranged. roughly evenly, from 1.53 to 4.38. For this
study, the children were divided into four groups, based on their MLUs. Table
1 shows the distribution of children by age and MLU.
Observers were asked to note clarifying context, in particular so that impera-
tives could be distinguished from declaratives which were lacking a subject. This
was done for the purposes of Valian (in press), where the grammaticality of
children’s utterances was coded. (An imperative would be classified as grammat-
ical, but a declarative missing a subject would be classified as ungrammatical.)
For the higher MLU children (above 3.0) the surrounding verbal context was
usually sufficient to determine whether an imperative was being spoken, but for
the lower MLU children (under 3.0) the observers’ clarifications were helpful.
The present study was not contemplated until most of the grammaticality assign-
ments had been made. It is therefore unlikely that there was any theoretical bias
in deciding whether an utterance was an imperative.

Utterances used in analyses. Three classes of child utterances were eliminated

from analysis. The first class, discards, consisted of utterances unintelligible in
whole or in part, interrupted utterances. and utterances that consisted solely of
single-word assents, dissents, or hesitations. (Unintelligible utterances, inter-
rupted utterances, and hesitations were also excluded from MLU calculations.)
Table 1. American children: .\IL LT.ngr. trnd ~rtrrrtrnce inforrfw~ior~

,LlLU ALIe Number total Number + \’ Proportion + V

ufteranccs” utteranccsh ufteranccs

Group I 1.53 2: I 47-I 47 10

1.74 1:ll) 491 169 .34
1.79 7.3 322 76 .2-l
_._
1.81 1:lO 259 62 .2-l
1.99 1.)
_._ 187 76 .Jl
‘Mean 1.77 -.
‘.i, 347 86 ?7
._

Group II 2.24 25 264 In6 .40

2.28 2:3 442 227 ._il
2.52 2: 7 566 271 .48
2.66 _.C
-I.$ 330 IS2 .55
2.76 7.i
_._ 390 250 .6-l
Mean 2.49 _‘.<._ 398 207 5’
.__

Group 111 3.07 25 S34 406 .76

3.15 2: 6 2% 217 .76
3.16 2:3 520 287 .ss
3.31 1.;
-._ 418 289 .69
3.44 ,.;,
_._ 437 322 .7-l
3.62 -I.$
_._ 141 309 .70
3.68 7.q
_._ 359 275 .77
3.72 1.T
_._ 2’31 182 .63
Mean 3.39 25 411 286 .70

Group IV 1.12 2:s 268 22s .84

4.17 2:s 333 251 .75
4.38 ‘.6
_. 347 273 .79
.Mean 4.22 217 316 250 .79

aNumber of non-imperative non-imitative utterances: this is the denominator for calculating the prop-
ortion of total utterances with a subject.
bNumber of non-imperative non-imitative utterances containing a verb; this is the denominator for
calculating the proportion of +V utterances with a subject.

The discarded class was roughly 40-50% of the child’s utterance tokens. Utter-
ances with doubtful portions were included in usable utterances. Repetitions of
whole utterances were counted as separate utterances; we counted utterance to-
kens, not types.
Since the focus of the study was the child’s spontaneous productions, kve sub-
tracted two other classes from the set of usable utterances: imitations and stock,
routine utterances. An imitation was an utterance which directly followed an
adult utterance. was either a full or partial repetition of that utterance, added no
new material, and made no morphemic alterations (Ervin, 1964). There was one
 S_vntacticsubjects in ecrrly speech 39

exception to this. If the child repeated a parental utterance that was itself an
imitation of the child’s spontaneous precedin g utterance. the child’s repetition
was not scored as an imitation. A routine was a stock utterance type used re-
peatedly by the child with no variation from one time to the next (cf. Brown.
1973: Peters, 1983). Table 1 shows the number of usable utterances per child.
which ranged from 187 to 566.
Questions as well as declaratives \vere included in the analyses (in both cases
only if the utterance included a verb). W/z-questions required a scoring decision.
In the adult grammar, most of the wh-words would not be considered subjects.
but (underlying) objects or adjuncts of the verb (moved into COMP). The ques-
tion -*What is that?” would be considered to have “that” as its subject in the adult
grammar, not “what”. Of the lc*h-aords the children used. whnt and who could
be (underlying) subjects or objects in the adult grammar, depending on the sen-
tence; *‘What happened?“. for example. would be considered to have “what” as
its underlying subject in the adult grammar. Where, how, and why would be
classified as objects or adjuncts.
In the child grammar. \ve do not know how the child is analyzing questions
like **What is that?” For the child “ivhat” could incorrectly be base-generated as
the subject, or as the object moved into COMP, or as the object moved into
some other position. We included such question tokens as part of the children’s
data for two reasons. First. in Italian. it is possible grammatically to say the
equivalent of “what is?“. or “nhere is?” If American children include both a
wh-word and a “subject” NP. that is noteworthy. In fact, almost never did a child
omit the true subject: out of a total of 552 wh-questions where the wh-word would
not be classified as the subject in the adult grammar, the children supplied a
subject in all but 9 cases, evenly scattered among the MLUs. Second, for all the
children (except MLU 1.81, who produced no wh-questions), their questions
appeared at least minimally productive. Even the lowest-MLU child, whose only
wh-word was “what”, and whose only accompanying verb was the copula “is”.
showed some productivity. using ‘?t”. “this”. and “that” as the “subject”, using
both the contracted and uncontracted form of “is”, and once using no verb.
For the subject counts, then. it did not matter whether in the child’s grammar
the wh-word or the other NP was classified as the subject. because the children
included both. We did. however. also perform an analysis excluding tvh-ques-
tions, to see whether that would reduce the children’s percentage of subjects. In
counting the number of pronominal subjects, we performed two analyses: one in
which we included wh-questions and one in which we excluded them. When we
included wh-questions we decided to err on the side of conservatism, and scored
whar and who as the subject. In almost all such cases, the true subject was itself
a pronoun (usually “this”. “that”. or .-it”), and thus a pronominal subject would
have been scored either way.
 Italian corpora

Children, taping, and transcription. Five northern Italian children were audio-
taped in conversation nith an observer. Transcription was performed by G.
Tirondola. and the transcripts were lent to the author by F. Antinucci. The
children were observed 11 times. beginning at age 1;6 or 1:7. The observations
occurred once a month. escept for a two-month hiatus encompassing a summer
vacation: the break occurred during Month 6.
From the length of the transcripts. each recording session appears to have
lasted no longer than 15-30 minutes. For that reason, it was necessary to pool
sessions. We divided the data into Time I and Time II. Time I covers the first 5
sessions, before the summe; break: the children were about 1;6 to 1:lO during
that time. Time II covers the last 6 sessions, after the summer break: the children
were 2;0 to 2;5 during that time. Table 2 shows the number of usable utterances
per child, and the number of utterances containing a verb per child. Since MLU
cannot be calculated in the same way with Italian children as with American
children, it was not computed.’
A native Milanese. lvho was also a graduate student in linguistics in hew York,
and who speaks English fluently, translated the transcripts into English. and
coded them. The author reviewed the translations and coding in discussions with
the translator.

Utterances used in analyses. In discarding utterances we used the same criteria

that we used with the American children. Almost none of the utterances were
unintelligible; we surmise that unintelligible utterances were not transcribed.
From the set of usable utterances we subtracted imitations and routines. We did
not separate declaratives and questions since there were so few utterances avail-
able per child. For the 5 months comprising Time I there was an average of 150
usable utterances per child: for the 6 months comprising Time II there was an
average of 202 usable utterances per child.

Comparison of American and Italian children

Age. The American children in Group I ranged from 1;lO to 2~2. lvith an
average age of 2;0. The Italian children at Time I ranged from 1;6 to 1:10, and

‘English has relatively little morphological complexity compared with Italian. In American chil-
dren’s speech bound morphemes are largely confined to plural -s. third person singular present tense
-5. progressive -ing, and past tense -ed. Italian children have gender and number for nouns and
adjectives. person. number. and tense for verbs. and some preposition-determiner combinations. It
is difficult to know when these morphemes are genuinely analyzed as morphemes by the children. If
they are assumed to be analyzed from the beginning, the Italian child’s ,MLU is very high relative to
the American child’s,
Table 2. Itdiritt cltiltirrtl: uttermcr it~fortntrtiot2

Child Number total Number + V Proportion + V

utterances’ utterancesh utterances

Time I E I19 -I3 .36

0 133 32 .24
M 168 -I2 .25
G 13Y 41 20
D 102 3Y .20
1Mean 150 79 .27

Time II E 16-t 5.i .3-l

0 1Yh 3’z
< .43
M 219 S6 .3Y
G 211 S-1 .-to
D 220 S-l 3s
Mean 202 78 3

NOW. Children were aged I:6 or I:7 at start of taping. Children were taped once monthly.
escept for Month 6. Time I comprises Months l-5: Time II comprises IMonths 7-12.
“The number of non-imperative non-imitative utterances: this is the denominator for
calculating the proportion of total utterances with a subject.
*The number of non-imperative non-imitative utterances containing a verb: the de-
nominator for calculating the proportion of +V utterances with a subject.

at Time II from 2;0 to 2;5. The Italian children were thus younger than the
American Group I at Time I and older at Time II. They were slightly younger
than Group II at Time II.

Verb usage. In order to have a linguistic way of comparing the children, we

calculated the proportion of non-imitative, non-imperative utterances with verbs
for all groups of children. The proportions for American children are shown in
Table 1, and those for Italian children in Table 2. The Italian children at Time
I produced the same proportion of utterances with verbs as the American children
in Group I (.27), and at Time II produced a higher proportion (.39).’

‘The present study compares cross-sectional data from Americans with longitudinal data from
Italians. It is necessary to ask whether cross-sectional data can be used to investigate children’s use
of subjects. It could have turned out that there was so much variation among children in their rate
of production of subject noun phrases and other constituents that a cross-sectional study based on
either age or MLU would be unrevealing. If subject use and related phenomena are unrelated to age
or MLU. then cross-sections made on the basis of age or MLU will not show an orderly development.
As will be seen. however, the data reported here are orderly. On most measures there is a relation
with MLU. the primary basis on which the children were grouped. Severtheless. future work should
also be undertaken to confirm our cross-sectional data with longitudinal analyses. One advantage of
cross-sectional data should also be noted: a IarSer sample of children is thereby possible.
 :Vleasures of subject use
M’tt measured subject use two ways: (1) total appearance in all usable non-im-
itative. non-imperative utterances, and (2) appearance within utterances contain-
ing a verb.

Memire I: Total subject appearance. The denominator consisted of all non-

imitative non-imperative usable utterances. including questions (shown in Table
1). The numerator consisted of utterances containing verbs which also contained
a subject noun phrase (as in Measure 2 below). as well as utterances consisting
of a subject NP plus an object NP. with a missing verb, typically a copula.
Examples of strings classified as having a subject despite verb absence are: **this
a good story”, “birdie inside”. “eggs in house”. “this outside”, “hand in there’?“.
‘-it noise”. “yeah. I my diaper on now”. The parent’s response was used to resolve
difficult cases: “Baba [referring to the child] popcorn. too” was scored as contain-
ing a subject: the parent’s response was “you like popcorn. too”. W/r-questions
lacking a verb but containing a &-word plus object were included in the
numerator.
Cases where neither the form nor the parental response justified coding a
subject were not included. A fragment consisting of a single noun phrase. for
example. is impossible to categorize as subject or object NP. and would be in-
cluded only in the denominator. Cases of a noun followed by an adjectiv.e were
typically scored as NPs rather than as a noun and a predicate adjective. As a
result. the complement of subject appearance cannot be interpreted as subject
omission.

Measure 2: Subject appearance in utterances n*ith verbs. The denominator

consisted of all non-imitative, non-imperative usable utterances with verbs. in-
cluding questions (shown in Table 1). The numerator consisted of the subset
containing subjects. We also performed an analysis on the American children
excluding w/z-questions. In Measure 2. the complement of subject appearance cnn
be interpreted as subject omission.

Measures of pronominal subject use

We had two measures of children’s use of subject pronouns: (1) total subject
pronoun use, and (2) for American children only. total subject pronouns exclud-
ing &-questions.

Measure 1: Total subject pronoun use. The denominator consisted of all utter-
ances including a verb and including a subject (the numerator from the measure
in Figure 2). The numerator consisted of all subject pronouns in those utterances.
including personal pronouns. demonstrative pronouns (-‘this’*. “that”. etc.). and
the interrogative pronouns “who” and “what”. Variants of “What’s this?” (such
Figure 2. At?lericon chiltlretr ‘s use of .mbjects in non-imirrrri~~e rlorl-irrlperoril~t, lctferrrnce.s
hirll ~w-hs.

