Journalof Speech and Hearing Research, Volume 38, 850-863, August 1995
Specific Language Impairment
as a Period of Extended
Optional Infinitive
Mabel L. Rice
University of Kansas
Lawrence
-L---·IPIICIIII
Kenneth Wexler
Massachusetts Institute of Technology
Cambridge
Patricia L. Cleave
University of Kansas
Lawrence
English-speaking children with specific language impairment (SLI) are known to have
particular difficulty with the acquisition of grammatical morphemes that carry tense and
agreement features, such as the past tense -ed and third-person singular present -s. In this
study, an Extended Optional Infinitive (EOI) account of SLI is evaluated. In this account, -ed, -s,
BE, and DO are regarded as finiteness markers. This model predicts that finiteness markers are
omitted for an extended period of time for nonimpaired children, and that this period will be
extended for a longer time in children with SLI. At the same time, it predicts that if finiteness
markers are present, they will be used correctly. These predictions are tested in this study.
Subjects were 18 5-year-old children with SLI with expressive and receptive language deficits
and two comparison groups of children developing language normally: 22 CA-equivalent (5N)
and 20 younger, MLU-equivalent children (3N). It was found that the children with SLI used
nonfinite forms of lexical verbs, or omitted BE and DO, more frequently than children in the 5N
and 3N groups. At the same time, like the normally developing children, when the children with
SLI marked finiteness, they did so appropriately. Most strikingly, the SLI group was highly
accurate in marking agreement on BE and DO forms. The findings are discussed in terms of the
predictions of the EOI model, in comparison to other models of the grammatical limitations of
children with SLI.
KEY WORDS: specific language impairment, grammatical development, grammatical
impairment, morphosyntax
The condition of specific language impairment (SLI) is an unresolved puzzle.
Children with SLI seem to have the necessary prerequisite competencies needed for
language acquisition, yet their language milestones are delayed in emergence and a
protracted period of time is needed for acquisition of some fundamental linguistic
features. In fact, there is now reason to wonder if certain features are ever fully
mastered. Plausible explanations for this phenomenon are matters of current inquiry
and dispute (cf. Watkins & Rice, 1994).
For some time it has been noted that grammatical morphology is a particular locus
of difficulty for children with SLI (cf. Bishop, 1992; Leonard, 1989). Certain closedclass free and bound morphemes are typically delayed in appearance for Englishspeaking children with SLI, and when they do appear they are used less accurately
than would be expected for their general level of linguistic productivity. This is
evident in comparisons of preschool children with SLI and younger non-SLI children
at equivalent mean lengths of utterance (MLU), where the children with SLI are less
accurate than their MLU-matched peers. The children with SLI are considerably
delayed relative to their age peers with regard to this set of morphemes.
Within the set of morphemes known to be problematic are ones that carry
tense-marking. This is typically reported with regard to the English affix that carries
past tense (e.g., the -ed inwalked). A number of investigators have documented that
© 1995, American Speech-Language-Hearing Association
850
00Z2 -4685/95/3804-0850
Rice et al.: Extended Optional Infinitive 851
-ed is often omitted by children with SLI (e.g., Leonard,
Bortolini, Caselli, McGregor, & Sabbadini, 1992, for preschool children; Marchman & Weismer, 1994, for school-age
children). Unmarked tense has also been reported for adults
with positive histories of SLI (Tomblin, 1994; Ullman &
Gopnik, 1994). In addition, it is also reported that other
verbal morphemes are also likely to be omitted (i.e., -s in he
walks, and the copula and auxiliary forms of BE). Available
treatments of the evidence tend to regard -s as an agreement marker (cf. Rice, 1994; Rice & Oetting, 1993) and the
BE forms as copula and auxiliary forms of the verb.
A question of interest is what these surface forms have in
common that may be difficult for a child with SLI. In the
descriptive methods used to date, this collection of morphemes can be interpreted in terms of their surface properties-that is, as small, unstressed, function forms. Alternatively, it can be noted that these morphemes all appear in the
verb phrase. What neither of these levels of description
capture is how these morphemes share a common linguistic
feature-that of finiteness. Recent studies show that, for
normally developing children, finiteness of the verb form is
related to linguistic representations that control rules for
verb movement. Thus, the way in which children mark
finiteness in their grammar is of central importance for their
representations of sentence structure. If the condition of SLI
involves incomplete representation of finiteness, the grammatical implications would go beyond the level of affixation
of bound morphemes.
In this study, we set out to evaluate the ways in which
children with SLI do or do not mark finiteness (tense and
agreement) in the verb phrase, and to compare their morphology to that of two groups of peers: one selected for age
equivalence, and the other selected for utterance-length
equivalence. The major purpose is to uncover some important new properties of the condition of SLI. On the basis of
our empirical results, we propose a new characterization of
the grammars of children with SLI, and the ways in which
they are "impaired" with respect to non-SLI children. Although these results indicate that the grammars of children
with SLI differ from adult grammars in highly specific ways,
they also suggest that children with SLI have a good deal of
closely related grammatical knowledge. Moreover, the patterns of knowledge and nonknowledge of children with SLI
look very similar to recently discovered patterns of knowledge and nonknowledge of younger normally developing
children. In effect, children with SLI seem to have "young"
grammar for a very extended period of time. This "young"
grammar, in contrast to existing explanations of SLI, is
better characterized by properties of finiteness-marking in
the underlying grammar than by surface properties of grammatical forms.
Finiteness in the Verb Phrase
Of interest to this study are some generalizations about
ways that finiteness operates in English grammar (cf. Chomsky, 1957; Halle & Marantz, 1993; Quirk, Greenbaum, Leech,
& Svartvik, 1985). Verbs can appear in finite and nonfinite
forms. Finite forms are those marked for tense and agreement. Finite verbs appear in main clauses, where tense/
agreement marking is obligatory. In examples 1 and 2 below
the forms of walk are finite, whereas inexamples 3 and 4 the
forms are nonfinite.
1. a. They walk.
b. They walked.
2. a. She walks.
b. She walked.
3. She liked to walk.
4. She made him walk.
Note that, in English, the finite/nonfinite distinction is not
always marked by a surface form. In la the bare stem walk
is finite. In this case, the bare stem carries invisible features
for tense/agreement that are checked in the syntactic level
of the grammar. In contrast, in 3 and 4 the bare stem, walk,
form is nonfinite.
Finiteness can be marked on lexical verbs or on BE and
DO forms. For example:
5. She does/did not walk.
6. She is/was walking.
7. She is happy.
If BE or DO is present as the first auxiliary verb (see 5 and
6) or as the main (copula) verb (see 7), it is finite. Omissions
of BE and DO in these contexts, then, can be considered as
omissions of tense-agreement-marking.
Finite verbs follow certain principles that guide movement
of the verb within the clause. In English, tense-marking can
become separated from the main lexical verb in certain
contexts. In those contexts, auxiliary BE or DO forms appear
to carry finiteness. This is evident in sentences with a
negative operator, where BE and DO forms appear to the left
of the negative operator, or in verb-initial questions, where
BE and DO forms appear at the beginning of the sentence.
In contrast, lexical main verbs do not move to the left around
negative forms or in questions.
8. a. *They walk not.
b. They do not walk.
9. a. She is not happy.
b. *Shenot is happy.
10. a. She is not running.
b. *She not is running.
11. a. Do they want a book?
b. Is this my book?
c. *Wantthey a book?
In summary, finiteness can be evident on lexical main
verbs, as well as on main verb and auxiliary BE and DO
forms.
Optional InfinitiveAccount
Wexler (1994) has shown that there is a stage in the
development of young nonimpaired children in which they
do not obligatorily mark tense in main clauses, but in which
they nevertheless know the grammatical properties of finiteness. This is known as the Optional Infinitive stage (henceforth, 01). In non-English languages young children sometimes use infinitival forms of verbs where they should use
finite forms. For example, samples from French-speaking
children yield utterances such as
12. Voir 'auto papa
See [- finite] the car of daddy
852
Journal of Speech and Hearing Research
As noted above, in English the infinitival forms are not as
easily detected, because the bare lexical verb is also the
infinitival form. Wexler argues that child utterances such as
13. He go.
14. He not go.
can be regarded as infinitival forms, lacking tense and
agreement, in the same way that infinitival forms appear in
children's utterances in other languages.
