int. j. language & communication disorders, 1999, vol. 34, no.

2, 117134

Sentence processing strategies in children

with expressive and expressivereceptive
speci c language impairments
Julia L. Evans and Brian M acW hinney

U niversity of W isconsin, M adison, W I, U SA

Carnegie M ellon U niversity, Pittsburgh, PA, U SA

(R eceived July 1997; accepted January 1998 )

Abstract
The purpose of this study was to investigate the sentence comprehension
strategies used by children w ith expressive and expressivereceptive speci c
language impairments (SLI ) within a language processing framework. Fourteen
children with SLI (ages 6;107;11) meeting strict selection criteria were compared to seven age-m atched and seven younger normal controls. Children were
asked to determine the agent in sentences composed of two nouns and a verb
( NVN, NNV, VNN ) with animacy of the noun as a second factor. Results of
group comparisons revealed that children with E-SLI and ER-SLI diV ered from
each other in the comprehension strategies they employed as well as diV ering
from both age-m atched and younger normal language control groups. Children
with E-SLI relied exclusively on a rst noun as agent strategy across all conditions, whereas children with ER-SLI used animacy cues when available.
Addit ionally, maximum likelihood estimates were calculated to investigate individual patterns of performance under diV erent cue conditions. Results revealed
a sign i cant correlation between severity of receptive language abilities and the
type of strategy used, with better receptive language skills being highly correlated
with childrens use of word order cues.
Keyword s : language
comprehension.

disorders,

language

processing,

SLI

subgroups,

Introduction
The term speci c language impairments (SLI ) refers to a group of children who demonstrate language disorders in the absence of any clearly identi able aetiology.
Traditionally, these children have been treated as a single population de ned on
the basis of exclusionary criteria (e.g. normal non-verbal intelligence, normal hearing,
A ddress correspondenc e to: Julia L. E vans Ph.D ., Waisman R esearch Cen ter, R oom 445, U niversity
of W isconsin , M adison, W I 53705, U SA.

International Journal of Language & C ommunication Disord ers

ISSN 1368-2822 print/ISSN 1460-6984 online 1999 Royal College of Speech & Language Therapists
http://www.tandf.co.uk/JNLS/lcd.htm
http://www.taylorandfrancis.com/JNLS/lcd.htm

118

J. L. Evans and B. MacW hinney

and no visible sign of neurological or emotional impairments; Stark and Tallal 1981).
It appears, however, that SLI is not an homogeneous group. In particular, research
suggests that children with both expressive and receptive de cits ( ER-SLI ) can be
contrasted to those having only expressive de cits ( E-SLI ).
Findings from studies employing status assessment measures (e.g. Aram and
Nation 1975, Wolfus et al. 1980, Stark and Tallal 1988) indicate that children with
E-SLI demonstrate good receptive vocabulary, syntax comprehension, normal
mem ory span, and good phonological discrimination, but have de cits in expressive
semantics and syntax as well as formulation diY culties in rapid motor sequencing.
In contrast, children with ER-SLI exhibit de cits in receptive vocabulary, syntax
comprehension, phonological discrimination and decreased mem ory span in conjunction with de cits in expressive syntax and semantics. However, production of
consonant clusters, diphthongs, and multisyllabic words is signi cantly better for
ER-SLI than for E-SLI children.
Experimental studies also re ect subgroup diV erences. At the lexical level,
children with poor receptive and expressive language abilities are slower and less
e Y cient in word- nding, picture naming, and auditory rehearsal skills ( Leonard
et al. 1983, Kail and Leonard 1986, Gathercole and Baddeley 1990). At the discourse
level, these childrens conversational responses are slower and less e Y cient than
children with good receptive language abilities (e.g. Craig and Evans 1989, 1993).
In contrast, in spontaneous speech, SLI children with good receptive language
abilities make more gram matical errors, and omit more function words in utterances
that have greater verbal and non-verbal processing demands, compared to children
with poor receptive and expressive skills ( Evans 1996, Evans et al. 1997).
Studies of the comprehension strategies used by children with SLI also demonstrate subgroup diV erences that seem to vary with the severity of the childrens
receptive language de cits. In particular, children with good receptive skills appear
to use more word order strategies, whereas children with poor receptive language
abilities rely more on semantically based strategies. For example, on standardized
assessment measures (e.g. Test of Receptive Gram mar ( TROG ); Bishop 1979), SLI
children with receptive de cits had diY culty using word order cues to interpret
sentences. In sentence processing tasks designed to investigate childrens use of
word order versus semantic strategies, children with poor receptive and expressive
skills (e.g. ER-SLI ) exhibited chance levels of performance for sentences when only
word order cues were made available (van der Lely and Dewart 1986, van der Lely
and Harris 1990).
In contrast, in studies of children with good receptive language skills (e.g.
E-SLI ), it can be seen that these children seem to be able to use word order cues
in sentence comprehension tasks ( Bishop 1982, Adam s 1990). For example, Adam s
(1990) found that children with good receptive language skills were able to use
word order cues to correctly comprehend reversible active and double object
sentences; constructions requiring the use of word order cues. Further analysis of
these childrens errors patterns for embedded clauses, embedded phrases, and
passives also revealed their use of word order based strategies. Thus, it appears that
SLI childrens ability to use word order cues to comprehend sentences may vary
with severity of receptive language de cits. However, this subgroup question has
yet to be addressed directly.
Recently, several processing based models have been proposed as possible
accounts of speci c language impairments (e.g. Bishop 1992, Johnston 1995,