100 7
95 -

90 -

ri
85 -
80 -

5:-

z-
0 Group1
55 -
2iJ 50 - n Group II
45-
: 0 Group III
R 3":-
30- A Group IV

E-
15-
lo-
5-
0 . I I I I I I I

1.5 20 2.5 3.0 3.5 40 4.5

Mean Length of Utterance

as “What is this?“. “What’s that?“) were scored as having an interrogative pro-

nominal subject. Only utterances with verbs were included.

Measure 2: Subject pronoun llse, excluding interrogatives. To ensure that n~it-

questions were not skewing the results. they were excluded from Measure 2. The
denominator thus consisted of all utterances including a verb and including a
subject, minus &-questions. The numerator consisted of all subject pronouns in
those utterances, including personal and demonstrative pronouns.

Results

We first present the American and then the Italian data

American children

Subject use. Measure 1 of subject use was the percentage of subjects in all
non-imitative non-imperative usable utterances, with or without a verb. In Group
I the mean percentage is very low - 23%: it increases to 50% in Group II. 65%
in Group III and 77% in Group IV.
4-i L.. Cblian

Figure 2 plots the results for Measure 2 of subject use (percentage of subjects
in non-imitative non-imperative usable utterances with verbs). for each child indi-
vidually. The MLU group into which each child was placed is shown in the
legend.
Figure 3 shows the means of subject use (Measure 2) for each MLU group. If
one concentrates on the bars for American children, one sees that the 5 children
in Group I showed an average use of subjects of 69% (s.d. = 12). There is a
clear increase in per cent usage between Group I and Group II: the 5 children
in Group II showed an average use of 89% (s.d. = 4). Performance was consis-
tently high thereafter, increasing slightly in the nest two groups (Group III mean
= 93% (s.d. = 4): Group IV mean = 95% (s.d. = 3)). As the standard deviations
indicate. Group I was the most variable group. Table 3 presents the data individu-
ally for the children in Groups I and II. An important question is the significance
of the 69% figure for Group I, in light of the variability in this group, where two
children show subject use as low as 55% and 58%.
When b&-questions were excluded. the results were almost identical escept for
two children in Group I. For the lowest-MLU child. subject use declined from
55% to 35%. and for the child at MLU 1.74 the decline was from 76% to 69%.
That reduced the average subject use for Group I from 69% to 64% (s.d. = 17).

Figure 3. Comparison of American anti Italian ckildren ‘s productiort of mbjects md pro-

nominnl subjects in utterances with verbs.

Group I Group II Group I II Group IV Time I Time II

Americans Italians
63 % Subjects N % Pre-Verbal Sublects

0 % Pronominal Subjects H % Post-Verbal Sub]ects

0 % Pronominal Sba!ects
 Syntactic .subjects in et+ speech 45

Table 3. American children: close-up of Groups I and II. Proportion of utterances con-
taining a verb which contain a subject noun phrase

MLU Proportion of Frequency of

subject use subject use”

Group I 1.53 ‘:I .55 26147

1.73 I:10 76 1291169
1.79 _._
7.7 17-1 56/76
1.81 1:lO 58 36l62
1.99 _,i
7.7 .Y? 62176
Mean 1.77 2:o .69 (s.d. = .12)

Group II 2.24 2:3 .90 95/106

2.28 2:3 .81 1911227
2.52 2:7 .Y5 23427-l
2.66 2:8 .91 1661182
2.76 25 .9-I 236i250
Meall 2.49 25 .89 (s.d. = 44)

“The denominator is the number of non-imitative non-imperative strings containing a

verb: the numerator is the number of those strings containing a subject.

For 18 of the remaining 19 children there was either no change or a decline of

one percentage point.

Pronominal subject use. Whether Measure 1 or Measure 2 is used to assess

pronoun use, at every MLU group a majority of the children’s subjects were
pronouns. Further, for every child pronouns are more than half their subjects
(the lowest figure was 59%), and usually more than three-quarters. Measure 1
includes w/z-questions and classifies interrogative w&words as pronominal sub-
jects. On that measure, .77 (s.d. = .15) of Group I’s subjects are pronouns, .87
(s.d. = .06) of Group II’s, .8.5 (s.d. = .05) of Group III’s, and .84 (s.d. = .04)
of Group IV’s Had the true (adult underlying) subjects of &-questions been
scored as the subject, instead of the w&words, the figures would have been
almost identical, since in almost all cases the true subject was “this”, “that”,
“these”, “those”, or ‘*it”. The results are shown in Figure 3 to the right of each
bar representing subject use. Pronoun use is represented as the percentage of +V
utterances containing a pronominal subject (Group I = 53; Group II = 77; Group
III = 79; Group IV = SO).
Measure 2 excludes n&-questions and includes only personal and demonstra-
tive pronominal subjects. Here the averages are slightly lower. For Group I there
is a decline of 4 percentage points: 73% of all subjects were pronouns. For the
remaining three groups the decline is only 1 point.
 Parrntal use of subjem. In order to determine \vhat the input to the child is.
we calculated the parents’ use of subjects in their immediate replies to children’s
utterances. We included only non-discards. non-imitative. and non-imperative
utterances with verbs. Parents are very consistent in using subjects. In Group I,
.96 (s.d. = .0-I) of usable parental replies included a subject. in Groups II. III.
and IV. .9S (s.d. per group always belou .02). \vhen imperatives are included.
the percentage of replies including a subject drops to .90 at Group I. and to .Y3
for Groups II. III. and IV.

Since so few utterances per child were available for the Italian ciata. fewer
analyses were performed.

Use of subjects and pronominal subjects. The total proportion of non-impera-

tive non-imitative strings containing subjects. whether a verb was present or not.
is .20 at Time I and .23 at Time II. Those figures are comparable to American
children in Group I.
With respect to appearance of subjects in utterances containing a verb. Italian
children look completely different from American children. The right-hand por-
tion of Figure 3 shows the Italian children’s use of subjects in utterances contain-
ing a verb. The top and bottom halves of the subject bars show pre-verbal and
post-verbal subject use. From Time I to Time II the overall proportion of subject
use stays constant at about .30 (average s.d. = .OS), less than half the average
usage of Group I Americans.
At Time I the children produce twice as many post-verbal (.21, s.d. = .04) as
pre-verbal (.09, s.d. = .06) subjects. By Time II the children have almost doubled
their production of pre-verbal subjects (to a proportion of .15, s.d. = .05), but
total production of subjects remains the same from Time I to Time II. The Italian
children differ from American children, in producing few subjects and in showing
no increase in production within the time period sampled.
Compared to the American children’s, the Italian children’s total subject ap-
pearance is high relative to the use of subjects in utterances with verbs. In the
American Group I data. the percentage of subjects in a11 strings was one-third
the percentage of subjects in strings with verbs. whereas here it is two-thirds. In
the Italian coder’s opinion, many of the Italian strings lacking verbs consisted of
a subject plus an object or adjective, or an adjective plus a postposed subject.
The latter structure in American children’s speech was considered to be simply
a noun phrase.
The Italian children also produce less than half as many pronominal subjects
as the American children. Because of the small number of utterances with sub-
jects, the data were pooled rather than averaged over subjects. Pooling data
resulted in slightly larger averages than averaging over subjects would have.
Figure 3 presents the proportions of subjects containing any sort of pronominal.
whether it be a personal pronoun, a demonstrative. or an interrogative. At Time
I 22% of the Italian children’s subjects were pronouns, and at Time II 35% were
pronouns. (For Group I of the American children, about 75% of the subjects
were pronouns.) At both Times I and II. pronouns were more common in post-
verbal than pre-verbal position.

Discussion

The data comparing American and Italian children show that American children
look very different from Italian children with respect to use of subjects. Italian
children look the way one would expect the learner of a prototypical Romance
null subject language to look. From age 1:6 to 2:6 they include subjects in a
minority of their sentences with verbs - about 30% - and they use pronoun NPs
as a minority of their subjects - about 20-35%.
In classifying the Italian children’s utterances we have assumed that the post-
verbal subjects are genuine subjects that were postposed from pre-verbal position.
On some analyses, however, the post-verbal subject is generated in situ. and pro
is generated as the pre-verbal subject. If pro were considered to be the subject
in all Italian sentences except those with a lexical pre-verbal subject. Italian
children at Time I would be using subjects only 9% of the time, and at Time II
only 15% of the time.
In contrast. American children in each of our hILU groups include subjects in
most of their sentences with verbs. In Group I (our lowest-MLU group: MLU
1.53-1.99). almost 70% of the children’s utterances with verbs include subjects -
more than double the rate of Italian children. Even when &-questions are
excluded from the American data, the American children produce subjects at
double the rate of Italian children. Group II children (IMLU 2.25-2.76) use sub-
jects consistently; the lowest percentage of subject use is 84% of utterances with
verbs - still a high level of performance.
Further, Group I uses pronoun NPs for the majority of their subjects - over
70% - again more than double the Italian rate. We note a parallel in use of
pronouns: just as American percentage of pronoun use increases, so does Italian.
From Group I to Group II the American use of pronominal subjects increases
from 73% to 86%; from Time I to Time II the Italian use increases from 22%
to 35%. This suggests that in both groups of children their initial use of pronom-
inal subjects is depressed for reasons independent of the status of subjects in the
language.
How has the impression of low subject use among American 2-year-olds devel-
oped? Examination of Group II’s total subject use indicates why subject use may
appear to be low even in children who are using subjects consistently. The percen-
tage of subjects in the whole corpus (and even these data do not include the
entire corpus, since they exclude discards, imitations. and imperatives) is low -
50%. Thus. in scanning a corpus, one can be misled by the large number of
utterances that lack a verb into thinking that the child’s use of subjects is low
even when it is high.
If we take subject usage from 84% to 94% as evidence that children under-
stand that subjects are obligatory, we can conclude that at least soon after MLU
2 American children exhibit no competence deficit. If the pro or VP hypotheses
describe a stage in acquisition. it would have to be a stage before Group II.
What about Group I? Here we must pause. Two children (at MLUs 1.53 and
1.81) showed a low use of subjects: 55% and 5S% subjects. When u*h-questions
were excluded. the lowest-MLU child’s usage was only 3S%. The other three
children in the group averaged 77% (74% when t&-questions were excluded).
Group I may consist of two subgroups. for one of which either the pro or VP
hypothesis holds. Under MLU 2.0, then. some children may exhibit a competence
deficit while others do not. With our sample we will be unable to resolve this
question. Our procedure will be to continue to treat Group I as a group. while
keeping in mind that the lowest-MLU child. in particular. may not conform to
the rest of the group.
Another caveat concerning our data is that Bloom, Miller, and Hood’s (197.5)
investigation of 4 children appears to show less subject use for 3 of their -I children
than our data would predict. Their data cannot be directly compared with ours.
since they excluded most intransitive verbs. Nevertheless, the comparison of
Bloom et al.‘s data with ours suggests that there may be developmental differ-
ences in children’s use of subjects. Only a larger database will answer this ques-
tion.
We can now consider how well the competence deficit and performance
theories account for Group I’s data, and for the differences between the Ameri-
can and Italian children. What do the theories predict? Are they confirmed or
disconfirmed?
If the competence deficit theories are interpreted as allowing no role for input
effects or performance constraints. they would then predict equal usage by
American and Italian children. On that interpretation the theories are discon-
firmed, because the data show unequal usage. If the theories allow input to
influence the child, and assume no performance constraints, they would then
predict that Italian children’s output would match Italian parents’ and American
children’s output would match American parents’. On that interpretation also the
theories are disconfirmed, since neither Italian nor American children produce
subjects at the same rate that adults do. (Bates. 1976. supplies figures for two
Italian parents in a total of three samples: the parents used subjects about 50%
of the time. (Bates states the percentages as 3040%. but my calculations from
her data yield 46-56%.))
Only if the theories include a role for input effects and performance constraints
would they be confirmed by the data. If children try to match their input regard-
less of their grammar. and if they are limited in how fully they can match the
input, then American children could have pro or an incomplete grammar. but be
producing subjects a majority of the time and at a higher rate than Italian chil-
dren.
Thus. the competence deficit theories can account for Group I’s data and for
the differences between the American and Italian children, but only if they add
the ancillary assumptions about input and performance. Neither competence
theory by itself can make any predictions about how the child’s grammar will be
manifested in the child’s output.
The performance account directly includes a role for performance constraints.
claims the child’s knowledge is correct. and is neutral about the role of input.
On that basis the performance account predicts that American children will use
subjects less than 10070 of the time, and predicts they will use subjects more than
Italian children do. The data thus confirm the performance account with no fancy
footwork needed.
On balance. then. our provisional conclusion is that American children know
early on that their language requires subjects: neither the pro hypothesis nor the
VP hypothesis consistently describes children above MLU 1.5. Since two of the
Group I children show relatively low usage, however, and since ancillary assump-
tions could bolster the competence deficit hypotheses, our conclusion is tentative.
The following studies look at other aspects of the children’s grammars in an
attempt to determine the soundness of our provisional conclusion.