A very important point is that when children do mark
finiteness, they show knowledge of the related linguistic
processes that apply to finite verbs. For example, in French
(see Pierce, 1992; Weissenborn, 1994) even very young
children know that finite verbs precede the negative marker
pas, as in utterances such as
15. II est pas mort
He is [+ finite] not dead
and that nonfinite verbs follow pas, as in utterances such as
16. pas manger la poupee
not eat [- finite] the doll
The interpretation is that French-speaking children know
that verbs raise, or move to precede the negative marker, if
the verbs are marked as finite-that is, if they are marked for
tense and agreement features. What children do not seem to
know is that such features must be marked on the first verb
of a main clause. That is, they sometimes chose an infinitive,
in which case the finite features are not registered and the
related processes do not appear. At the same time, children
do seem to know that finite verbs must show agreement and
tense and that they must appear in certain positions.
On the basis of cross-linguistic data, Wexler (1994) proposes that when children in the Optional Infinitive stage use
finite tense, they also show correct subject-verb agreement.
We will therefore take correct agreement for finite verbs to
be a property of the 01 stage.
Predictions of the Optional Infinitive
Stage for English
Wexler (1994) argues that the occurrence in children of
bare stems in third-person singular matrix verbs is an
example of the Optional Infinitive stage in English. That is,
when children produce forms like "she talk," what appears
to be a bare stem, "talk," is a nonfinite form for the child. In
other Romance and Germanic languages, the nonfinite form
has a distinctive added inflection (e.g., er in French). In
English, however, the nonfinite form appears as a bare
stem.'
For English-speaking children, children in an 01 stage of
development are predicted to demonstrate certain grammatical properties (see Wexler, 1994 for the theoretical
arguments to support these predictions). Of most importance for this investigation are predicted uses of tense
markers on lexical verbs and on BE and DO. Here we will not
1
Note that in English there is a certain ambiguity about bare stem forms of the
main lexical verb, because bare stems may be either finite or nonfinite,
depending on the sentential context. (Recall that in example la, "they walk,"
walk isfinite; in example 8b "they do not walk," walk is nonfinite). Finite forms
of the verb carry features for tense and agreement that are checked in the
syntactic level of the grammar, where principles of verb movement apply.
38
850-863
August 995
derive the predictions in detail, but will give just a short,
informal description of how they are made. The essential
assumption is that in the Optional Infinitive stage, tense is
optional for the child. That is, the child will have some
sentences with finite tense and some sentences without
finite tense. This predicts that lexical verbs in main clauses
can appear in nonfinite form, meaning that verbs can appear
as bare stems in English-speaking children's sentences and
in the infinitive form in French-speaking children's sentences, contrary to the adult versions of the respective
languages. A second assumption is that BE and DO are
inserted in a sentence only to carry tense-marking, so that,
if the sentence does not have tense, BE and DO will not
appear. This also predicts that BE and DO will not appear in
the infinitival form, because if the sentence does not have
tense, BE and DO won't be inserted. 2 Furthermore, we
assume that the child in the 01 stage has knowledge of tense
and agreement, and that these features will not be used in
contexts in which the adult grammar does not allow them.
For example, a child will not show a third-person singular
agreement marker on a verb when the subject is first-person
singular (e.g., *I1goes). In summary, we can make the
following predictions:
1. For -s and -ed markings on lexical verbs, bare stems
may optionally be used where inflected forms are required.
2. For -s, in contexts other than third-person singular,
there will be no overt marking; that is, *they walks is not
predicted to be a productive error.
3. -ed will be restricted to past-tense contexts.
4. Auxiliary and main verb (copula) BE may be omitted. 3
5. Auxiliary DO may be omitted.
6. When BE and DO forms are used in contexts where the
adult grammar requires a finite form, children will give
correct agreeing forms.
Extended Optional Infinitive Stage
The 01 account of language acquisition in nonimpaired
children can be applied to children with SLI. Our hypothesis
is that the predictions of the Optional Infinitive stage will hold
for children with SLI, with the further stipulation that this
stage will be extended or prolonged for a greater period of
time for these children. We do not know, in fact, if individuals
2
Note that there is no prohibition against BE and DO being used as nonfinite
forms in grammatical contexts in which nonfinite forms are required-for
example, imperatives ("Be happy"), VPs containing modals ("You can be my
friend"), infinitival complements ("You want to be my friend?"). What should
not appear are nonfinite forms of BE and DO in grammatical contexts that
require finiteness; for example, we do not expect forms like "She be happy,"
"Do she have a book?" In fact, as Wexler (1994) points out, it is a general
property of the 01 stage cross-linguistically that auxiliaries do not appear as
nonfinite forms in contexts where adult grammar requires that they be finite.
The prediction is that the 01 and EOI stages of English follow this general
pattern.
3
1nthis section and throughout the paper, when we say "a form F may be
omitted," we mean that the child's grammar accepts the sentence with and
without F, although the adult's grammar requires F. Sometimes we speak of
the omission of F producing an "incorrect" form. We mean by this that
omitting F is incorrect for the adult grammar. Of course, given the 01 model,
the form is correct for the child's grammar.
Rice et al.: Extended Optional Infinitive 853
with SLI will ever fully leave this stage. Thus, we predict an
Extended Optional Infinitive stage (EOI).
It is important to be clear about what the term optional
means with respect to the "Optional Infinitive stage." The
term is used in the sense of "optional" in grammatical
theory. To say a form F is "optional" means that F may or
may not appear in a sentence; both the forms with and
without F are grammatical. There is no implication that
sentences with and without F should appear equally in
production; it may very well turn out that the form with or
without F is preferred and thus might appear more often.
Wexler (1994) pointed out that it appears as if normally
developing children produce the finite form more often in
obligatory contexts as they grow older; moreover, it might
even be the case that extremely young children hardly ever
produce the form inobligatory contexts. Thus the proportion
of use of nonfinite forms in contexts where finite forms are
required inthe adult grammar can serve as a measure of the
extent to which the child is emerging from the Optional
Infinitive stage. We should point out that it is not clear to
what extent the emergence from the 01 stage is gradual; the
data are not yet available for non-SLI children. Nor is it clear
why emergence should be gradual, if indeed it is. The 01
model conceives of competence in the 01 stage as accepting both finite and nonfinite matrix sentences as grammatical; why the percentage of finite utterances in production
increases with age is a question for future research. Perhaps
the grammatical model of the 01 stage is correct as a model
of competence, but older children "prefer" to produce finite
forms more often, although both finite and nonfinite forms
are grammatical for them.
In a similar spirit, if EOI is correct, children with SLI will
accept as grammatical both finite and nonfinite matrix
clauses at a much later age than do normal children.
Moreover, given that the proportion of finite forms in obligatory contexts seems to be a measure of emergence from
the 01 stage in normally developing children, we would
expect that children with SLI will show a higher proportion of
nonfinite forms than will normal children of the same age.
The standard question asked in SLI research about children's knowledge of a particular form F is: Iscompetence on
this form F even less than the competence on the form
displayed by a normally developing child with "overall"
linguistic ability equivalent to the child with SLI? "Overall"
competence is conventionally indexed by MLU (though
certain caveats apply to this design; cf. Plante, Swisher,
Kiernan, & Restrepo, 1993). Thus, following the logic of a
language-matched control group, we investigate whether
children with SLI produce finite forms in obligatory contexts
less often than do MLU-equivalent nonimpaired children.
Following the above discussion, the EOI model predicts that
the answer to this question is yes.
The value of the EOI for adding to our understanding of
the condition of SLI lies in the predicted cluster of affected
grammatical morphemes, the predictions of certain kinds of
adult-like grammatical knowledge co-existing with nonadult-like properties of the grammar, and the possible parallels to the grammar of non-SLI children. In this paper we
report the empirical outcomes of an initial investigation of
the predictions of this account, suggest a new characteriza-
tion of the condition of SLI, and consider implications for
alternative models of SLI.
Method
Participants
Sixty children participated in this study. Eighteen of the
children were diagnosed as SLI; the other 42 children were
non-SLI children whose language acquisition met normative
expectations. All of the children were drawn from native
English-speaking homes and were enrolled in a preschool or
day-care program.