Sentence processing in subgroups of SLI

119

Leonard 1995). However, these accounts focus primarily on the expressive language
de cits characteristic of this population. Although there is reason to suspect a
relat ionship between processing de cits and expressive language impairments, little
research has focused on comprehension de cits in this group from a processing
based framework. The model employed in this study to investigate a performance
based account of sentence comprehension in children with SLI was the competition
model developed by Bates, MacWhinney and their colleagues ( Bates et al. 1984,
Bates and MacWhinney 1987, 1989).
During sentence comprehension the child is required to attend simultaneously
to multiple transient and eeting acoustic, phonetic, semantic, morphological,
syntactic and intonational cues. The Com petition Model provides an account of
language comprehension which focuses on this process of converting multiple
competing cues into a linear speech stream . Three key concepts cue validity, cue
strength, and cue cost are included in the model to capture the multidimensional
nature of language comprehension. W ithin this theoretical framework, cue valid ity is
de ned as the information value of a given lexical or gram mat ical feature.
Operationally, it is the product of a cues availability and reliability. Or more simply
put, cue validity is the proport ion of the time a cue is present and indicates the
correct interpretat ion ( i.e. role assignment) in a given language.
Cue strength, in contrast, is a measure of the strength to which a given individual
has approximated the validity of a given cue in his or her language. During language
acquisition, the process of determ ining the roles of nouns as actor or patient is a
diY cult task. The mappings between cues that mark these roles and the roles
themselves are many-to-many. For example, the cue for noun animacy tends to
mark the role of actor in many cases (e.g. the boy hit the ball). Yet the an imacy
cue is not always present (e.g. the bat hit the ball), or correct in marking the role
of actor (e.g. the ball hit the boy). Thus, according to the competition model, the
strength of a cue for an individual speaker is proportional to its informational value
or cue valid ity.
Individual limitations in processing capacity during real-time language use will
a V ect cue strengths for individual speakers. The third theoretical concept, cue cost,
is a general measure of the processing costs associated with the real-time use of a
given cue (e.g. mem ory, speed of processing limitation, fatigue). The Com petition
Model holds that the processing costs of diV erent linguistic cues will diV er across
individual speakers, re ecting the individuals information processing abilities.
Cue cost and cue validity a V ect not only the degree to which children will
believe or trust certain cues, but also the developm ental order with which children
come to rely on diV erent cues. For example, children will pick up cues in order of
their validity w ith the rst cues learned by the child being the most reliable ones
in his or her language. However, if a given cue has high cue validity but also high
processing cost, there will be a delay in the childs use of the cue. Studies of typically
developing children reveal that, in the early stages of language developm ent ( i.e. 2
years of age), children rely primarily on an imacy cues to interpret sentences, yet by
4 years of age, English speaking children begin to rely more on word order cues
to process canon ical word order (SVO), but continue to rely on animacy cues for
non-canon ical word orders (OSV, and VOS). By 7 to 12 years of age, however,
children are able to use word order cues to process non-canonical (OSV and VOS)
forms as well ( Bever 1970, Nelson 1974, Bates et al. 1984, von Berger et al. 1993).
Many of the important predictions of the Com petition Model are based on the