Summary

American and Italian children at roughly comparable ages and levels of linguistic
sophistication use subjects to different degrees. Group I Americans (MLUs be-
tween 1.53 and 1.99). while not entirely consistent in their use of subjects, use
subjects and pronominal subjects about twice as much as Italian children. Neither
of the two competence-deficit hypotheses explored would predict such a pattern,
without being bolstered by ancillary assumptions. A performance hypothesis
straightforwardly accounts for the data.

Study 2: Expletives and pronoun types

While sheer amount of pronoun use by the American children suggests that they
are. at least, not operating with the wrong (Italian) value of the null subject
parameter, the types of pronouns children use might help clarify matters in two
ways. The first way concerns whether the children use expletive pronoun subjects,
such as it. The second way concerns whether the subject pronouns the children
use receive nominative case.
 If the children think their language is a null subject language, and if being a
null subject language entails lack of expletives. one might predict. as Hyams
(1986) did. that children’s speech would correspondingly lack expletive subjects.
Hyams has claimed that American children only begin producing expletive sub-
jects once they use subjects consistently. We therefore searched for the existence
of expletive subjects in the American children’s speech, and for “expletive” con-
texts in the Italian children’s speech. An expletive context was defined as one in
which a subject was obligatorily absent, as with the verb piovere - “to rain”.
On the VP hypothesis children’s grammars lack elements of the INFL system.
(In some recent treatments. e.g.. Pollock, 1959. INFL is not a node: instead.
tense and agreement each have their own separate node. We shall retain the
more usual conflation of both features under a single INFL node.) The feature
tense or Agr, within INFL, assigns nominative case to subjects. If we add no
input assumptions, it follows that if children lack tense. or an INFL system gen-
erally, they will be unable to reliably mark the case of subject pronouns. The
case of pronouns should be random. One vvay. then. that we can determine
whether the child’s grammar is lacking INFL is to determine whether subject
pronouns reliably receive nominative case. Accordingly. we established which
pronouns were used as subjects, and what case they appeared in.

Method

The corpora from Study 1 were used. Utterances labeled discards were elimi-
nated, as were imitative and imperative utterances.

Expletive subjects
American transcripts were searched, using a custom search program. for all
occurrences of it, as well as for likely expletive contexts. The expletive contexts
searched were ruin, snow. hot, cold, dark. and seem. The Italian corpora were
searched for verbs which require the absence of subjects: bisognare (--to need”),
piovere (**to rain”), basture (**to be enough”), accadere (*‘to occur”), avvenire (“to
happen”), sembrure (“to seem”), nevicare (“to snoa”), fare freddo (**to be cold”
- weather), fare caldo, fare notte, fare giorno, etc.

Pronominal subjects
We counted the personal pronouns the American children used as subjects via
a computer-assisted search. Among sentences with subjects, the following pro-
nouns were searched for: I. me, you, he, she. it. him, her, we, us. they. them,
em, my, your, his, hers, their, theirs. (For you and it case cannot be assessed.)
We also manually searched for pronominal objects of verbs and prepositions, to
determine whether children’s nominative case-marking was vacuously correct. If
children use the nominative form regardless of the position of the pronoun. they
cannot be described as knowing that the nominative case should be restricted to
subjects. The Italian corpora were not searched for pronouns because the
database was too small.

Results

Expletive subjects
For the American children. expletive use is infrequent at all MLUs. To the
extent that there is use of expletives. it occurs across the board. Among the 21
children there were only 12 candidate instances of expletive it:

Child MLU 1.53 “(When it’s noe-y)”

1.79 “When it rains”: “It rains”
2.76 “It’s dark”: “It’s dark outside”
3.07 “It’s winter time?”
3.15 “It looks summertime, ‘cause it’s summertime”
3.31 .* ‘Cause it’s dark”: “When it’s hot”
4.12 “Okay, it’s very cold”: ‘*It’s hot”

There were no instances of the expletive contexts that we searched that lacked
an expletive subject. Thus, as far as can be determined. there were no instances
where a context required an expletive and the child failed to provide one. Un-
doubtedly such instances occur in child speech, as they do in adult speech. But
since the base rate of expletive contexts is low, it will be difficult to find examples
in only one and one-half hours of taping. Given how infrequently high-MLU
children, who are producing subjects more than 90% of the time, produce exple-
tives, the low rate of production by low-MLU children is not meaningful. Exple-
tives are not a good diagnostic for our data.
An incidental fact can be noted: 5 of the 12 expletives were in subordinate
clauses. Considering how few subordinate clauses are produced, that percentage
is very high.
The Italian children only produced examples of bisognare (**to need”). piovere
(**to rain”), and basture (**to be enough of’). At Time I there was one instance:
at Time II there were a total of 11 instances, produced by 4 of the 5 children.
All of the examples correctly lacked a subject. While the rate of production of
expletive contexts is low, it is much more frequent than the American children’s
production.

Pronominal subjects
All the American children produced personal pronouns as subjects: no child
produced fewer than three types. The lowest-MLU child, however, produced a
total of only 5 tokens, 19% of subjects. Further, Group I as a whole produced
a smaller percentage of personal pronouns as subjects (52) than did Groups II
(7s). III (67). or IV (71). Group I, then. and especially the lowest-MLU child.
need further scrutiny.
With respect to the case of the pronouns used. Table -I shows that the children
overwhelmingly used the nominative case when they produced pronominal sub-
jects. Group II produced the largest proportion of incorrectly cased pronominal
subjects. at 2% of total pronominal subjects: Group II was the only group in
which every child produced at least one incorrectly cased pronoun.
There were no examples of nominatively marked pronouns in object positions.
(A possible exception was the utterance “Putting they. in their bed”. which ap-
peared to us to be a false start.) For all groups. when the children used a pronom-
inal object which received overt case-marking (e.g.. rite. !zirn. /zer. 11s. ther77). the
form was correct. But there were few examples of markable pronouns in object
positions in Groups I and II: most pronouns used after a verb were if. JXN~. this,
or that. In Groups I and II only 25% of the pronouns were markable pronouns,
contrasting with 80% object in subject position. In Groups III and IV objectively
marked pronouns were common.

Disxssion

When expletives are used, they are used by our children below MLU 2 as fre-
quently as by our children above MLU 4. But espletives are rarely used by
American children across the MLU range we observed. Even when children are
producing subjects for 90% of their utterances with verbs, they appear to have
little reason to utter expletives, doing so only 12 times in our corpora. The
children do not use verbs, such as seem, which take expletive subjects, and they
probably only speak about the weather when it is relevant to undertaking some
activity. Contrary to Hyams’ (1986) claims. then. we see no relationship between
expletive use and subject use.

Table 4. American chiiciren: proportion of different types of pronominal stlbjects

Group I you he ir ,rv they my him em All All .I

she me his born *
her
--
I .63 .05 .I5 .I3 .02 .02 ,003 0 0 .S2 ,003 34
II .66 .06 .07 .12 .O? .OJ .Ol .Ol .OOJ .so .02 111
III .49 .17 .14 .I2 .03 .05 0 ,003 0 .71 .003 174
IV .4 .I0 .16 .li .13 .06 0 ,002 0 .79 ,002 169

‘Lore. The final column is the mean number of personal pronouns used per child. “All Sam” is the
proportion of nominative pronouns (I, he/she, n’e. they): “all -.’ is the proportion of incorrectly cased
pronouns (nix. me. him, hir. her, em). The remaining pronouns (FOU and it) are not overtly case-
marked.
 The data from the Italian children support the interpretation that American
children use few expletives for semantic rather than syntactic reasons. Although
the Italian children produced expletive contests more often than the American
children did. their use appears to reflect the meanings of the verbs involved rather
than the grammars of the two target languages. Of the three “expletive” verbs
the Italian children used. two are clearly of importance to children: bisognnre has
to do with there being a need for something. and basrnre with there being enough
of something.
Certainly there is nothing in our data to suggest that expletive use is related
to subject use. If expletive use is taken to be diagnostic of knowledge that subjects
are obligatory, then even Group I children understand that. Their inconsistent
usage of subjects could not be esplained by the pro hypothesis.
The predictions of the VP hypothesis. which would explain the inconsistent
subject usage by our lowest-MLU group as due to a grammar lacking INFL. are
not borne out by our data. Contrary to prediction. even Group I children show
uniform use of nominatively cased NPs in subject position. Nevertheless. there
are two countervailing factors.
First, the lowest-MLU child produced particularly few personal pronouns as
subjects. This was also the child vvho produced the smallest percentage of sub-
jects. This child, then, may be best described in terms of the VP hypothesis. and
the period between MLU 1.5 and 2.0 may be best viewed as a transitional period
when some children have a full grammar and others have only a VP grammar.
The remaining children in Group I showed usage of personal pronouns slightly-
below the higher-MLU groups. but within their range.
Second, the clearest distributional pattern would be uniform use of nominative
pronouns in subject position and of objective pronouns in object position. as vvas
the case for Groups III and IV. But Groups I and II behaved differently. Those
children consistently used nominative pronouns in subject position but used few
markable pronouns in object positions (using primarily _vou. it. and demonstra-
tives). Groups I and II appeared to “avoid” using markable pronouns as objects.
although they correctly case-marked the markable pronouns they did use in object
position.
The most reasonable interpretation of Groups I and II’s pattern (excluding the
child at MLU 1.53) is that the children primarily knew the nominative forms of
the pronouns they used. and also knew that the forms bcverenominative. Other-
wise they should have used the nominative forms freely in all object positions.
which they did not do. If the only way of establishing nominative case is via
INFL. then the children also have INFL. If that is so, then the VP hypothesis
cannot account for Group I’s inconsistent use of subjects, and a performance
analysis is correspondingly strengthened.
Whether the children assigned nominative case via INFL, however. or simply
understood that the nominative form appears in subject position, cannot be
answered by our data. Input may play a role: the child might restrict herself to
the pronominal forms that she has heard in subject position. which will generally
be nominative. No change in the child’s output would then be observed vvhen
case was assigned via INFL. The effect of adding this assumption is to rob the
VP hypothesis of predictive force. since it will now predict both random use of
pronouns and consistent use of nominative pronouns. A different ancillary as-
sumption could be derived from Lebeaux (lYS7). who proposes that case is first
assigned via the phrase structure configuration. and then via tense. Again. how-
ever, predictive force regarding case would be lost.
The lack of objectively or possessively cased pronouns is surprising. given cited
examples of both (e.g.. Budwig, 19S5. 19S9: Radford. 1990). At present we
cannot say why such discrepancies exist. It may be that. although errors of using
objective case in nominative position exist. they are generally infrequent in each
child’s usage relative to the number of correct uses. or that there is wide variation
among children in their use of non-nominative forms.
Previous data summaries do not provide the comparative figures required to
determine how common errors of incorrect case are. Data on first-person sub-
jects. provided by Budwig (personal communication). indicate that, of the six
children she observed, one used “me” for subjects about 25% of the time. while
the rest showed little or no usage. Three frequently used “my” as first-person
subject. with percentages ranging from 21% to 11%. Thus. there may well be
children who show extensive usage of non-nominative forms in subject position.
We need data from large numbers of children. so that developmental differences
in acquisition can be identified and evaluated. Reanalysis of others’ transcripts
and data is likely to underestimate the true variation in acquisition. Our data
strongly suggest that, from MLU 1.5 on. there is consistently correct use of
subject pronouns.