The children in the SLI group ranged in age from 55 to 68
months (M = 60.17; SD = 4.42). They were diagnosed as
language-impaired by a certified speech-language pathologist and enrolled in a preschool program for children with
language impairments. Their intellectual functioning was
above clinical levels of intellectual impairment (all subjects
scored an age deviation score of 85 or above on the
Columbia Mental Maturity Scale [CMMS]; Burgemeister,
Blum, & Lorge, 1972), and their hearing was within normal
limits (according to a hearing screening or audiological
examination conducted within 6 months of the study). Their
receptive and expressive language performance fell below
age expectations. Each child scored one or more standard
deviations below the mean on the receptive measure, the
Peabody Picture Vocabulary Test-Revised (PPVT-R; Dunn &
Dunn, 1981), and on the expressive measure, the mean
length of utterance (MLU), calculated from a sample of at
least 100 spontaneous utterances and the normative information provided by Leadholm and Miller (1993). Two additional children for whom there were data were ultimately
dropped from the analyses. One child was a twin sister of
another subject. Because of the independent observation
assumptions of the quantitative analyses, and the likelihood
that a twin's grammar will be more similar to that of a sibling
than to that of an unrelated child, the twin was dropped. The
other child's mother was found to have a clinically significant
hearing loss, and her speech and grammar were affected by
the hearing loss. Because of the unconventional input received by this child, she was dropped from the sample.
Individual subject profiles are presented in Table 1.
The 42 non-SLI children formed two comparison groups.
Twenty-two were of the same chronological age, about 5
years old (henceforth, the 5N group), as the SLI sample.
They ranged in age from 55 to 67 months (M = 60.91; SD =
3.82). The remaining 20 children were at an equivalent level
of language, as indexed by their mean length of utterance.
The children in this group were about 2 years younger than
the children in the SLI group (henceforth the 3N group). The
age range for the 3N group was 30 to 40 months (M = 35.50;
SD = 3.24). For the 3N group, the mean level of MLU was
3.49-the same as that for the SLI group. The hearing levels
of children inthese groups was normal, according to teacher
report, and their language performance was within normal to
high normal range on the PPVT-R. For children in the 3N
group, their MLU values were within or above 1 standard
deviation of the mean expected for age. In order to ensure
854
Journal of Speech and Hearing Research
38
TABLE 1. Subject profiles: SLI children.
ID
Sex
Age
Columbia'
August
995
TABLE 2. Subject profiles: 5N children.
PPVTb
MLU
ID
Sex
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
M
F
F
F
F
F
M
F
F
F
M
F
M
M
F
M
F
F
F
M
M
F
1
M
65
103
75
3.70
2
M
59
111
79
3.12
3
M
59
115
84
3.60
5
M
63
85
72
3.93
7
M
59
99
61
3.72
8
M
55
107
68
3.04
9
F
57
93
81
3.19
11
F
66
99
85
4.03
12
F
57
99
76
3.34
13
M
59
86
69
3.26
14
F
68
95
83
2.78
15
F
58
115
75
3.14
16
M
68
96
65
3.92
17
M
55
108
73
4.44
18
F
65
102
82
3.38
19
M
57
100
79
3.48
20
F
56
89
83
3.32
21
F
57
90
73
3.19
M
60.17
99.56
75.72
3.48
SD
4.42
9.17
6.93
55-68
85-115
61-85
Range
850-863
0.42
2.78-4.44
aColumbia Mental Maturity Scale; age deviation score reported.
bPeabody Picture Vocabulary Test-Revised; standard score re-
ported.
equivalent levels across the two groups, each subject in the
MLU group was within .10 morphemes of at least one child
in the SLI group. Subject profiles for the 5N and 3N subjects
are listed in Tables 2 and 3.
The obtained mean values for the CMMS index of intellectual functioning were within normative expectations:
99.56 for the SLI group and 105.55 for the 5N group. The
difference between the two means is statistically significant:
t (38) = 2.21, p = .03. Zero order correlations between the
CMMS and the target variables (i.e., proportion correct for
-s, -ed, and BE-pooled across probes and spontaneous
samples) were calculated for each group. They yielded no
significant associations for any of the variables. Thus, it is
unlikely that performance on the CMMS is associated with
performance on the grammatical variables of interest in this
study, in the normative range sampled.
Because of the known tendency for children with SLI to have
speech errors, and the potential ambiguities such errors pose
for morpheme analysis, children with multiple and severe
articulation errors were excluded from the study. The Goldman
Fristoe Test of Articulation (Goldman & Fristoe, 1986) was
administered to describe their single-word pronunciation. Inspection of the children's errors revealed that none of the SLI
subjects' articulation errors involved the final /t/ or /d/ phonemes, and all subjects had intelligible approximations of final
/s/ and /z/. In addition, each child was administered a probe for
final /s/ and /z/ as part of the protocol for a study of their plural
M
SD
Range
Age
Columbiaa
PPVTb
MLU
7.44
4.01
4.29
4.49
5.26
4.69
3.98
3.96
4.35
4.51
4.82
4.36
4.86
4.73
4.85
5.85
4.64
4.25
5.79
3.37
6.03
6.23
62
57
62
63
58
57
58
55
65
59
67
58
65
65
57
57
66
62
66
57
59
65
113
105
101
101
111
106
100
120
101
105
103
123
94
93
100
111
111
94
106
102
106
116
114
107
98
118
129
107
109
112
104
118
104
131
118
91
97
101
120
99
128
94
103
117
60.91
3.82
55-67
105.55
7.98
93-123
109.95
11.46
91-131
4.85
.93
3.37-7.44
aColumbia Mental Maturity Scale; age deviation score reported.
bPeabody Picture Vocabulary Test-Revised; standard score reported.
productions (cf. Getting & Rice, 1993). All subjects demonstrated final /s/ and /z/, or a consistent approximation of /s/
and /z/, on this probe. Thus, agreement and tense omission
are not confounded by faulty articulation. The way in which the
children's speech is not like that of the normal control group is
largely attributable to their mispronunciation of /s/, /sh/, /ch/,
/r/, and /I/. Overall, the intelligibility of the SLI group was
equivalent to that of the MLU group. In their spontaneous
samples, the percentage of intelligible utterances for the SLI
group was 92%, compared to 91% for the MLU group, as
calculated by the Systematic Analysis of Language Transcript
(SALT) software (Miller & Chapman, 1991). Therefore, there is
reason to believe that the children's morphological omissions
are not readily attributable to faulty phonology or speech
production, and the lower MLU estimates are not an artifact of
articulation errors. Further evidence in support of this conclusion is the finding that the children's use of plural morphology
is highly accurate (cf. Getting & Rice, 1993).
Procedures
Experimental measures were derived from language samples and linguistic probes designed to elicit instances of
Rice et al.: Extended Optional Infinitive 855
TABLE 3. Subject profiles: 3N children.
ID
Sex
Age
PPVT
MLU
93
3.76
---
31
32
33
34
35
37
38
39
40
41
42
M
F
M
F
F
F
F
M
F
F
M
43
M
44
M
45
F
46
M
47
48
F
F
49
M
50
51
F
M
M
SD
Range
36
30
33
97
3.12
117
3.57
34
116
3.37
38
39
33
39
39
40
35
30
36
33
32
108
3.90
3.89
40
95
99
100
3.14
3.28
125
3.55
3.81
115
3.31
108
3.42
102
2.89
112
3.13
109
3.95
96
4.35
36
38
32
37
113
35.50
3.49
3.24
105.60
9.67
30-40
93-125
2.89-4.35
103
93
3.24
3.39
117
3.43
94
3.34
.36
aPeabody Picture Vocabulary Test-Revised; standard score reported.
past tense -ed and third-person singular -s forms on lexical
verbs, as well as productions of copula and auxiliary BE in
questions and statements and auxiliary DO in questions. The
testing and experimental tasks were administered over a
3-day period to each subject individually in a quiet room
near the preschool or day-care class. Each session was 15
to 30 min in length.
The language samples were collected using a standard
set of toys that included a garage/store set, zoo animals,
and a baby doll. The samples were audio-recorded, using a
dual-microphone setup. The samples were subsequently
transcribed and coded for grammatical morphemes following the conventions of the Kansas Language Transcript
Database (KLTD) as outlined in Howe (1992). Grammatical
analyses were conducted by means of the SALT transcript
analysis procedures (Miller & Chapman, 1991). Standard
analyses were supplemented by analyses conducted from
listings of each utterance in which a targeted form was used
or an obligatory context for use occurred. Visual inspection
of these listings were carried out to document all errors of
form use.
The third-person singular probe (e.g., -s) consisted of 12
pictures that depicted a person in a particular occupation.
The examiner showed the child a picture and said "This is a
fire fighter. If I'm a teacher and I teach, he's a fire fighter so
he..." and the child completed the utterance. The child's
responses were coded on-line.
The past-tense probe (e.g., -ed) consisted of pictures
depicting people in the process of doing something or
having completed an activity. There were a total of 11
regular past-tense verbs included. The child was shown a
picture of an ongoing action and told, for example, "The boy
is climbing the ladder. Now he's all done." The child was
then shown a picture depicting the completed action and
asked "What happened?" The child's responses were
coded on-line.