There is no support in our data for either competence-deficit hypothesis as an

explanation for Group I’s behavior. Group I’s inconsistent subject usage does not
co-occur with other properties that would be expected if either competence-deficit
hypothesis were correct. The pro hypothesis is not confirmed by the children’s
use of expletives: to the extent that children use expletive pronouns, they do so
across a broad MLU range and subject use range. If the expletive data are
meaningful, either our lowest-MLU children already understand that subjects are
required, or expletives are irrelevant to the status of subjects. The VP hypothesis
is also not confirmed, because the children appear to have correctly cased
nominative pronouns in subject position. If the only way nominative pronouns
can arise is via marking from tense or Agr, then Group I children, by implication.
have INFL, or at least a grammar larger than VP.
Study 3: Modals

Thus far. the most straightforward predictions of competence-deficit theories have

not been borne out by the data. American children show inconsistent use of
subjects only at lLlLUs below 2. and even then they use subjects and pronominal
subjects much more than Italian children do. They also use ex-
pletive subjects very early in acquisition, and have correctI!. case-marked pronom-
inal subjects. Nevertheless. the children’s early usage is limited. Children’s use
of modals potentially provides another way of assessing the significance of the
subjects they use. since most competence-deficit theories have tried to explain a
presumed absence of modals in young children’s speech. It is therefore germane
to measure children’s use of modals. first to determine Lvhat the facts are. and
second to assess the hypotheses about children’s early grammars.
In English. modals originate under the INFL node. rather than under the VP
node (see. e.g.. Pollock, 19S9). Modals do not have an infinitival form (* “to
can”). nor do they inflect for person and number (:’ “she cans”). Modals in
English do carry tense, and exist in present (can, cvill. mc7y) and past (could,
broil/d. miglzr) tense forms. In Italian (and French) modals are main verbs and
originate under the VP node. In Italian (and French), there is an infinitival form.
and inflections for person and number.
The pro and VP hypotheses appear to be on their weakest ground with respect
to modals. While both have predicted absence of modals. there is little theoretical
basis for the prediction. In Hyams’ (1956) formulation of the pro hypothesis
children projecting a null subject language had an element in INFL that simul-
taneously licensed pro subjects and ruled out the co-presence of a modal. Thus,
an American child with the null subject parameter set incorrectly would also have
the INFL element that prevented classification of modals as part of INFL. Hyams
went on to argue that the children would also be unable to treat modals as main
verbs, because of their lack of inflection. Having no place to put the modals. as
it were. the children would fail to produce them altogether. Once their grammar
changed so that the element in INFL licensing null subjects was absent, modals
could then be classified.
As Lebeaux (1957) points out. however, the reasoning does not go through.
Nothing in principle stops the child from initially treating modals as defective
main verbs within VP. Indeed, if the child has the null subject value of the
parameter, that should force her to classify modals as main verbs.
To rule out modals as main verbs, one would have to introduce an ancillary
assumption stating that a verbal element must behave like other main verbs in
all respects for the child to classify it as a main verb. Medals do behave like main
verbs in some respects (e.g.. they carry tense, though we do not know if the child
recognizes this). The assumption therefore has to be that the child fails to classify
modals as main verbs because they do not act like main verbs in nil respects.
 56 L’. li7lif7l7

Further. the assumption would also have to apply. generally to all verbs. requiring
a child to see each verb in all its forms before classifying it as a verb. Since the

assumption is not only unmotivated syntactically. but is also unlikely as a principle

of acquisition. there are no good grounds for predicting absence of mod&. only
grounds for ruling them out as part of INFL.
As Lebeaus (1987) also points out. there is a more serious problem with the
pro hypothesis. Say that the child is unable to classify modals as main verbs and
say that the incorrect setting of the null subject parameter blocks the analysis of
medals as part of INFL. If modals are recognized as having any verbal character
at all. then they should force a change in the grammar. and. given their frequency
in adult speech, do so reasonably quickly. In other words, on the logic of Hyams’
(19S6) analysis, modals should act as a telltale and cause the child to reset the
parameter. (While Hyams footnotes a similar possibility, she does not recognize
it as a problem.)
The situation is somewhat different for the VP hypothesis. since it is often
accompanied by a maturation mechanism (Guilfoyle & Noonan. 1989). Rather
than the input forcing a resetting of the parameter, the child’s grammar changes
as a function of maturation (see Borer & Wesler. 1987, for discussion of matura-
tion). With such a mechanism the input is not a causal agent in grammar change.
Thus, while the VP hypothesis alone does not allow a prediction that the child
will ignore medals. the maturation hypothesis does (and in fact does so whether
associated with the VP or the pro hypothesis).
The argument would take the following form for the VP hypothesis. The child
has only a VP grammar; she cannot classify modals as verbs because of their
inadequate verbal morphology (this reasoning has already been questioned): she
cannot classify modals as modals because her grammar lacks INFL - therefore,
no modals. Once the grammar matures to include INFL, she can recognize mod-
als and include them in her output. The argument could be used for the pro
hypothesis as well. The child has pro; she cannot classify modals as verbs; she
cannot classify modals as modals because her grammar will not allow them as
part of INFL - therefore, no modals. Once the grammar matures to allow modals
as part of INFL. the child can recognize them and include them in her output.
The similar nature of the two arguments shows that the hypothesized form of
the child’s grammar is insufficient to predict absence of modals. The critical work
is done by: (a) the assumption that the child cannot classify modals as verbs. and
(b) the maturation mechanism. The prediction follows mainly from the claim that
at Time 1 the child’s grammar is too immature to allow correct classification. and
at Time 2 it is mature enough to allow correct classification. In this case. at least,
appeal to a maturation mechanism is an ad hoc and circular answer to the ques-
tion of how children’s grammars change.
Given the weak theory-internal basis for predicting absence of modals. it is
reasonable to ask why absence has been predicted. The principal reason seems
 Syrltnctic subjects in earIF speech 57

to be the presumption, drawn from previous summary data (e.g., Bellugi, 1967),
that modals are absent, even when semi-auxiliaries (such as u~~nna, gonna, and
hcrftrr) are present. Here we measure the use of modals and semi-auxiliaries to
determine whether there is an absence to be accounted for.

Method

The corpora from Study 1 were used. Utterances labeled discards were elimi-
nated. as were imitative and imperative utterances.

Measurement of modais
Using computer-assisted search procedures, we counted the number and types
of modals the children produced. We searched for can, can’t, could. couldn’t, ‘ld,
will. ‘11,won’t, would, wouldn’t, shall, should, shouldn’t. may, might, must. and
ought. We searched the Italian corpora for all forms of dovere (“must”) and
potere (“can”).

Measurement of semi-auxiliaries
We searched for gonna, wanna, hafta, and gotta. We also searched for and
included in our tabulations going to, want(ed) to, have to. had to, and get to,
even though those forms could also be considered as full verbs with infinitival to.
We did not search the Italian corpora for such verbs.

Data tabulation
To calculate the proportion of modals and semi-auxiliaries we used the number
of utterances containing a verb as the denominator. (Some utterances contained
more than one verb; such utterances were only counted once.) The numerator
consisted of the number of modals or semi-auxiliaries (seldom more than one per
utterance).

Results

All American children except the lowest-MLU child produced modals. In every
group, can and ‘II/will were the two most common modals. Figure 4 graphs the
percentages of modals, semi-auxiliaries, and, for comparison, verbs, in each
group. Modal usage was infrequent in Group I: only 3% (s.d. = 3) of utterances
with verbs contained modals, a total of 14 tokens. There was a steady gradual
increase: to 6% (s.d. = 2) in Group II (61 tokens), 9% (s.d. = 4) in Group III,
and 14% (s.d. = 2) in Group IV.
Italian children produced even fewer modals than Group I Americans, at both
Time I and Time II. At Time I there was only one modal: at Time II 1.5% of
utterances with verbs contained modals. Since the status of the modals in the
American children is the main interest, we will present those results in detail.
5s v. ~idian

Figure 4. r\fnericurl children’s prohcrion of verbs, semi-Am’s, cd tr~otirrls.

Group I Group II Group III Group IV

Verb

- Semi-Auxiliary

I Modal

American medals
In Group I. can, ‘II, and will were produced. In Group II, can’t. could. won’t,
would, and may were added. In Group III, couldn’t, should, might, and must
were added. In Group IV, shall and shouldn’t were added, but might and must
did not appear. Fewer past tense modals were produced than present tense ones.
The first past tense modal appeared in the child at MLU 2.52.
Contrary to what one might have expected from Bellugi’s (1967) data, the
earliest medals were not contracted with a negative. Not until the child at MLU
2.28 did a modal appear contracted with a negative; can’t appeared with the child
at MLU 2.28, and won’t appeared with the child at 2.52. Only one child, at MLU
2.76, produced only modals contracted with a negative: can’t and won’t.
Modal usage increases gradually and steadily as a function of age and MLU
across all four groups, and there is no hint of a step function. The Pearson
product-moment correlation between MLU and modal usage was .71. p < .OOl.
and between age and modal usage r = .66, p < .OOl (two-tailed). The partial
correlation between MLU and modal usage (with age partialled out) just missed
significance, r = .43, p = .056. The partial correlation between age and modal
usage (with MLU partialled out) was low, r = .28. n.s. Thus, MLU and age
jointly correlate strongly with modal usage, and MLU alone shows a stronger
relation with modal usage than does age alone.
Modal usage neither suddenly begins when subject use becomes more consis-
tent in Group II, nor dramatically increases when subject use becomes more
consistent. To examine the relationship between modal usage and subject usage,
we correlated the proportion of subjects in utterances with verbs (data shown in
Figure 2) with modal usage. The simple correlation was high, r = .SS. p = .006.
However, when MLU and age were partialled out, the correlation dropped to
-.04. Thus, modal usage has no independent relation with subject usage. Chil-
dren’s use of subjects is roughly constant regardless of how many modals they
produce.
To investigate more closely the relation between subjects and modals we also
calculated what proportion of utterances with modals included a subject. The data
can only be suggestive in Group I, due to the small number of tokens. In Group
I, the average proportion was .94; in Group II, .95: in Group III, .9S; in Group
IV, .99. Children use a higher proportion of subjects in utterances with a modal
than they do on average in all their utterances with verbs (compare with Figure
3).

American semi-Au’s
The lowest-MLU child produced no semi-Aux’s, but all other children pro-
duced them. The results are graphed in Figure 4. As with modals, production of
semi-Aux’s was infrequent in Group I: 5% (s.d. = 6) of utterances with verbs
contained semi-Aux (19 tokens). As can be seen from Figure 4, usage of semi-
Aux’s shows a complex relation with age and MLU. Usage peaks during Group
II (16%, s.d. = S) and then gradually diminishes in Groups III (12%, s.d. = 8)
and IV (S%, s.d. = 5). There was also much more variability of semi-Aux usage
within each group than there was for modal usage. Gonna and wanna were the
two most common semi-Aux’s.
We calculated what proportion of utterances with semi-Aux’s included a sub-
ject. Again, for Group I, the data are only suggestive. In Group I. the average
proportion was .63; in Group II, .89; in Group III, .87; in Group IV, .95. The
proportion of subjects in utterances with semi-Aux’s roughly parallels the propor-
tion of subjects in all utterances with main verbs. in contrast to the proportion
of subjects in utterances with modals, which is higher than the proportion of
subjects in all utterances with main verbs.

Italian modals
The Italian children produced fewer, rather than more, modals than did the
Group I Americans. For the Italian children at Time I there was 1 example of a
modal (“dopo poi ?” - “later can (you)?“). At Time II there were 6 examples -
60 V. Vdian

about 1.5% of the verbs (-l produced by 1 child: 1 each produced by 2 children.
out of a total of 392 utterances with verbs). A possible seventh example (“devi
mica” - “(you) must not”) was interpreted as an imperative and therefore not
included.

Discussion

Contrary to the predictions and claims of competence-deficit models. there ap-

pears to be no direct or primary relationship between onset of use of modals and
obligatory use of subjects. Since neither model had strong grounds for predicting
the absence of modals. their presence in American children’s speech need em-
barass neither. The hypotheses were used to explain a presumed fact: once the
fact vanishes. there is nothing to explain.
The children use more subjects and more modals as MLU advances. but once
MLU or age is partialled out there is no correlation between modals and subjects.
Further, the American children’s usage is greater. not less, than Italian children’s
usage, even though modals are main verbs in Italian. (It should be noted. how-
ever. that there are more modal forms in English - can, will, shall. and may -
than in Italian - porere and tioverr.) Finally, modals are not first apparent as
carriers of negation, nor do semi-Aux’s appear before modals appear. The conclu-
sions based on the three children Bellugi (1967) examined are not verified by a
larger sample of children. and the differences Hyams (1986) claims between
American and Italian children are not confirmed in our study.
Our claim that Group I children have the category modal hinges on whether
the early uses are genuine uses, in which the words are classified as modals or
main verbs, rather than simply being uncategorized words. Only if the child’s
initial correct, albeit limited and sporadic, use of a form (e.g., modals) is genuine,
do we have evidence that. say, the child has an INFL node.
We have adopted the criterion of any amount of correct use. Although it runs
the risk of letting in purely formulaic uses, there are also considerations in favor
of it. First appearance correlates highly with more stringent criteria (Stromswold.
1989). suggesting that the child’s initial correct uses are the genuine seeds of
knowledge out of which further knowledge and performance develop. It is also
extremely unlikely that the first appearance in a corpus is the first time the child
herself has used the form. Our samples are a minute fraction of the child’s
productions. Sampling considerations alone would suggest that a child has been
using a form before our data demonstrate it. Further. acquisition of any form has
to start somewhere. There is no reason to think that the initial genuine uses would
be widespread and consistent; quite the contrary.
At present any criterion beyond initial correct use appears arbitrary. Since
Group I children made no distributional errors with the modals they used, we
have called their use genuine. Although genuine, however. the modals may have
 Syntactic subjects in early speech 61

been classified as main verbs by the children rather than modals. But there is
distributional evidence that the children have the modals categorized correctly.
First, they use a different proportion of subjects before modals than before verbs:
second, they use a different proportion of subjects before modals than before
semi-Aux’s. Modals and semi-Aux’s are similar, in that they both precede main
verbs, but they are treated differently by the children. Semi-Aux’s are treated as
main verbs, but modals are not. O’Grady et al.‘s (1989) data roughly confirm
ours. Third, the children do not use an infinitive after a modal.
We can advance an explanation for the asymmetry in subject usage with mod-
als and semi-Aux’s, harking back to an issue raised in the general introduction.
It is less acceptable in adult speech to begin an utterance with a modal than with
a semi-Aux. While both are ungrammatical, “Will play” is less acceptable than
“Gonna play”. The children’s input probably reflects that difference: a parent
might ask a child, “Gonna play?“, or “Wanna play?“, but will never ask her.
“Will play?” The children appear sensitive to that asymmetry.
One reason, then, that the Group I children might have fewer modals than
other groups is that producing a modal requires them to produce a subject as
well. In contrast, producing a semi-Aux does not require them to produce a
subject. A different reason for the asymmetry is that the extended projection
principle obligates the child to include a subject if a modal is used.
Although we cannot, from our data, unequivocally conclude that the children
have correctly categorized modals as modals, we should note that there is no
distributional evidence in favor of the main verb analysis. The forms look like
modals, and with development there is no indication of reanalysis; rather, a
steady progression is seen. Thus, we tentatively conclude that even Group I
children have a rudimentary use of modals.

Summary

Both the data in Study 2, indicating basic, and correct, nominative case-marking,
and the data from this study, indicating primitive but correct use of modals from
early in Group I on, argue that the child has an INFL node before MLU 2. If
that is so, then neither competence-deficit analysis of Group I’s inconsistent use
of subjects is supported. The children’s inconsistent use of subjects would reflect
neither the presence of pro in the grammar (since pro is incompatible with a
modal in INFL on one analysis; Hyams, 1986), nor an immature grammar consist-
ing simply of a VP.
At the same time, the lowest-MLU child has consistently lacked the elements
that the other children have produced. The VP hypothesis thus remains a good
candidate description of the child’s first grammar, and allows us to suggest how
competence and performance factors might interact. The first grammar is ideally
suited to the child’s production constraints: the grammar will only allow short
62 V. Vdian

utterances and the child’s limited production abilities only allow short utterances.
The child’s reach and grasp have the same span. After the child’s grammar ex-
pands, however, the child’s reach exceeds its grasp. The grammar allows for
longer utterances than the child can consistently produce.

Study 4: To, tense, and subordinate clauses

INFL plays a central role both in the adult grammar and in language acquisition
accounts. The principal elements in INFL are tense features and agreement fea-
tures. Tense can be realized either as -tense, with infinitival ro. or +tense, with
the two possible values being present or past. INFL elements assign nominative
case to subjects; the subjects of untensed clauses do not receive nominative case
but objective case from the preceding verb (contrast, e.g., “I wanted her to go”,
with, *“I wanted she to go”).
The two competence-deficit theories make opposite claims about the existence
of INFL. The pro hypothesis predicts the presence of INFL. and the VP
hypothesis the absence of INFL. Since pro requires a licenser, which is presumed
to be a feature in INFL, INFL must be present if pro is. Since the VP hypothesis
limits the child to lexical categories. INFL is required to be absent. The most
direct prediction the pro hypothesis makes is therefore the presence of INFL
features. such as tense or agreement. in child speech. The most direct prediction
the VP hypothesis makes is the absence of those features.
In both cases, however, the predictions are only possible if there are no other
factors affecting children’s productions. But Lebeaux (1987) and Kazman (1988),
for example, note that the child could have tense or agreement features as an
affix on the verb, in the absence of an INFL node. Ancillary assumptions have
to be appeared to in order to secure a prediction of absence or presence of an
element in speech. Neither hypothesis can make predictions about the child’s
productions solely on the basis of the hypothesized formal representation. In the
case of O’Grady et al.‘s (1989) proposal concerning the relationship between
subject use and tense, their prediction would be little or no use of tense in
children who use subjects under 90% of the time.
Tense is also important in Roeper and Weissenborn’s (1990) analysis. On their
analysis there is one and only one characteristic of null subject languages: null
subjects (pro) can exist in embedded tensed clauses; in non-null subject languages
they cannot. The interesting language acquisition consequence of Roeper and
Weissenborn’s claim is that only one feature in the input can serve as a telltale
to set the correct value: the presence of null subjects in tensed subordinate
clauses. Roeper and Weissenborn claim empirical support for their position. For
example, they state that French and German children continue to use null sub-
jects even after they use tense productively and use expletives. Therefore, exple-
 Sytnctic subjects in early spwch 63

tives cannot be serving as a telltale for null subjects, nor can INFL be sufficient
for setting the correct value.
With respect to the American child, Roeper and Weissenborn have predicted
(1990) that obligatory subjects will appear once the child has subordinate clauses.
Since, on their analysis, the telltale of a null subject language is that a null subject
can appear in tensed subordinate clauses, Roeper and Weissenborn predict a
linkage between the child’s production of subordinating conjunctions and com-
plementizers (such as that) and the child’s consistent production of subjects in
subordinate clauses and in matrix clauses.

Method

The corpora from Study 1 were used. Utterances labeled discards were elimi-
nated, as were imitative and imperative utterances.

Znfinirival fo
We searched for ro, and tabulated the number of infinitives. We did not in-
clude wanna or gonna, but we did include cases of wanr to, where the transcript
thereby indicated that two separate words were used.

Tense and agreement

We used a custom-made frequency count program to list all the lexical items
in each child’s corpus. We manually examined each child’s list for any words that
could be a past tense verb or a verb marked for third person singular. We then
searched each corpus for the candidate verbs, and tabulated the instances of past
tense verbs and third person singular verbs.

Subordinate clauses
We searched each child’s corpus for subordinating conjunctions which intro-
duce a tensed clause: that, what, because, who, where, when, why, how, if, so,
for, after, and before. We tabulated the instances of subordinating conjunctions
in tensed clauses and calculated the percentage of subject use in that context.

Results

Infinitival fo
The infinitive was used by every group, but infrequently in Groups I and II.
When infinitival use is relativized to verb use, the percentage of infinitives by
group is 1.7, 1.4, 4.6, and 5.6. To usage increased irregularly with group, and
most markedly between Groups II and III. There was a great deal of variability
in frequency of usage.
In Group I, there was a total of 7 to’s (produced by 30 children; range per
64 V. Vulian

Table 5. American children: mean token freqtrenc_v anti proportion use of put tense

Mean frequency per child

Regular Irregular Total Mean total proportion”

Group I 2.6 4.4 7.0 .09

Group II 4.2 9.0 13.2 .06
Group III 5.3h 13.6 LS.9 .06
Group IV 12.7’ 10.7 23.3 .09

“The numerator is total number of past tense examples and the denominator is number
of utterances with verbs.
?hese figures include irregular verbs which Here incorrectly given a regular past tense
suffix.

child from 0 to 3). Even the child at MLU 1.53 produced two examples of to:
“(Oranges to) eat” and “Right, trying to go through it”. In Group II, there were
17 to’s (produced by 41.5children: range from 0 to 8); in Group III, 92 (produced
by 818 children; range from 1 to 41. with 65 being produced by 2 children); in
Group IV, 40 (produced by 3/3 children; range from 4 to 24).

Tense and agreement

Every child in every group produced tokens of past tense. Range of usage
varied from 2% to 22% of verbs (with no correlation with MLU). Two children
produced no examples of regular (-en) past tense: MLU 1.53 and MLU 3.72. All
other children produced examples of both regular and irregular verbs in past
tense. Table 5 shows that while the average frequency of past tense use increased
with group, the percentage of verbs that were tensed did not. Since the percentage
of utterances with verbs increases dramatically with group, it is necessary to use
verb production as a baseline. (The relatively high figure of 9% for Group IV is
due to one child who produced past tenses for 22% of her verbs: the other 2
children produced 4% and 2%.)
There were no errors of incorrect form of the past tense of irregular verbs
(e.g., “runned”, “maked”) until Group III, where 3/8 children produced one such
mistake, and 2/S produced 3 each. Even here, then, the errors were infrequent.
Only a total of 9 incorrect tokens were observed in Group III, compared to 109
correct tokens, a proportion of .08. In Group IV, l/3 children produced tensing
errors: that child persistently mistensed one verb 15 times.
All but one child (MLU 1.81). produced at least one example of third person
present singular -s. Across all the children there was a single error of inappropriate
use of -s: “I shows you a cookie” (MLU 2.76). As with past tense, there was an
increase in frequency with group, but no increase in proportion until Group IV.
In Group I the children produced an average of 3.8 -s’s, or 4.2% of their verbs.
 S_vntczcticsllbjects in early speech 65

Group II produced an average of 8.4, or 4.2% of their verbs. Group III produced
an average of 13.3, or 4.4%, and group IV produced an average of 22.6. or 9%.

Subordinate clauses
No child ever failed to use a subject after a subordinating conjunction introduc-
ing a tensed clause, but there were few examples in Groups I and II. In Group I
there were only 4 instances of a tensed subordinate clause (among 3/.5 children,
including the lowest-MLU child); in Group II there were 11, in Group III 77; in
Group IV 40.

Discussion

In all groups the children produced examples of all forms. but Groups I and II
produced few infinitivals and few subordinate clauses. (The lowest-MLU child
produced two infinitives, no regular past tense forms, one third person singular
form, and one subordinate clause.) When there were major increases in frequency
of usage of to, tense, and agreement (relative to number of verbs used), they
occurred either between Groups II and III, or between Groups III and IV. Without
ancillary assumptions, neither competence-deficit model accounts for the pattern
of the data.
The children produced a relatively stable percentage of past tense verbs (5-10%
of all verb uses) and third person present singular verbs (4% of all verb uses) over
Groups I-III. There were few examples of past tense and agreement in Group I,
but considering how few verbs those children used, they would have had to produce
a very high percentage of tense and agreement to bring their frequency figures up.
The increase of use of to between Groups II and III corresponds to findings by
Bloom, Tackeff, and Lahey (1984) showing a similar increase in ro somewhere
between MLU 3.0 and 3.5.
As with subject use, it is clear how an impression of lack of tense, agreement,
and infinitives could be gained from perusing young children’s corpora. There are
few examples in low-MLU children, and many more examples with higher-MLU
children. In the case of past tense, however, there is no proportional increase in
our data, once use of tense is relativized to verb use. O’Grady et al. (1989) state
that the three children they observed produced almost no examples of tense until
they began using subjects 90% of the time. The statement is slightly misleading,
however, since their figures do show inflected verb use for all three children, and
many examples for one child. Their tabulated data appear to agree with ours,
though their conclusion differs.
We can briefly consider whether the children’s early uses of ro, tense, and
agreement are genuine. With respect to tense, we note that errors of creating a
regular past tense for an irregular form did not occur until Group III. Even in
Group III there were very few examples - about 9% of the irregular verbs; errors
66 1;. Vnlinrf

of overgeneralization are always fewer than correct uses. Marcus et al. (1990).
analyzing data from a child database, similarly report low rates of overgeneraliza-
tion over a wide age range. including ages comparable to children in this study.
It is often claimed that pre-overgeneralization children have simply memorized
the past tense forms, and have performed no analysis that identifies the verbs as
past tense (see Marcus et al.. 1990. for discussion). That claim seems untenable
unless there are specific errors, as have been attested for some children for some
verbs (Kuczaj. 1981). of actively treating the past tense forms as untensed bare
forms, by, for example, affixing a third person singular s, or ing, to the past tense
verbs. Such errors appear to be rare and sporadic, however. and do not occur in
our corpus. We favor the analysis of Marcus et al. that children’s “overgenerali-
zation” errors are actually retrieval errors. and consider the children’s past tense
uses imperfect but genuine.
With respect to agreement, the children’s uses also seem genuine. Only one
error of using the s ending inappropriately was found, and it is difficult to see
how the child could restrict use of s to third person present singular without
having coded the restrictions. That need not mean that the agreement feature is
present in INFL, only that it is analyzed as a labeled agreement affix on the verb.
Both tense and agreement, then, could be analyzed as features on the verb. We
note that Meisel (1990) has independently concluded, in a study of three bilingual
French-German children, that children show evidence of INFL before MLU 2.
To some extent the issue of genuineness is orthogonal to an evaluation of the
two competence-deficit explanations for Group I children’s inconsistent subject
usage. Assume first that the early usage of IO and tense in Groups I and II is not
genuine, and that only in Group III, when there occurs a large increase in use
of infinitives, the first errors of over-regularization of the past tense, and wide-
spread use of subordinate clauses, is INFL well documented. That would be a
problem for both hypotheses.
The problem for the VP hypothesis is that those changes occur at the wrong
point. Instead of occurring between Groups I and II, which is when the major
change in frequency of subject usage occurs, they occur between Groups II and
III. The VP hypothesis would be forced to claim that the consistent subject usage
in Group II still reflected a VP-only grammar, and would have no explanation
for the large increase in subject usage from Group I to Group II.
For the pro hypothesis, which, on my interpretation, requires an INFL node
to license a pro subject, there is a different problem. If there is no INFL until
Group III, then there is no basis for a pro subject until Group III. In that case
the inconsistent subject usage in Group I must have a different source than a pro
subject, and the shift in usage between Groups I and II could not be explained
by the pro hypothesis.
Assume now that the early uses are genuine (with the possible exception of
the child at MLU 1.53). The VP hypothesis still cannot explain Group I subject
 S_vntactic subjccrs in etrrly speech 67