The BE and DO forms were elicited in a task involving
stuffed animals (one cat and three bears) and a puppet that
the examiner manipulated. The child was told that the
puppet could talk to the animals and so he (the child) had to
ask the puppet questions to find out about the animals. The
examiner then elicited questions for the various forms. For
example, to elicit copula BE contexts, the examiner said "I
wonder if the cat is cold? The puppet knows. Ask him." For
the auxiliary contexts, the eliciting statements took the form
"I wonder if the bears are crying?" For the DO context, the
eliciting questions followed the form, "I wonder if the puppet
wants a blanket?" Questions were elicited in present tense
for second- and third-person contexts for copula and auxiliary BE and auxiliary DO. A total of three examples of each
context was elicited. In addition, statements were elicited for
third-person BE by prompting the child to describe the
animals as she or he played with them. For example, the
experimenter would say "What's happening?" or "Tell me
about the kitty." Three examples each of third-person singular and plural forms of copula and auxiliary BE forms were
elicited. This probe was audio-recorded, and the child's
responses were later transcribed.
Reliability. Language sample transcription and coding
involved a number of different steps and checking passes.
Language sample elicitation, transcription, and coding of the
child utterances were completed by two graduate students;
two additional graduate students assisted in the transcription of the adult utterances and the various checking procedures. The reliability of initial transcription was calculated
by having the two graduate students who completed the
transcription check each other's work. Ten percent of the
data was checked and was found to be 95%-99% consistent with the conventions outlined in Howe (1992). Finally, in
the preparation of the listings for line-item analyses further
checking of coding was carried out for the few miscodings
that may have persisted through previous passes.
Reliability estimates for the probe data were obtained in
the following way. The children's responses on the pasttense and third-person singular probes were recorded live,
and these were then scored by a trained graduate student.
A second graduate student checked the scoring, and disagreements were settled by consensus. The DO/BE probe
was audio-recorded and later transcribed and scored. For
10% of the samples (two children from each group) a
second examiner independently transcribed and scored the
children's responses. Agreement was over 95% for both
questions and statements.
856
Journal of Speech and Hearing Research
38
Results
850-863
Augus 1995
TABLE 5. Grammatical morphemes on matrix verbs in elicitation probes: percent correct.
The findings are reported according to lexical verbs versus BE and DO, for each group of children, for spontaneous
and probe data. The results are compared to the predicted
outcomes.
Main lexical verb marking of tense and agreement.
Main lexical verbs such as walk carry -s for third-person
singular present tense and -ed for regular past tense.
Uninflected lexical verbs, as argued above, can be interpreted as infinitival forms. The results for lexical verbmarking of tense and agreement are reported in Tables 4
and 5. Table 4 reports the percentage correct in obligatory
contexts in spontaneous samples and means and standard
deviations for each group. (Included in these calculations are
participants with a minimum of three obligatory contexts for
-s. N = 15, SLI; 17, 3N; 20, 5N.4) These findings document
that SLI children are less likely to mark -s on lexical verbs for
tense/agreement than are control groups of children, with
means of 34% for the SLI group, 51 % for the 3N group, and
85% for the 5N group. An ANOVA yielded an F (2, 49) =
13.5, p < .001. Pairwise comparisons were as follows: SLI <
3N, t (49) = 1.62, p = .11; SLI < 5N, t (49) = 5.01, p = .001;
3N < 5N, t (49) = 3.45, p = .001.
A similar pattern of group differences holds for the pasttense marker -ed-with means of 18% for SLI, 56% for 3N,
and 90% for 5N-for the children with at least three obligatory contexts (N = 9, SLI; 7, 3N; 14, 5N). Perhaps these
numbers should be regarded with caution because the
remaining N is reduced. An overall ANOVA yielded an F (2,
27) = 25.55, p < .001; pairwise comparisons as follows: SLI
< 3N, t (27) = 3.18, p = .004; SLI < 5N, t (27) = 7.13, p =
.000; 3N < 5N, t (27) = 3.11, p = .004.
In Table 5, data from probe elicitation items are presented
for the lexical verb markings. There were 12 items presented
for -s and 11 for -ed. Children were included in the analyses
if they produced at least five grammatical contexts in which
the target morpheme would be obligatory in the adult
grammar. The dependent variable is the percentage correct
use of the morpheme in contexts where it is required. For
4
Because the number of responses for a given variable could vary from child
to child, and sometimes a child provided no or only a very few utterances with
obligatory contexts for the target morphemes, it was necessary to set, for
each variable, a minimal number of occurrences of obligatory contexts for
inclusion of a child in the analyses. Whenever feasible, the minimal number
was five occurrences. For some variables, however, this criterion would lead
to the loss of many subjects. In those cases, three or four occurrences was
the standard for inclusion, following the strategy of invoking the criterion that
maximized both the number of surviving cases and the number of occurrences in the denominator for calculation of the proportions of correct use in
obligatory contexts. The criterion employed for each variable is reported.
TABLE 4. Grammatical morphemes on matrix verbs in spontaneous samples: percent correct in obligatory contexts.
SLI
3N
5N
-s
34%
(SD = 26%)
51%
(SD = 40%)
85%
(SD = 20%)
-ed
18%
(SD = 20%)
56%
(SD = 32%)
90%
(SD = 20%)
-s
-ed
SLI
3N
5N
26%
(SD = 23%)
38%
(SD = 41%)
95%
(SD = 13%)
27%
50%
92%
(SD = 26%)
(SD = 36%)
(SD = 14%)
example, if a subject produced eight sentences in which -s
is required and actually produced -s twice, the percentage
would be 25%. It is clear that the same pattern of group
differences holds here. For the -s morpheme, the group
means are 26% for SLI, 38% for 3N, and 95% for 5N (N =
18, SLI; 19, 3N; 22, 5N). The overall ANOVA produced an F
(2, 56) = 35.10, p < .001, with pairwise contrasts indicating
SLI < 3N, t (56) = 1.29, p = .20; SLI < 5N, t (56) = 7.70,
p = .001; 3N < 5N, t (56) = 6.46, p = .001. For the -ed
analyses, the means are 27% for SLI, 50% for 3N, and 92%
for 5N (N = 17, SLI; 12, 3N; 22, 5N). The overall ANOVA
yielded an F (2, 48) = 36.44, p < .001. Pairwise contrasts
were as follows: SLI < 3N, t (48) = 2.56, p = .014; SLI < 5N,
t (48) = 8.37, p = .001; 3N < 5N, t (48) = 4.84, p = .001.
Note that the similarity infindings between the spontaneous
samples and the probe-elicitation items bears on the possibility that the low performance of the children in the SLI
group could be attributable to a lack of comprehension of
the probe-elicitation constructions. The lower performance
of the SLI group is evident in spontaneous productions as
well, thereby ruling out a probe effect alone.
Three observations are of interest from these analyses of
lexical verbs. The first is that Prediction 1 is confirmed;
children in the SLI and 3N groups often use bare stems
instead of inflected forms for -s and -ed. The second
observation is that the groups are differentiated on the basis
of overall percentage. Children with SLI are consistently at
much lower levels of performance than their age-matched
peers, a difference that meets conventional levels of statistical reliability. The numerical means of the SLI group are
also consistently lower than those of the younger children of
equivalent language levels, the 3N group. These differences
hover around conventional levels of statistical reliability,
reaching that level unequivocally for the -ed probe data.
With regard to the -s data, we are inclined to view the
pairwise comparisons as lacking in sufficient power to
detect real differences that might exist between the SLI and
3N group. We draw upon three kinds of reasoning to support
the conclusion that a real difference exists. First are the
reports from other investigations of comparable children
with SLI, in which reliable levels of difference between the
two groups are reported for -s (Leonard et al., 1992; Rice &
Getting's study of 81 SLI children, 1993). Second is the
observation that repeated findings of a difference in the
predicted direction can constitute evidence of real differences. As stated by Stevens (1986, p. 137), "several low
power studies that report nonsignificant results of the same
character are evidence for an effect." In our case, a predicted direction of difference is found for two different
measures (spontaneous language and probe items) of the
Rice et al.: Extended Optional Infinitive 857
morpheme -s, with p values of .11 and .20, respectively.
Third, in the -s probe data comparisons, the standard
deviation in the 3N group was .42, as compared to .26 in the
SLI group, which would also have worked against detection
of a real difference. Considering these arguments, we believe the evidence supports the interpretation that children
with SLI at 5 years of age are likely to lag behind children 2
years younger in the likelihood that they will use -s and -ed
in contexts where these morphemes are obligatory in the
adult grammar.