inconsistency, because INFL will already have been in existence. and therefore
something else - such as production limitations - will be accounting for subject
inconsistency. The pro hypothesis might be thought to fare better. If INFL exists
even in Group I, then a pro subject is possible for Group I children. We would.
however, still be left with the absence of a development between Groups I and
II that would indicate a restructuring of the grammar.
Thus, the timing is off for both competence-deficit hypotheses. Whether we
suppose the early uses of to and tense are spurious or genuine. we do not have
the relationship between those elements and use of subjects that either hypothesis
would require.
The data on the production of subordinate clauses were intended to bear on
Roeper and Weissenborn’s (1990) analysis. Our data suggest either that Group I
children already have knowledge of subordinate clauses. or that knowledge is
delayed until Group III. If the former. then Group I children should be taken
to understand that subjects are obligatory: their inconsistent production of sub-
jects would require another explanation. If the latter, then the Group II children
are a problem, because they appear to have an obligatory subject. but produce
few subordinate clauses.
If the early subordinate clauses are taken to be genuine, and if the children
in Group I are taken to understand that subjects are obligatory. there is some
support for Roeper and Weissenborn’s (1990) prediction that subordinate clause
use will be linked with subject use. If. however, subordinate clauses are inter-
preted as not developing until Group III, when they dramatically increase in
frequency, then Roeper and Weissenborn’s prediction is disconfirmed, since the
children in Group II show good evidence of understanding that subjects are
obligatory.
How should we understand the low production of subordinate clauses in
Groups I and II? Subordinate clauses both require knowledge of individual sub-
ordinating conjunctions, and, unless the clause is used alone (as it was for the
child at MLU 1.53). of embedding. Neither would be expected to be well devel-
oped at very low MLUs. A similar reason can be offered to explain why there
are so few infinitives in Groups I and II (less than 2% of the utterances with
verbs). Infinitives could increase because the higher-MLU children can handle
the concomitant increase in sentence complexity and length that use of infinitives
entails. We therefore favor the interpretation that the children’s early uses are
genuine.

Summary

If the children’s uses of to, past tense. and third person singular present are
considered to be genuine from the time of first appearance, the children then
must be credited with having INFL at least in the middle of Group I. If the uses
68 V. L’fllirin

are not considered genuine until they are more frequent. then the children do
not show clear presence of INFL until after they are already using subjects con-
sistently. In either case, both competence-deficit hypotheses fail to account for
the facts.
We appealed very broadly to considerations of complexity and length as an
explanation of increase in usage of infinitives and subordinate clauses. but in so
doing we did not account for the precise timing of the increases. The next study
spotlights performance issues more directly.

Study 5: Verbs, direct objects, and subjects

Thus far we have found no support for the pro hypothesis. There is some slight
support for the VP hypothesis in the data of the lowest-MLU child, who produced
the smallest number of subjects, the smallest percentage of nominatively cased
subjects, no medals or semi-Aux’s, no regular past tense, only two to’s and one
subordinate clause. While one child’s data are only suggestive, there is at least
one child whom the VP hypothesis appears to describe reasonably accurately.
If that child does have a VP grammar. then the other children in Group I,
who look importantly different from that child. have a grammar larger than a
VP. In that case, they either have some other, unknown, competence def-
icit, or there are performance factors that constrain how often they produce
subjects and other sentence elements. In other words, if we accept the lowest-
MLU child as having a VP grammar, we must accept the others as having a fuller
grammar, and correspondingly search for a different source for those children’s
subject omissions.
In Study 5 we look more directly at performance measures: does VP length
vary as a function of type of subject: is subject use related to age, MLU, or verb
use; are subjects and objects omitted equally often?
Performance accounts predict that the longer the VP the less likely a lexical
subject will be chosen. L. Bloom (1970) and P. Bloom (1990) have proposed that
the longer the VP, the more cognitive load is imposed, and therefore the greater
the likelihood of dropping a subject. VP length was analyzed by L. Bloom (1970)
for one child’s utterances with the verb make, and by P. Bloom (1990) for three
children’s utterances with past tense verbs and verbs that cannot be used in
imperatives (to exclude ambiguous utterances). In both cases, longer VPs were
associated with subject absence. In addition, P. Bloom found that VP length was
shortest when the child used a full lexical subject, longer with a pronominal
subject, and longest with no subject. Since in both studies the sample size was
small, and since past tense forms are, as we have seen, a minority of children’s
productions, it would be desirable to replicate those findings.
(Bloom, Miller, and Hood (1975) demonstrated that many factors affect chil-
dren’s utterance length. Unfortunately, the large number of coding differences
between their study and the present one precludes direct comparisons. Further
research, however, should be aimed at systematically examining the effects of
discourse and different types of complexity on subject use in particular. Here we
restrict ourselves to a small subset of possible performance predictions.)
Performance accounts also predict that subject use will increase regularly as
age and MLU increase. Since MLU is a measure of utterance length, it is obvious
that increasing subject usage will increase MLU. But MLU could also increase
independently of subject use, since increased use of any sentence element (e.g..
complement clauses) will increase MLU. Age, though an imperfect measure of
performance capability, is independent of any of the language variables. If a
positive relation between age and subject use is found throughout the age range.
that would indicate a performance component.
Performance accounts thirdly predict a relation between verb usage and subject
usage. Verb production is operationally independent of subject use as measured
here. The children could maintain a constant proportion of subject usage as their
verb production goes up. But verbs introduce structural complexity into an utter-
ance, especially if the child understands that utterances with verbs require sub-
jects. If children’s omission of subjects is related to a general limitation on the
number of major sentence constituents that can be included, then one would
espect subject usage and verb usage to be highly correlated. As children become
able to handle the complexity involved in including verbs, they should corres-
pondingly become able to handle the complexity involved in including subjects.
That development should hold for both American and Italian children, since both
should experience production limitations. However, it should be more extreme
in American children, because the American range of subject use can be much
greater.
Performance accounts might also predict asymmetries between use of subjects
and objects. Pragmatic factors would lead to asymmetries because the given, and
hence dispensable, information will tend to be the subject, and the new, and
hence important, information will tend to be the object. Production factors could
also be important, since the beginning of an utterance will be more effortful than
the end. Finally, the utterance-initial location appears prone to omission and
reduction effects, perhaps for prosodic reasons (Gerken, 1991).
Competence theories have been inconsistent in their predictions about asym-
metries between subject use and object use. One theory (Kazman, 1988) has
directly predicted absence of objects as well as absence of subjects. Hyams (1987).
in contrast, explicitly claimed an asymmetry between subjects and objects. Rad-
ford (1990) has argued that both subjects and objects in fact are absent in early
child speech, since the child freely allows “implicit” arguments. Thus, there is no
uniform prediction from one competence-deficit account to another.
Finally, a performance account predicts two other developmental changes. The
first is an increase in the use of purely transitive verbs. verbs for vvhich objects
are obligatory: one way the beginning speaker can lighten the burden of produc-
ing objects for verbs is to produce more verbs that do not require objects. The
second is an increase in how often children supply objects for “mixed” verbs -
verbs that are grammatical with or without an object.

Method

The corpora from Study 1 were used. Utterances labeled discards vvere elimi-
nated, as were imitative and imperative utterances.

Relation between type of subject and length of VP

We examined simple one-clause, one-verb sentences, excluding utterances with
modals or negatives, and excluding imperatives and imitations. We computed the
length in morphemes of verb phrases with a full lexical subject. a pronominal
subject, or no subject; all verbs were used, including be as main verb. All material
in the VP (complement or adjunct) was included.

Verbs and objects

We divided the children’s verbs into three categories: those which we consid-
ered to be pure intransitives; those which we considered to be pure transitives;
and those which could be transitive or intransitive. For Group I an average of
64 verb tokens per child contributed to the analysis: for Group II, 13-l: for Group
III, 160; for Group IV, 168. We tabulated the proportion of verbs of each categ-
ory as a function of MLU group.
We also tabulated the number of times direct objects appeared with the verbs
of each category. Examples of utterances we classified as having an object with
an intransitive verb are “he. he talking a duck”, “go school”, “yeah, it looks a
lady”, “I come teeth”. Be was excluded from the tabulation; as a copula, it does
not take objects in the same sense as the other verbs. The verbs put and get were
also excluded; put requires two objects and thereby differs from the other verbs;
get proved very difficult to score. Examples with an S complement or an adjunct
phrase (e.g., “I go now”) were excluded.

Results

Relation between type of subject and length of VP

The performance prediction that subject presence or absence would be related
to VP length was supported, replicating L. Bloom (1970) and P. Bloom (1990).
The predicted ordering is: longest VP with no subject, intermediate with a pro-
nominal subject, and shortest with a full lexical subject. There are six possible
orderings, so that the probability of any particular ordering appearing is .17.
 For Group I, which would be expected to show the pattern most strongly. the
average VP length was 2.53 with no subject. 2.45 with a pronominal subject. and
2.3 with a lexical subject. with 3 of the 5 subjects showing the phenomenon. For
Group II, the figures were 3.05. 2.S-l. and 2.67, with. again, 315 children showing
the phenomenon. Since there are six possible patterns. by chance 17% of the 10
children would show the predicted pattern: instead. 60% showed it. Application
of the binominal test shows that the probability of 60% of the subjects showing
such a pattern is less than ,003. For Groups III and IV there was no consistent
pattern. Only 3 out of the 11 children in those two groups showed the pattern
which was dominant in Groups I and II.

Correlations among subject use, MLU, age, nnd verb use

Both predictions concerning correlations with subject use were borne out in
our sample. The simple correlation between MLU and subject use is .77 @ <
.OOl), and the simple correlation between age and subject use is .74 (p < .OOl).
When a partial correlation between MLU and subject use is computed, the cor-
relation is .4S (p = .03), a still significant, though much lower, correlation. The
partial correlation between age and subject use is .41 (p = .075). MLU and age
together predict subject use better than MLU or age alone.
Children’s verb use increases markedly from Group I to Group IV, as Table
1 shows. Verb use is very highly correlated with MLU, even with age partialled
out (r = .Sl. p < .OOl), but there is no correlation between verb use and age
when MLU is partialled out (r = .20, n.s.). Verb use (unlike modal use) is highly
correlated with subject use. even when MLU and age are partialled out. The
partial correlation is .7S @ < .OOl). As children produce more utterances with
verbs, they are correspondingly more likely to produce subjects for those verbs.
Because of the small number of Italian children, it is not possible to assess the
relationship there between subject and verb use. Using a Spearman rank order
correlation we find that at Time I the correlation is .50, and at Time II .90, but
neither is significant at the .05 level (two-tailed). Nevertheless, it appears that
there is a rough relation which is similar to that found in the American children.

Verbs and objects

As would be expected on a performance explanation, the children produced
more purely transitive verbs as development proceeded. Figure 5 shows how the
proportions of the children’s verbs in the three major categories changed as a
function of group, with the largest changes occurring between Groups I and II:
the proportion of pure transitive verbs increases at that point, and the proportions
of the other two verb categories decrease.
First, the proportion of pure transitive verbs increases from .45 (s.d. = .13)
in Group I to 59 (s.d. = .lO) in Group II. There is another increase from 57
(s.d. = .lO) in Group III to .69 (s.d. = .ll) in Group IV. Second, the proportion
Figure 5. American children’s production of pure transitive. pure intransirive, and mixed
t.erbs.

100 -

5 1:: /_A

>

8 4oS *

2oI w
O- 1 I 8 I
Group I Group II Group III Group IV

Pure Transitive

- lntransfiransitive

- Pure Intransitive

of mixed verbs decreases from .35 (s.d. = .16) in Group I to .29 (s.d. = .lO) in
Group II; there is another decrease from .33 (s.d. = .ll) in Group III to .25
(s.d. = .06) in Group IV. Third, the proportion of pure intransitive verbs de-
creases gradually from .20 (s.d. = .lO) in Group I, to .ll (s.d. = .05) in Group
II, to .09 (sd. = .05) in Group III, to .06 (s.d. = .06) in Group IV.
As can be seen from Figure 6, the children were reasonably constant in their
use of objects for verbs which were either pure intransitives or pure transitives,
but they slightly increased their use of objects for verbs which could be intransi-
tive or transitive. The children seldom used objects with pure intransitive verbs:
Group I used objects 4% of the time (s.d. = 4); Group II 8% (s.d. = 5); Group
III 4% (s.d. = 6); Group IV 4% (s.d. = 8).
The children consistently used objects with pure transitive verbs, even in
Group I, where the highest individual percentage of omission was 14%. With
pure transitive verbs, Group I used objects 93% of the time (s.d. = 5); Group
II 93% (s.d. = 8). Group III 98% (s.d. = 2); Group IV 97% (s.d. = 4).
The children’s use of objects with mixed verbs increases somewhat from Group
 S_wtoctic subjects in enr!\ speech 73

Figure 6. Atnericnn children’s production of objects Hit/tdifferent verb types.

20 -

n
0 I I I I
Group I Group II Group III Group IV
with Transitive
with IntraWTrans
with Intransitive

I (.49, s.d. = .13) to Group II (.66. s.d. = .30), and remains roughly at that
level (Group III averages .66, s.d. = .27; Group IV averages .59, s.d. = .OS).

Discussion

Performance factors of some type do appear to play a role in children’s use of

subjects. First, we find support for the prediction that children will use shorter
VPs with full lexical subjects than pronominal subjects, and shorter VPs with
pronominal subjects than no subjects. Although only 6 out of the 10 children in
Groups I and II showed the predicted pattern, that is more than three times as
many as would be expected by chance, given that there are six possible orderings.
Further, in Groups III and IV, where we would predict a weaker performance
limitation, only 3 of the 11 children show the same pattern - a chance effect. For
full satisfaction, however, Group I should have displayed the pattern more than
Group II, and that did not occur.
Second, the other performance measures also showed effects. MLU and age
combined predicted use of subjects well. The fact that age is related to subject
use almost as strongly as MLU suggests that some language-independent factors
74 C’. C’dian

are involved in children’s use of subjects. Increased subject use would, all other
things being equal. contribute to increased hlLI_J. so that relation. although pre-
dicted. is not especially strong confirmation of a performance account. But sub-
ject use obviously cannot contribute to age. and thus the independent contribu-
tion of age is noteworthy.
More interesting still is the correlation between subject use and verb use.
Subject and verb use are linked across the entire range of American children.
There is some evidence that subject and verb use are also linked for the Italian
children. though within a narrower range of subject and verb use. but because
of the small number of Italian children it is difficult to assess. Some effect in both
language groups was predicted, since both should be subject to production limita-
tions. When the American correlation between subject and verb use is put to-
gether with their much higher rate of subject use. it suggests that the American
children know that verbs require subjects. but use them less at low MLUs because
of performance limitations.
If American children do understand that their sentences require subjects, that
raises the question of how they deal with direct objects. There is a marked
asymmetry in how often Group I children use subjects and how often they use
objects for pure transitive verbs. Since even Group I children produce objects
when required over 90% of the time, we see little to support the hypothesis that
they have pro, or any null or implicit argument. in object position. Bloom. 1Iiller.
and Hood (1975) similarly report a very high percentage of objects for verbs
which appear to be pure transitives. Radford’s (1990) report of null objects may
be due to the lower age of the children he examined; alternatively. since he
reports no quantitative data, the examples may be infrequent for each child.
Any theory, competence or performance. which predicts an asymmetry be-
tween subjects and objects between MLUs 1.5 and 2.0 is on firmer ground than
one which predicts that they will be treated the same. The question then is
whether the asymmetry means that Group I children fail to understand that
subjects are required. When all the data are considered. the answer appears to
be no.
The children’s performance on verbs of different types, and their production
of objects, is noteworthy in several respects. First, our data show that children
typically do not use a verb unless they know how it subcategorizes with respect
to objects. There are relatively few errors at any MLU of incorrectly using an
object with a pure intransitive verb (and the errors that occur might be best
described as omission of a preposition), and relatively few errors of incorrectly
omitting an object with a pure transitive verb.
The most telling contrast is that between provision of objects with pure trans-
itive verbs and mixed verbs. The children have those two classes separated: they
provide objects much more frequently for pure transitive than mixed verbs. indi-
cating that they recognize the difference between when an object is obligatory
and when optional. That suggests that the children are tracking parental input
closely. since the only way of distinguishing a pure transitive verb like “hold”
from a mixed verb like “eat” is by noticing whether adults consistently use an
object after each of them.
Second. our data show that the use of optional objects in mixed verbs increases
(from 49% to 66%) between Groups I and II. That increase seems best under-
stood as a decrease in performance limitations. As the children can produce
longer utterances, they provide more optional objects for mixed verbs. Third. the
children increase their use of pure transitive verbs (from 45% to 69%) as develop-
ment proceeds. That increase also seems best understood as a decrease in perfor-
mance limitations. As the children can handle longer length. they increase their
use of verbs which require objects.
We can now note that the child has different degrees of freedom with respect
to subjects and objects. Subjects are always grammatically required (though they
can be acceptably omitted in some circumstances) once the child’s grammar ex-
ceeds the scope of a VP. Objects are always required for purely transitive verbs
(both for grammaticality and acceptability), but the child has the option of using
more intransitive and mixed verbs to get around the cognitive load that additional
constituents would appear to impose (if there is a synonymous intransitive or
mixed verb in the child’s vocabulary).
Our data suggest that the Group I children make use of that option: they
produce the lowest percentage of purely transitive verbs and the highest percen-
tage of purely intransitive verbs: they also produce the lowest per-
centage of objects. With objects. the Group I children can grammatically avoid
the burden objects impose. With subjects, the child does not have a comparable
option. Our data on the increase in children’s use of purely transitive verbs, which
first occurs at the same time as the increase in children’s subjects, suggests that
children are able to increase the number of constituents that they can handle
around MLU 2. and then again around MLU 4.
It may also be the case that semantic factors interact with cognitive load. The
conceptual content of pure transitive verbs may be more complex than that of
intransitive or mixed verbs. It is not clear, however, that there are any meaning
differences which reliably correlate with transitivity, so this must remain specula-
tion.

Summary

Study 5 has presented evidence that performance factors are important in chil-
dren’s production of subjects. Children in Groups I and II produced their longest
VPs when no subject was present. next longest with a pronominal subject, and
shortest with a full lexical subject. Subject use is highly correlated with MLU and
age together, and with verb use. Children’s production of purely transitive verbs
76 V. Valirrn

increases most between Groups I and II. as does their production of optional
objects. The facts are best explained by performance theories.

General discussion

We have had three goals in this series of studies. The first was to determine the
facts about use of subjects and related sentence constituents among young Ameri-
can children. The second was to decide, from those facts, when American chil-
dren understand that their language requires subjects. The third was to evaluate
competence and performance explanations of children’s use of subjects.
At a factual level, our studies show that American children betvveen MLU
1.50 and 2.00 (ages 1:lO to 2;2) use subjects in almost 70% of their utterances
with verbs. Between MLU 2.00 and 3.00, subject use averages almost 90%, and
around MLU 4.00 reaches 95%. Subject use is never lower than 84% after MLU
2. We first conclude that the children with MLU greater than 2 understand that
subjects are required. and that neither the pro hypothesis nor the VP hypothesis
holds for those children.
Our second conclusion is that American children understand even earlier.
somewhere between MLU 1.5 and 2.0, that English requires subjects. A multi-
factored performance explanation is a more tenable explanation of the children’s
behavior than a competence deficit. Group I, taken as a whole. provides no direct
evidence in favor of the pro hypothesis (e.g.. Hyams, 1986, 1987) or the VP
hypothesis (e.g., Guilfoyle &: Noonan, 19S9: Kazman, 1988) and some evidence
against. Features that, on the basis of either of those two competence explana-
tions, would be expected to co-occur with inconsistent use of subjects. do not.
Further, the children do not lack tense (O-Grady et al., 1989). If our subordinate
clause data are taken at face value, there is support for Roeper and Weissen-
born’s (1990) suggestion that subordinate clauses are linked with obligatory sub-
jects.
Let us review the evidence for the conclusion that, despite their inconsistent
usage. Group I understands that sentences require overt subjects. Group I pro-
duced twice as many subjects as Italian children between the ages of 2:l and 2:5
(Study l), contrary to the pro hypothesis in Hyams (1986). They used pronouns
for most of their subjects. again contrasting with Italian children, who used pro-
nouns for a minority of their subjects (Study 1). They produced few expletive
subjects, but as many as higher-MLU children did (Study 2). Not only did low-
MLU American children differ from Italian children, but they also used subjects
more often than the Japanese children studied by Mazuka et al. (1986). The
subjects used by Group I look like real subjects rather than VP subjects. because
the children consistently used nominatively case-marked pronouns in subject po-
sition (Study 2), contrary to the VP hypothesis.
 Syntactic subjects in rat+ spercl~ 77

Both competence hypotheses have predicted (or assumed) lack of modals, but
the Group I children produced a few medals, and produced them more often
than Italian children (Study 3). Further, there was no correlation between how
frequently children produced subjects overall and how frequently they produced
modals (Study 3). The only link between medals and subjects. one best accounted
for by facts of English usage, was that children uniformly produced a subject if
their utterance had a modal. Group I children also produced examples of infinit-
ival to, past tense, third person singular present, and subordinate clauses (Study
4), all contrary to what the VP hypothesis would predict.
Some of the early infrequent uses of sentence elements that we have
documented may be spurious, rather than genuine: both competence-deficit
hypotheses would interpret them thus. The competence theories would then pre-
dict an abrupt quantitative or qualitative change in how children use those ele-
ments, a change which would occur at the same time as the large increase in use
of subjects between Groups I and II. But the kind of patterned timing of
emergence of elements that would be expected if either competence hypothesis
were correct does not emerge from our detailed examination of the children’s
productions, with the possible exception of the difference between the lowest-
MLU child and the remaining children. Almost all the observed changes in use
appear to fall into four categories, none of which fit the predictions:

(1) The changes are gradual rather than abrupt. An example is the increase
in number of types and tokens of modals.
(2) The changes occur at the wrong time. Examples are the increase in infinit-
ival to between Groups II and III; over-regularization of the past tense between
Groups II and III; the increase in third person singulars between Groups III and
IV; and the increase in subordinate tensed clauses between Groups II and III.
(3) The changes are predicted by performance accounts rather than by compe-
tence accounts. Examples are the higher than chance usage of longer VPs with
no subject than with a pronominal or lexical subject; the correlation between
MLU, age, and subject use; the correlation between verb use and subject use;
the increase in purely transitive verbs; the increase in objects provided for mixed
verbs.
(4) The changes are independent of any theory at present (which is natural,
as there will be some simultaneous changes that are due to other grammatical
developments). Examples of independent changes occurring between Groups I
and II are the increase in use of pronominal subjects (already high in Group I),
also evident in the Italian children; the increase in the use of semi-Aux’s between
Groups I and II.