The third observation of interest is with regard to the ways
in which the morpheme use was incorrect. Predictions 2 and
3 state that -s and -ed will not be misapplied to contexts
where the adult grammar does not allow them. For all of the
children, the errors for these lexical verb-markings were
omissions. In the context of he walks they said *he walk.
Recall that the bare-stem form of lexical verbs can be
viewed, in the framework adopted here, as infinitival forms of
the verb. Collapsing the spontaneous and probe data for -s,
the omission rates are 70% for SLI, 55% for 3N, and 10% for
5N. That is, the entire 100% of responses are accounted for
by correct uses of the morpheme and omissions. For -ed, in
the context of he walked the children often said *hewalk. For
-ed the omission rates are 78% for SLI, 47% for 3N, and 9%
for 5N. Again, all the responses in the past context are
accounted for by omissions of -ed and correct uses. The
children did not seem to avail themselves of other possible
ways to be wrong. For example, they did not misapply the
affix in he walks to *youwalks, nor did they say he walks
where he walked would be correct. Thus, we can infer that
they do not make positive agreement errors, in the application of -s in contexts other than third-person singular, nor did
they ever use the past tense -ed suffix in a present-tense
context. They sometimes but rarely misapplied the -ed in he
walked to irregular stems (e.g., he goed), but of course this
conforms to our predictions that a past-tense morpheme is
used in a past-tense context. Overall, the children never
used a finiteness marker (tense or agreement) in the wrong
grammatical context. This result follows from Predictions 2
and 3.
BE and DO forms. In contrast to the lexical verbs that
carry semantic information, BE and DO are syntactic forms
inserted for tense and agreement-marking (Chomsky, 1957;
Halle & Marantz, 1993; Wexler, 1994). Marking of person,
number, and tense is accomplished by means of the forms,
am, is, and are (e.g., I am, you are, he is, they are, etc). The
infinitival phonetic form of BE (i.e., be) does not appear in the
person, number, and tense paradigm. The infinitival phonetic form of DO is used in all present-tense contexts but the
third-person singular context (e.g., he does). In other words,
the finite and nonfinite forms of present-tense DO are not
distinguishable phonetically except in the does context.
Predictions 4 and 5 state that BE and DO are likely to be
omitted. The use of BE forms in the children's spontaneous
utterances was analyzed for percentage of correct use in
obligatory contexts (see Table 6). Collapsing across copula
(e.g., he is cold) and auxiliary contexts (e.g., he is sleeping),
the group means and standard deviations are as follows:
SLI, 46% (24); 3N, 68% (24); 5N, 97% (5) (N = 18, SLI; 20,
3N; 22, 5N). Because of the multiple occurrences of BE
TABLE 6. BE and DO in spontaneous samples: percent correct
in obligatory contexts.
BE (collapsing across main
verb and auxiliary)
Third-person irregular
(does)
SLI
3N
5N
46%
68%
97%
(SD = 24%) (SD = 24%) (SD = 5%)
76%
55%
100%
forms for all children, there was no need to invoke a minimal
use criteria and no subjects were dropped for that reason.
The overall ANOVA yielded an F (2, 57) = 34.4, p < .001,
with pairwise comparisons as follows: SLI < 3N, t (57) =
3.48, p = .001; SLI < 5N, t (57) = 8.22, p = .001; 3N < 5N,
t (57) = 4.79, p = .001. Thus, it appears that the same
general pattern holds as for the lexical verbs-that is, the SLI
children's performance on BE is lower than that of the
control groups. As in the case of the lexical verbs, almost all
the noncorrect cases were accounted for by omissions.
The children's spontaneous use of third-person singular
does is also reported in Table 6. This was the most frequently occurring present-tense form of DO in these samples (where DO often appeared in past tense). This form is of
interest because finiteness is phonetically marked. At the
same time, the frequency of use is low (less than 3 occurrences per child, on average), so the actual numerical values
must be interpreted with caution. The means are of interest,
however, in that they follow the same pattern as for the BE
forms (i.e., SLI, 55%; 3N, 76%; 5N, 100%). No ANOVA was
conducted because of the low frequency of observations.
Almost all of the errors are omissions of does.
Further evidence of this pattern of findings is supplied by
the probe data. The use of BE and DO is reported in Table 7,
broken out by statements and interrogatives, with the BE
forms subdivided according to auxiliary versus copula contexts. In the BE analyses, subjects were excluded if they
used ain't (1 child), if they were generally nonparticipatory (7
from the 3N group were lost in this way), if they did not use
alternating forms (1 child), or if they made fewer than 4
attempts (2 children). ANOVAs were conducted to evaluate
TABLE 7. BE and DO in elicitation probes: percent correct.
SLI
Statements
Interrogatives
Statements
Interrogatives
Interrogatives
3N
Auxiliary BE
40%
72%
(SD = 33%)
(SD = 35%)
53%
64%
(SD = 46%)
(SD = 41%)
Copula BE
49%
73%
(SD = 26%)
(SD = 30%)
38%
51%
(SD = 47%)
(SD = 45%)
DO
37%
47%
(SD = 41%)
(SD = 43%)
5N
99%
(SD = 3%)
97.5%
(SD = 6%)
100%
99%
(SD = 2%)
96%
(SD = 10%)
858
Journal of Speech and Hearing Research
38
the statistical reliability of the observed group differences.
For auxiliary BE statements, the overall ANOVA yielded an F
(2, 51) = 20.96, p < .001, pairwise contrasts of SLI < 3N, t
(51) = 3.37,p = .001; SLI < 5N, t (51) = 6.47, p = .001; 3N
< 5N, t (51) = 3.14, p = .003. For auxiliary BE in questions,
the results were overall F (2, 42) = 8.76, p = .001, pairwise
contrasts of SLI < 3N, t (42) = .85, p = .40; SLI < 5N, t (42)
= 3.92, p = .001; 3N < 5N, t (42) = 2.79, p = .008. For
copula BE in statements, the results were overall F (2, 54) =
24.67, p = .001; pairwise contrasts of SLI < 3N, t (54) =
3
.14, p = .003; SLI < 5N, t (54) = 7.00, p = .001; 3N < 5N,
t (54) = 3.74, p = .001. For copula BE in questions, the
results were overall F (2, 43) = 15.14, p = .001, pairwise
comparisons, LI < 3N, t (43) = .90, p = .375; SLI < 5N, t (43)
= 5.04, p = .001; 3N < 5N, t (43) = 3.93, p = .001. As in the
spontaneous data, almost all the noncorrect cases were
accounted for by omissions. The lower performance of the
SLI group relative to the 3N group was evident for BE in
statements, for copula and auxiliary, but the difference was
not as evident in the question contexts. This may be
because question uses are somewhat more difficult to elicit,
with larger standard deviations within the groups suggesting
some instability in measurement.
For the DO analyses, subjects were excluded if they made
fewer than 4 attempts. In the probe items, there is some
discourse optionality with regard to the use of DO in secondperson questions (e.g., You want a cookie?). Such utterances were not included in the analyses. The mean values
for the DO probes were as follows: SLI, 37%; 3N, 47%; 5N,
96% (N = SLI, 14; 3N, 12; 5N, 21). An overall ANOVA
reported an F(2, 44) = 16.99, p < .001. Pairwise comparisons were SLI < 3N, t (44) = .87, p = .39; SLI < 5N, t (44)
= 5.36, p = .001; 3N < 5N, t (44) = 4.17, p = .001.
To summarize the findings with BE and DO, they are very
often omitted by children in the SLI and 3N groups. This
conclusion is in line with Predictions 4 and 5 above.
What the analyses of lexical verbs and BE/DO show is
that, for children with a MLU of around 3.5, tense and
agreement markers tend to be omitted. This tendency is
especially pronounced for children in the SLI group, whose
level of performance is consistently below that of the 3N
group, a difference that reaches statistically reliable levels
for -ed and BE in this sample and, arguably, seems to be the
case for -s as well. These patterns are illustrated in a general
way in Table 8, in which the data are collapsed over elicited
and spontaneous samples, the two forms of BE, and interrogative versus declarative forms.
850-863 August 1995
Finiteness, Tense, and Agreement
Some of the findings reported thus far could be interpreted as evidence that children with SLI, as well as younger
normally developing children, lack knowledge of the system
for marking tense and agreement in English, or, plausibly, do
not know that every main clause must be marked + finite.
On the contrary, the 01 and EOI hypotheses claim that
children do know the properties of tense and agreement, but
they do not know that main clauses must be marked + finite.
Predictions 2 and 3, which the previous section demonstrated to be in line with the data, state that children have
knowledge of the tense and agreement system in lexical
verbs. That is, the -s and -ed were not used incorrectly in
lexical verbs. Prediction 6 makes the same claim for BE and
DO forms; namely, when these forms are used in finite
contexts children will give correct agreeing forms. In particular, children will neither use infinitival forms of these verbs
in finite contexts nor will they use finite forms that don't
agree with the subject. Note that because agreement is
more manifest on BE than on lexical verbs, the study of
agreement on BE will be particularly informative with respect
to the claim that children in the 01 stage show correct
agreement. If this prediction is supported, it would be further
evidence both that children know the tense and agreement
systems in English and that the generalizations predicted by
the 01 and the EOI models are correct. In order to test these
predictions, additional analyses were done.
The following forms were examined in the probe data: In
interrogative sentences, BE (copula and auxiliary) and DO; in
statements, BE (copula and auxiliary). Statements and questions are separated because questions require overt movement of BE and DO forms to the front of the sentence (e.g.,
Are you happy? Is he running? Do you want it?). If BE and
DO forms are moved to the front of the sentence, according
to the model guiding this work, they must be marked + for
tense and agreement (Poeppel & Wexler, 1993; Wexler,
1994).
Recall that first-person contexts were not elicited in these
procedures so the contexts of interest are second- and
third-person singular and plural (i.e., you are, he is, they are;
you do, he does, they do). All are in present tense. In the
TABLE 9. Finite/nonfinite analyses: probe data, number of
occurrences.
BE
DO
(third-person singular)
606
51
0
4
237
10
0
0
158
5
0
2
5N
TABLE 8. Percent correct in obligatory contexts, collapsed over
elicited and spontaneous samples.
Finite
Nonfinite
3N
-s
SLI
3N
5N
30%
45%
90%
Finite
Nonfinite
SLI
-ed
23%
53%
90%
BE
45%
65%
99%
Finite
DO
46%
62%
98%
Nonfinite
Rice et al.: Extended Optional Infinitive 859
analyses to follow, each of these linguistic contexts was
examined for each of the three groups of children.
As reported in Table 9, in all of the BE contexts, for all of
the subjects, when the children used a form they chose a
finite form-that is, am, is, or are. In none of the observed
contexts did they use the infinitival be. This indicates that if
they used a BE form, they seemed to know that it had to be
the proper form (i.e., carry tense and agreement). For BE,
that may not be too surprising because they do not hear be
in these contexts. The situation is different for DO, because
the infinitival form is used for all contexts except thirdperson singular. It would be reasonable to expect children to
choose this alternative if they were insensitive to the need to
mark agreement. However, the same pattern was observed
for the use of DO. That is, in third-person contexts, they
used Does he... ? with only minor deviations; for the 5N
control subjects, there were four occasions (contributed by
one child) out of 51 utterances total in which do was used
instead of does-for example, "Do he want the dog?" For
the SLI children, there were 2 utterances (contributed by one
child) out of a total of 7 utterances in which do was
substituted for does. The 3N children made no such errors in
a total of 10 utterances. Overall, the conclusion seems to
hold that if a BE or DO form is used, it is a finite form.
If the children use a form, do they know it has to show
agreement? That is, do they select the correct form? For
each group of children, there were 10 tested contexts for
agreement marking for BE and two contexts for DO. These
findings are laid out in Tables 10-14, 5 and summarized in
Table 15.
50ne of the SLI children was excluded from these analyses because this was
the only child who demonstrated agreement errors. This child made several
errors of the sort, "Is the bears are sleeping?" Furthermore, this child made
noninversion errors, such as "He is running?" It is not clear how to interpret
these errors, except to say that, for the analyses reported here, this child's
grammar seems to be not quite the same as that of the other children. The
generalizations that hold for the other children, drawn from the group
TABLE 10. Agreement for BE copula in interrogatives.
3
rd
person
singular
5N subjects (N = 21)
agree
nonagree
Total
% Agree
3N subjects (N = 12)
agree
nonagree
Total
% Agree
SLI subjects (N = 13)
agree
nonagree
Total
% Agree
3
rd
person
plural
TABLE 11. Agreement for BE auxiliary in interrogatives.
5N subjects (N = 21)
agree
nonagree
Total
% Agree
3N subjects (N = 12)
agree
nonagree
Total
% Agree
SLI subjects (N = 13)
agree
nonagree
Total
% Agree
3 rd
3rd
2nd
person
singular
person
plural
person
singular
57
0
57
100%
53
3
56
95%
64
1
65
98%
14
1
15
93%
9
2
11
82%
25
0
25
100%
10
0
10
100%
12
1
13
92%
14
0
14
100%
As reported in Table 15, for the 5N group, summing
across the contexts for BE, in 601 out of 606 contexts, they
chose the correct form, for 99% accuracy of agreement; for
DO, 106 out of 112 uses were correct, for 95% accurate
form choice. For the 3N group, for BE, in 221 out of 237
contexts they chose the correct form, for 93%; for DO,
19/23, for 83% accurate form choice. For the SLI children,
for BE, 149/158 for 94%; for DO, 14/17 for 82% accurate
form choice. (If we use a criterion of 80% as near accuracy,
summaries and applicable to individual children as well, do not seem to hold
in this individual case.
TABLE 12. Agreement for DO in interrogatives.
2 nd
3rd
person
singular
person
singular
person
plural
51
4
55
93%
55
2
57
96%
10
0
10
100%
9
4
13
69%
5
2
7
71%
9
1
10
90%
3
rd
5N subjects (N = 21)
58
0
58
100%
59
1
60
98%
61
0
61
100%
12
2
14
86%
9
3
12
75%
21
4
25
84%
10
0
10
100%
5
0
5
100%
10
0
10
100%
agree
nonagree
Total
% Agree
3N subjects (N = 12)
agree
nonagree
Total
% Agree
SLI subjects (N = 14)
agree
nonagree
Total
% Agree
860
Journal of Speech and Hearing Research
38
TABLE 13. Agreement for BE copula in statements.
5N subjects (N = 21)
agree
nonagree
Total
%Agree
3N subjects (N = 18)
agree
nonagree
Total
%Agree
SLI subjects (N = 18)
agree
nonagree
Total
%Agree
3rd
3rd
person
plural
64
0
64
100%
60
0
60
100%
48
1
49
98%
25
1
26
96%
32
0
32
22
6
28
79%
100%
11/12 contexts are at this level.) Overall, the children showed
a high probability of using the proper form-even children in
the SLI group. Of particular interest is the fact that for the SLI
children, in the 62 elicited questions with BE, there was only
one form choice error. Thus, as predicted by the model
proposed here, when the BE form was moved to the front of
the sentence, it was marked + for tense and agreement. The
one context that elicited an interesting low performance on
the part of the SLI children is in the "Does he want a
blanket?" context, where in 2 out of 7 cases the children
said "Do he want a blanket?" There were no instances of
does/do, as in *doesyou? or *doesthey?
Further evidence of accurate form choice is apparent in
BE
rd
5N subjects (N = 21)
agree
nonagree
Total
% Agree
3N subjects (N = 18)
agree
nonagree
Total
%Agree
SLI subjects (N = 18)
agree
nonagree
Total
% Agree
3
rd
person
plural
99%
95%
3N
(601/606)
93%
(106/112)
83%
(221/237)
(19/23)
94%
(149/158)
82%
(14/17)
SLI
the children's spontaneous samples. Table 16 shows the
percentage of correct uses of a form of BE, given that a form
is used. The table shows that the children's form choice was
highly accurate. The normally developing children almost
always chose the proper form. This characterization holds
for the children in the SLI group as well; in a total of 342
uses, there were 9 wrong choices.6 Note that agreement
was evident even for the less frequent BE forms (e.g., am). In
these data, when the children used questions-that is, when
BE was moved to the front of the sentence-the form, if
used, was almost always correct for agreement.
Evaluation of the children's spontaneous uses of DO
forms revealed a slightly different pattern. In these samples,
in contrast to the elicited probe items, DO appeared in
negation contexts. A number of occurrences of nonstandard
third-person singular uses of DO appeared in the form of *he
don't, *it don't, or *this don't. Because this is an acceptable
dialectal variant for many speakers in this region, it is not a
6
The use of infinitival BE as a main verb is very rare in these samples, as
predicted by the EOI. Comparison data, from a different sample of 11 3- to
5-year-old children with SLI, shows a similar picture. Hadley and Rice (1995)
studied the emergence of BE and DO. Inthat study, six instances of infinitival
BE as main verb were observed in a total of 1656 obligatory contexts. An
example is "The lion be here."