While we have concluded that performance explanations account better for the
overall pattern of the children’s development, child,
we note that the lowest-MLU
at MLU 1.53, provided some evidence for the VP hypothesis. This child not only
7s V. Lidinn

used subjects less than any other child. but had few personal pronouns as subjects.
no modals, no semi-Aux’s. and no regular past tense. There were, however, two
infinitives, one third person singular. and one subordinate clause. While not
presenting a completely consistent picture. this child came closer to fitting the
VP description than did any other. Accordingly. the initial stage for all children
may be a version of the VP grammar illustrated in Figure 1 (e.g., Guilfoyle &
Noonan. 19S9).
Our suggestion, however, is that the child’s first syntax is a skeletal mobile
consisting of the entire tree in Figure 1 (minus modals), with the nodes unfilled
and with the order of SPEC. head. and complement left free. We favor this
version on the assumption that all the elements in the skeletal tree (except mod-
ais) are universal and innate. The developing child learns how to lexicalize the
nodes, learns what the proper phrase orders are in her particular language, and
learns what empty categories exist in her language. The child in our study at
MLU 1.53 is clearly deficient in knowing how to lexicalize different sentence
elements; whether the child also lacks the full skeletal structure we cannot say.
With respect to pro, the developing child will wait for evidence before project-
ing that empty category, and will require a licenser of some sort for it. The
American child does receive some misleading input, namely utterances without
subjects. But there is no evidence that American children are ever seriously
misled by such utterances to project pro. Further, it seems unlikely that the
American child can start off very much in error if she already has the correct
notion of subjects before MLU 2.0.
The Italian child receives a great deal of information that empty subjects are
possible, in the form of utterances without subjects (on the basis of Bates’s (1976)
data, about 50% of adult input) and in the form of “perfect” verb endings. Unlike
the American child, the Italian child not only receives evidence that null subjects
are possible. but evidence about how they are licensed. Since the empty subject’s
identity can be read off the verb endings, identification is guaranteed. and if
identification is guaranteed, so is licensing. (Whether such reasoning will actually
guarantee the correct outcome depends in part on whether it is universally true
that languages with completely transparent verb endings have pro subjects.) In
the case of the child learning Japanese. Chinese, and many other ‘null subject”
languages, we can say nothing further at present, because the nature of the empty
subject is not clear, nor how the grammar licenses the empty categor?;.
All other things being equal. we want to construct an acquisition mechanism
which projects the minimum number of incorrect hypotheses (Guilfoyle Br
Noonan. 1989; Lebeaux, 1989). because grammar reorganization is costly. At the
same time, we have to account for the errors the child produces. An incomplete
grammar, representing absence of knowledge, rather than an incorrect grammar.
representing false information, is one way to achieve both. That is the attraction
of the VP hypothesis: it postulates a limited grammar which can serve as a correct
 S_vntuctic subjects in rorl! speech 79

foundation for later grammatical development, rather than an incorrect grammar

which must be revised.
Recall that the pro hypothesis had no theory-internal basis for starting the
child off with the null subject value of the null subject parameter (Hyams, 1987):
it was in response to presumed facts about children’s early productions - pre-
sumptions which our data show tvere incorrect. While the subset principle has
sometimes been invoked as a reason for starting the child off with the non-null
value, that reasoning receives no empirical support and also appears logically
flawed (Valian, 1989, 1990a).
Our suggestion - a skeletal grammar - is similar to the VP hypothesis in
proposing that the child’s initial knowledge is incomplete rather than in direct
cornmissive error. It differs from the VP hypothesis in invoking no mechanism
other than learning as the effector of changes. The child learns, via the evidence
to which she is exposed, how tense and agreement are coded in her language,
how elements are ordered within a phrase, what empty categories exist, and so
on. As the child learns, her performance mechanism is concurrently developing.
affording longer utterances and more complete expression of what the child
knows. Performance factors must play a role in production, regardless of the
child’s target language. The question is how those factors interact with children’s
knowledge to produce the observed patterns of use. We suggest that the inconsis-
tent use of subjects for most of the children below MLU 2.0 is due to performance
limitations, and have presented evidence to that effect. The work for the future
is to develop performance models that will allow us to isolate both what children
know, and what they can express.

References

Antinucci. F., & Parisi, D. (1973). Early language acquisition: A model and some data. In C.A.
Ferguson & D.I. Slobin (Eds.), Studies of child language development. New York: Holt.
Rinehart, and Winston.
Bates. E. (1976). Langlrage and conrexr. F;ew York: Academic Press.
Bell+. U. (1967). The acquisition ofnegorion. Unpublished Ph.D. dissertation. Harvard University.
Blake. J., Quartaro, G., Austin, W., & Vingilis, E. (1989). Does memory span consrrain language
complexity? Paper presented at the Biennial Meeting of the Society for Research in Child
Development, Kansas City. MO.
Bloom, L. (1970). Language developmenr: Form and function in emerging grammars. Cambridge,
MA: MIT Press.
Bloom, L. (1991). Language developmentfrom two to three. Cambridge: Cambridge University Press.
Bloom. L., Lightbown. P.. & Hood, L. (1975). Structure and variation in child language. Monographs
of the Society for Research in Child Development. 40.
Bloom. L., Miller, P., & Hood, L. (1975). Variation and reduction as aspects of competence in
language development. In A. Pick (Ed.). Minnesota Symporia on Child Psychology (Vol. 9,
pp. 3-55). Minneapolis: University of Minnesota Press.
80 V. Valian

Bloom. L.. Tackeff. J.. Br Lahey. 51. (19%). Learnin g 10 in complement constructions. Jom~al of‘
Child Latrglrage. I I, 391106.
Bloom. P. (1990). Subjectless sentences in child language. Linguistic Irquiry. 21. 49l-504.
Borer, H. (1989). Anaphoric AGR. In 0. Jae,,4i Sr K.J. Safir (Eds.). The rrdl mbjecf pcrrunwrer.
Dordrecht: Kluwer.
Borer. H.. & Wesler. K. (1987). The maturation of syntax. In T. Roeper 8: E. Williams (Eds.).
Paratnerer serring (pp. 122-172). Dordrecht: Reidel.
Bowerman. M. (1973). Earl! svnracric dnelopmenr: .4 cross-litrgrristic sntriv blith sprciul reference to
Finnish. Cambridge: Cambridge University Press.
Bowerman. M. (1978). Semantic and syntactic development. In R.L. Schiefelbusch (Ed.). Bases of
lurrgnage inrerv~enrion (pp. 97-159). Baltimore: University Park Press.
Braine. M.D.S. (197-l). Length constraints. reduction rules. and holophrastic processes in children’s
word combinations. Jonrrral of Verbal Learning and Verbal Behuvior, 13. -U8-456.
Brown. R. (1973). A first lungmrge. Cambridge. MA: Harvard University Press.
Brown. R.. Cazden. C.. Br Bellugi. U. (1973). The child’s grammar from I to III. In C.A. Fsrguson
& D.I. Slobin (Eds.). Snrdies of child Jangnage derelopmenr (pp. 295-333). New York: Holt.
Rinehart and Winston.
Brown, R.. & Fraser, C. (1963). The acquisition of syntax. In C.N. Cofer & B.S. Musgrave (Eds.).
Verbal behavior and learning (pp. 155-197). New York: McGraw-Hill.
Budwig. N. (1985). “I. me. my and ‘name”‘: children’s early systematizations of forms, meanings and
functions in talk about the self. Papers and Reporrs on Child Larzgnage Developmenr. 2-1. 30-37.
Budwig, N. (1989). The linguistic markin g of agentivity and control in child language. Jam-md of
Child Langnage. 16. 263-284.
Chi, M.T. (1978). Knowledge structures and memory development. In R.S. Siegler (Ed.), C/rildrert>
rhinking: What develops? Hillsdale. SJ: Erlbaum.
Chomksy. N. (1981). Lecrares on government curd binding. Dordrecht: Foris.
Chomsky. N. (1982). Sofne concepts and corrseque,rces of rhe rheory of governmern and binding.
Cambridge. MA: MIT Press.
de Villiers. J.G., & de Villiers. P.A. (1973). A cross-sectional study of the acquisition of grammatical
morphemes in child speech. Jonrnal of Psvcholitrgdisric Research. 2, 267-278.
Ervin. SM. (1964). Imitation and structural change in children’s language. In E.H. Lenneberg (Ed.).
New directions in the snrdv of larrglfage. Cambridge. .MA: MIT Press.
Gerken. L. (1991). The metrical basis for children’s subjectless sentences. Journal of Memory and
Lungmzge, JO.
Greenfield. P.M., & Smith, J.H. (1976). The strncmre of commanicarion in early language develop-
mem. New York: Academic Press.
Guilfoyle, E. (1984). The acqrrisirion of rease and the emergence of lexical sctbjecrs in child grammars
of English. Boston University Conference on Language Development, Boston.
Guilfoyle, E., & Noonan, M. (1989). Fm~criorzal categories and langnuge acqaisirion. Unpublished
manuscript. McGill University.
Huang, C.T.J. (1989). pro-drop in Chinese: A generalized control theory. In 0. Jaeggli & K.J. Safir
(Eds.), The null sabject paramerer. Dordrecht: Kluwer.
Hyams, N.M. (1986). Langauge ucqmsition and the theory of paramerers. Dordrecht: Reidel.
Hyams, N.M. (1987). The serring of rhe mrll snbjecr paramerer; A reurruJysis. Boston University Con-
ference on Language Development. Boston.
Jaeggli. 0.. & Safir. K.J. (1989). The null subject parameter and parametric theory. In 0. JaeggIi
& K.J. Safir (Eds.), The rzrcllsubj&r paramefer. Dordrecht: Kluwer.
Kazman, R. (1988). Null urgnmems and rhe acqaisition of Case and fr?fl. Boston University Confer-
ence on Language Development. Boston.
Kuczaj, S.A. (1981). More on children’s initial failure to relate specific acquisitions. Joarnal of Child
Language, 8, 485487.
Lebeaux, D. (1987). Comment on Hyams. In T. Roeper & E. Williams (Eds.). Parameter serring (pp.
23-39). Dordrecht: Reidel.
Lebeaux. D. (19s’)). A formal syuaas of telegraphic speech. Boston University Conference on Lan-
guage Development. Boston.
Limber. J. (1976). Unravelling competence. performance and pragmatics in the speech of young
children. /orrntal of Child Language. 3. 309-315.
Lust. B.. d Chien. Y.-C. (1981). The structure of coordination in first language acquisition of Man-
darin Chinese: Evidence for a universal. Cogtritiot7, 17. 49%S3.
Marcus. G.F.. Ullman. M.. Pinker. S.. Hollander. M.. Rosen. T.J.. Sr Su. F. (1990). O\,erreyulari-
zarion. Occasional Paper no. II. Center for Cognitive Science. MIT, Cambridge. MA.
Mazuka, R., Lust. B., Wakayama. T.. & Snyder, W. (1986). Distinguishing effects of parameters in
early syntax acquisition: A cross-linguistic study of Japanese and English. Arperr auti Reporrs
ot7 Child Language Der~elopmem. 25.
Meisel. J.M. (1990). INFL-ecrion: Subjects atld subject-verb agreemerrr br early child language. Eli-
dence from simulraneoirs acqaisiriotl of r)vo first larlguages: Germtrf7 artd Fretrch. Unpublished
manuscript. University of Hamburg.
O’Grady. W.. Peters, A.M., 8: Masterson. D. (1989). The transition from optional to required
subjects. JoLrrt7al of Child Latlguage. 16. 5 13-529.
Olson, G.M. (1973). Developmental changes in memory and the acquisition of language. In T.E.
Moore (Ed.), Cognitive development and rhe acqlrisifior7 of language (pp. l-l%157). Sew York:
Academic Press.
Peters. A.M. (1983). The w7irs of language acquisifion. Cambridge: Cambridge University Press.
Pinker. S. (198-l). Language learnability arrd lurlglrage developmem. Cambridge. MA: Harvard Univer-
sity Press.
Pollock. J.-Y. (19S9). Verb movement. universal grammar. and the structure of IP. Litrglrisric Inquir!.
20. 365--12-l.
Radford, A. (19SS). Transformafior~al grummart A firs! course. Cambridge: Cambridge Universit!
Press.
Radford, A. (1990). S~rrtacdc meor? atrd rhe acqltisiriot7 of English s~r7ra.r: The uarure of early child
grammars of Euglish. Oxford: Blackwell.
Riemsdijk. H. van. & Williams, E. (19S6). Theory of grammar. Cambridge. MA: XLIT Press.
Rizzi, L. (1986). Null objects in Italian and the theory of pro. Lirtgltisric Inquiry, 1;. 501657.
Roeper, T., &r W’eissenborn, J. (1990). How to make parameters work: Comments on Valian. In L.
Frazier & J. de Villiers (Eds.). Larlgtrage processing and language acqrrisiriorz (pp. 117-162).
Dordrecht: Kluwer.
Stromswold. K. (1989). Using narrrralisric data: Theorerical atrd methodological issues. Boston Univer-
sity Conference on Language Development, Boston.
Ullman. M.. Pinker. S., Hollander. M.. Prince. A., & Rosen. T.J. (19Y9). Grobc,rh of regular and
irregular vocabulary and 017the onset of overregrrlarizariorr. Boston University Conference on
Language Development, Boston.
Valian. V. (1956). Syntactic categories in the speech of young children. Developmemal Psycholog!.
22. 562-579.
Valian, V. (1989). Positive evidence, indirect negative evidence. paramerer-setting. and language leanr-
ing. Unpublished manuscript. Hunter College, New York.
Valian, V. (1990a). Null subjects: A problem for parameter-setting models of language acquisition.
Cognirion, 35, 105-122.
Valian. V. (1990b). Logical and psychological constraints on the acquisition of syntax In L. Frazier
& J. de Villiers (Eds.). Language processing and larrgrtage acqltisifion (pp. 119-145). Dor-
drecht: Kluwer.
Valian. V. (in press). Parental replies: Linguistic srutus and didactic role. MIT Press/Bradford, Cam-
bridge. MA.