TABLE 16. Accuracy of BE agreement: spontaneous samples.
5N
am
are
63
0
63
100%
36
1
37
97%
19
1
20
95%
62
0
62
100%
22
1
23
96%
15
1
16
94%
DO
5N
TABLE 14. Agreement for BE auxiliary in statements.
3
August 1995
TABLE 15. Accuracy of agreement: probe data.
person
singular
person
singular
850-863
is
3N
am
are
is
SLI
am
are
is
Statements
Questions
100%
(78/78)
100%
(115/115)
100%
(1/1)
100%
(19/19)
99%
99%
(528/530)
(116/117)
100%
(14/14)
98%
(57/58)
0
100%
(5/5)
99%
99%
(360/363)
(114/115)
98%
(44/45)
88%
(23/26)
99%
(211/214)
0
100%
(2/2)
96%
(53/55)
Rice et al.: Extended Optional Infinitive 861
TABLE 17. Accuracy of DO agreement in third-person (does):
spontaneous samples.
5N
3N
SLI
Statements
Questions
100%
(27/27)
89%
(17/19)
73%
(8/11)
100%
(29/29)
91%
(20/22)
100%
(4/4)
clear instance of an agreement error. Furthermore, this
structure appeared in all three groups, although there were
slightly more of these for the SLI sample.
The results are summarized in Table 17, excluding the
don't dialectal variants. For the 5N group, their choice of
does is always accurate. For the 3N group, there were four
instances of do, two times in questions and two times in
statements. For the SLI group, does was evident in 8 out of
11 statements. One of the errors was *he just do it, which
most likely is a main verb use of do, produced in the
infinitival form, a possibility consistent with the 01 model.
The other two instances might be performance errors,
although it is hard to tell because of the ellipsis: *somy mom
do and *shedo. Note that, as we have seen previously, when
DO appears in questions, the children are not likely to make
errors in their form choice.
In summary, almost none of the children make agreement
errors on forms of DO and BE, nor do the children use
infinitival forms of DO and BE when finite forms are required.
These results are consistent with Prediction 6.
Discussion
------`-`
The main findings from this study offer support for the
predictions of an Extended Optional Infinitive account of SLI.
As predicted, children with SLI are likely to omit tense
marking on lexical verbs. This is evident with regard to -s
and -ed affixation, in spontaneous samples and on elicitation
tasks. This optional marking is also evident for their younger
language-matched controls, but it is even more likely for
children with SLI. At the 5-year-old level sampled here, their
omission rates were around 75% for -s and -ed, compared
to around 50% for the 3N group.
With English lexical verbs it is not easy to sort out
important distinctions regarding what the children know
about finite versus nonfinite forms, tense, agreement, and
related rules for verb movement. With such high rates of
omissions, it may seem as if they would have little in the way
of underlying grammatical representations of tense-marking
in the verb phrase. This possibility can be evaluated within
the English BE and DO system, where finiteness, tense,
agreement, and movement to the front of the sentence can
be observed.
As with the lexical verbs, children in the SLI group
frequently omitted BE and DO at rates of 40%-60%, typically 20% more than their language-matched comparison
group. At the same time, however, when they did use a form,
it was a finite form and was almost always correctly marked
for tense and agreement.
Conclusions
We conclude that the evidence supports the EOI model of
SLI. In this model, the grammar of children with SLI, for the
features tested here, parallels that of younger normally
developing children. The children with SLI show extreme
variation in the mastery of a grammatical process that is
slow to appear in younger normally developing children.
These youngsters, and perhaps older individuals, do not
seem to know that tense-marking is obligatory in a main
clause. This lack of knowledge for children with SLI manifests itself in two ways: First, they produce a higher proportion of nonfinite matrix clauses than expected for their MLU
level; and second, they persist in producing nonfinite matrix
clauses to an older age than do normal children.
At the same time, other features of the linguistic representations are, apparently, intact. Thus, at the time when there
is optional marking of tense in the main clause, there is
knowledge about finite forms and obligatory ways to mark
finiteness. There is reason to believe, on the basis of the
findings reported here, related theoretical reasoning, and
observations from children learning languages other than
English (cf. Wexler, 1994), that children with SLI who have
demonstrated receptive and expressive language limitations
nevertheless know many important grammatical properties,
including the following:
* the distinction between finite and nonfinite forms of
verbs
* the requirement that if the surface form is finite it must
carry tense and agreement markings
* subject-verb agreement
Caveats
It must be acknowledged that the available empirical
evidence, from this and other studies, with regard to the
conclusions drawn above, is much stronger for the lexical
verb data than for the analysis of BE and DO. Although other
investigators have reported on copula and auxiliary uses of
BE, and those analyses confirm the likelihood that children
with SLI will omit those forms, there is limited evidence
available that differentiates between use in statements versus declaratives. This study is the only report of DO use by
SLI children to our knowledge and the only study (with the
exception of Hadley & Rice, 1995) that attempts to explore
the question of finiteness and agreement-marking on surface forms of BE and DO by children with SLI. For the latter
analyses, the data available here were relatively sparse, and
reported in aggregate form, across all the children in the
group. On the other hand, it is also the case that inspection
of the data reveals that the finiteness and agreement generalizations (summarized above) seem to hold for virtually
every child observed. The point here is that additional
evidence at the same level of detail will be necessary inorder
to arrive at empirically robust generalizations.
862
Journal of Speech and Hearing Research
Implications for Accounts of SLI
The EOI account posits that children with SLI are not
unlike nonimpaired children in their incomplete representations of the obligatory nature of tense marking, coexisting
with their knowledge of the finite/nonfinite distinction and
related processes. In this way, the EOI account is a version
of an Extended Development account, inwhich children with
SLI maintain a "young" grammar much longer than do
nonimpaired children. Furthermore, this grammar recognizes the distinction between finite and nonfinite forms of
verbs, the necessity of finite verbs' carrying tense and
agreement markings, and the necessity of subject-verb
agreement marking. Thus, this model shows, in detail, the
ways inwhich the grammar of affected individuals compares
to that of unaffected individuals.
What the EOI theory suggests is that a core grammatical
limitation of children with SLI is attributable to a prolonged
period in which they do not know that finiteness marking is
obligatory in matrix clauses. Grammatical "impairment"
under this model can be thought of as a "preference" for
nonfinite clauses, although "preference" implies a consciousness of choice that does not literally apply to this
phenomenon. "Impairment" can be manifest in three possible ways: (a)later-than-expected emergence of finite forms
in SLI children's utterances (This finding is reported by
Hadley & Rice, 1995, in their investigation of the emergence
of BE and DO forms in the utterances of young children with
SLI.); (b) lower-than-expected optional use of finite forms
once finiteness marking emerges (as demonstrated in this
study); and (c) a longer-than-expected period of 01. New
findings suggest this period may extend into the elementary
school years (Marchman & Weismer, 1994) and on into
adulthood (Tomblin, 1994; Ullman & Gopnik, 1994).
This model can be compared to a Language Delay (LD)
model of SLI (cf. Lahey, Liebergott, Chesnick, Menyuk, &
Adams, 1992), in which it is postulated that these youngsters
demonstrate a delay inall aspects of language. An LD model
presumes that emergence of the expected language developmental milestones is delayed in some way that affects the
general linguistic system. What is underestimated by the LD
model is the extent to which an EOI stage is evident, relative
to non-EOI grammatical morphemes (such as plurals and
prepositions), relative to general linguistic productivity (as
indexed by the mean length of utterance) and relative to
long-term outcomes. Children's appreciation of the need to
mark tense does not just appear somewhat later, along with
other nonrelated later manifestations of an emergent linguistic system. An EOI stage persists beyond the time of
mastery of other grammatical features (inthe same way as in
the nonimpaired children, but more so), persists to higher
levels of general linguistic productivity (indexed by MLU),
and may remain unresolved long into the developmental
trajectory. In particular, the results in this paper show that
although the optional use of tense is lower than expected in
children with SLI, other properties of the tense and agreement system are not affected-for example, subject-verb
agreement and the necessity of agreement and tense in
sentence-initial verbs.
The EOI can be compared as well to a general processing
38
850-863
Ag
1995
model. Several versions of processing models exist (e.g.,
Leonard, 1989; Tallal & Piercy, 1975; Tallal & Stark, 1981).
What is characteristic of this line of explanation is a hypothesized breakdown of some sort in the input available to
children with SLI, either at the level of auditory input or inthe
formulation of linguistic paradigms. The finding that children
with SLI seem to control agreement-marking for BE and DO
poses a challenge for a processing account of their grammatical limitations. The most detailed version of how a
processing limitation could affect the grammar is offered by
Leonard (1989) and is known as the Low Phonetic Substance model. In this model, the fact that children with SLI
omit surface forms is interpreted as a processing problem
attributable to the nonsalient properties of these singlesyllable or single-phoneme morphemes, perhaps in combination with the cognitive demands of paradigm-building for
the ways in which tense or agreement is marked in a given
language (e.g., Leonard, 1989). By this account, the input to
the language acquisition mechanism is somehow distorted,
such that the small unstressed parts get dropped out or
incompletely registered as candidates for paradigm placement. The learning of the BE and DO paradigm, then, should
be faulty. If the input forms are not registered, it is expected
that omissions should occur. At the same time, with incomplete or distorted input or faulty paradigm-building, errors of
form choice, such as *she am or I1is, should be expectedperhaps especially so for low-frequency forms. Yet the
children in this study were quite accurate in their choice of
the proper suppletive form, even for the low-frequency form,
am. It seems that if the children were able to process the
forms well enough to encode and store the proper form, and
to choose the proper form without error, in some sense the
paradigm is known. What is not specified, on a processing
account, is what would block access to those forms. The
advantage of the framework presented here is the posited
differentiation of tense and agreement, which allow for the
expression of agreement and tense on BE and DO at the
same time that tense-marking does not appear on the lexical
verb.
The import of the EOI account for models of SLI is that it
identifies a predicted cluster of affected morphemes. This
places those morphemes in an underlying morphosyntactic
structure in which finiteness is operative, shows how the
grammar of children with SLI is similar to and different from
that of nonimpaired children, and reveals phenomena that
challenge existing models of SLI. Evidence of an EOI stage
may prove to be helpful in the specification of clinical
markers of early stages of language impairment, and may
reveal ways in which properties of deviant language acquisition are manifest in different natural languages.
Acknowledgments
This study was supported by National Institute of Deafness and
Communicative Disorders Awards R01 NS26129 to Mabel L. Rice,
and R01 DC01803 to Mabel L. Rice and Kenneth Wexler. We
express special appreciation to Mary Howe for her assistance with
data management and data analyses, and to Janna Oetting for her
assistance with data collection. Our special appreciation is expressed to the children who participated inthis study, their parents
Rice et al.: Extended Optional Infinitive 863
who provided permission, and the day care centers and preschools
that supported this research. In Lawrence, Kansas they are: Children's Learning Center, Community Child Care Services, Head
Start, Hilltop Child Development Center, Stepping Stones, and
United Child Development Center. Also, the Hutchinson Early Education Center, Hutchinson, KS, and the public school systems of
Baldwin and Shawnee Mission, KS, whose cooperation made this
study possible.
References
Bishop, D. V. M. (1992). Biological basis of specific language
impairment. In P. Fletcher & D. Hall (Eds.), Specific speech and
language disorders in children: Correlates, characteristics, and
outcomes (pp. XXX). San Diego, CA: Singular Publishing Group.
Burgemeister, B., Blum, H., & Lorge, I. (1972). The Columbia
Mental Maturity Scale. San Antonio, TX: Psychological Corporation.
Chomsky, N. (1957). Syntactic structures. The Hague: Mouton
Publishers.
Dunn, A., & Dunn, A. (1981). Peabody Picture Vocabulary TestRevised. Circle Pines, MN: American Guidance Service.
Goldman, R., & Fristoe, M. (1986). Goldman-Fristoe Test of Articulation. Circle Pines, MN: American Guidance Service.
Hadley, P. A. & Rice, M. L. (1995). Categorical distinctions between
early markers of finiteness: Evidence from children with specific
language impairment. Submitted for publication.
Halle, M., & Marantz, A. (1993). Distributed morphology and the
pieces of inflection. In K. Hale & S. J. Keyser (Eds.), View from
building 20: Essays in linguistics in honor of Sylvain Bromberger
(pp. 111-176). Cambridge, MA: MIT Press.
Howe, M. L. (1992). Transcription and coding manual for analysis of
language samples: Kansas language transcript database. Unpublished manual, University of Kansas, Lawrence, KS.
Lahey, M., Liebergott, J., Chesnick, M., Menyuk, P., &Adams, J.
(1992). Variability in children's use of grammatical morphemes.
Applied Psycholinguistics, 13, 373-398.
Leadholm, B., & Miller, J. (1993). Language Sample Analysis: The
Wisconsin Guide. Milwaukee, WI: Wisconsin Department of Public Instruction.
Leonard, L. B. (1989). Language learnability and specific language
impairment in children. Applied Psycholinguistics, 10, 179-202.
Leonard, L. B., Bortolini, U., Caselli, M. C., McGregor, K. K., &
Sabbadini, L. (1992). Morphological deficits in children with
specific language impairment. The status of features in the
underlying grammar. Language Acquisition, 2, 151-179.
Marchman, V. A., & Weismer, S. E. (1994, June). Pattems of
productivity in children with SLI and NL: A study of the English
past tense. Paper presented at the Symposium for Research in
Children's Language Disorders, Madison, WI.
Miller, J. F., & Chapman, R. S. (1991). Systematic Analysis of
Language Transcripts. Madison: University of Wisconsin Language Analysis Lab.
Oetting, J. B., & Rice, M. L. (1993). Plural acquisition in children
with specific language impairment. Journal of Speech and Hearing Research, 36, 1236-1248.
Pierce, A. E. (1992). Language acquisition and syntactic theory: A
comparative analysis of French and English child grammars.
Boston: Kluwer Academic Publishers.
Plante, E., Swisher, L., Kiernan, B., & Restrepo, M. A. (1993).
Language matches: Illuminating or confounding? Journal of
Speech and Hearing Research, 36, 772-776.
Poeppel, D., & Wexler, K. (1993). The full competence hypothesis
of clause structure in early German. Language, 69, 1-33.
Quirk, R., Greenbaum, S., Leech, G., & Svartvik, J. (1985). A
comprehensive grammar of the English language. New York:
Longman Inc.
Rice, M. L. (1994). Grammatical categories of children with specific
language impairments. In R. V. Watkins & M. L. Rice (Eds.),
Specific language impairments in children (pp. 69-88). Baltimore,
MA: Brookes Publishing Co.
Rice, M. L., & Oetting, J. B. (1993). Morphological deficits of
children with SLI: Evaluation of number marking and agreement.
Journal of Speech and Hearing Research, 36, 1249-1257.
Stevens, J. P. (1986). Applied multivariate statistics for the social
sciences. Hillsdale, NJ: Erlbaum Associates.
Tallal, P., & Piercy, M. (1975). Developmental aphasia: The perception of brief vowels and extended stop consonants. Neuro Psychologia, 13, 69-74.
Tallal, P., & Stark, R. (1981). Speech acoustic-cue discrimination
abilities of normally developing and language-impaired children.
Journal of the Acoustical Society of America, 69, 568-574.
Tomblin, B. (1994, February). Family and twin studies of language
impairment. In M. Rice (Chair), Inherited speech and language
disorders: In search of a phenotype. Session conducted at the
AAAS Annual Meeting, San Francisco.
Ullman, M., & Gopnik, M. (1994). Past tense production: Regular,
irregular, and nonsense verbs. In Linguistic aspects of familial
language impairment (pp. 81-118). Special Issue of McGill Working Papers in Linguistics, Vol. 10. Montreal, Canada: McGill
University.
Watkins, R. V., & Rice, M. L. (Eds.) (1994). Specific language
impairments in children. Baltimore, MD: Brookes Publishing Co.
Weissenbom, J. (1994). Constraining the child's grammar: Local
wellformedness in the development of verb movement in German
and French. In B. Lust, J. Whitman, & J. Kornfilt (Eds.), Syntactic
theory and language acquisition: Crosslinguistic perspectives.
Vol. 1. Phrase structure. Hillsdale, NJ: Erlbaum.
Wexler, K. (1994). Optional infinitives. In D.Lightfoot & N. Hornstein
(Eds.), Verb movement. New York: Cambridge University Press.
Received June 1, 1994
Accepted December 22, 1994
Contact author: Mabel L. Rice, PhD, University of Kansas, Child
Language Program, 1082 Robert Dole Human Development Center,
Lawrence, KS 66045.