120

J. L. Evans and B. MacW hinney

processing of sentences in which multiple linguistic cues are in agreem ent or in

competition. The Com petition Model predicts that the cost associated with the realtime processing of multiple cues will diV er depending on the patterns of
co-occurrence of these cues. For example, in English sentences with the rst noun
an imate and the second inan imate (e.g. NVN-AI ) both word order and an imacy
cues suppor t the interpretation of rst noun as agent. However, for NVN sentences
where the rst noun is inanimate and the second is animate (e.g. NVN-IA), word
order and animacy cues are in con ict, with word order cueing the rst noun for the
role of agent, but animacy cueing the second noun. The use of word order cues
during language comprehension requires the child to process and temporarily store,
in serial order, both immediate and nal interpretat ions of multiple competing
mean ings. The competition model predicts that, if processing costs exceed available
resources, children with SLI may rely instead on comprehension strategies (e.g.
whole world semantic knowledge) which have high cue validity but lower cue costs.
Detailed examination of SLI childrens responses in competing and co-occurring
cue conditions provides a framework to investigate possible subgroup diV erences
in the use of word order versus animacy sentence processing strategies. The goal
of this study, therefore, was to employ a processing based model of comprehension
to investigate the sentence processing strategies of children with SLI diV ering in
receptive language abilities. Two speci c questions were posed:
(1) Do the sentence processing strategies used by children w ith SLI vary with
the severity of receptive language de cits?
(2) W ith changes in cue con ict levels, do the sentence comprehension strategies
change for children with E-SLI and ER-SLI?
M ethod
Participants
The subjects for this investigation were 28 children: 14 with SLI (ages 6;97;11),
seven chronologically age-m atched controls (ages 6;10 7;9) and seven younger
normal language controls (ages 4;0 4;2). The children with SLI were selected to
meet the following criteria:
(1) non-verbal IQ greater than 85 based upon either the Weschler Intelligence
Scale for Children-Rev ised ( W ISC-R ) ( Wechsler 1974), or the Columbia
Scale of Mental Maturity (CMMS) ( Burgemeister et al. 1972);
(2) speech intelligibility greater than 95% as measured by the Arizona
Articulation Performance Scale (AAPS) ( Fudala and Reynolds 1986 );
(3) hearing within normal limits;
(4) no signi cant emotional or frank neurological impairments.
Children were further selected for this study to comprise two distinct groups
diV ering in receptive language abilities ( table 1), based upon the composite receptive
and expressive language scores from the Clinical Evaluation of Language Functions
Revised (CELF-R; Semel et al. 1987 ). The CELF Expressive Language Score
comprises three subtests:
(1) word classes ( W S);
(2) formulating sentences ( FS); and
(3) recalling sentences ( RS).

Sentence processing in subgroups of SLI

121

Table 1. Means and standard deviation s for age, and the

expressive and receptive language scores for the CELF-R
(SS) for expressive (E-SLI) and expressivereceptive
(ER-SLI) subjects
CEL F-R

E -SLI
E R-SLI
CA

A ge
M ean (SD )

( R LS)
M ean (SD )

( E LS)
M ean (SD )

7;6 (3.6)
7;3 (3.8)
7;4 (3.5)

89 (6.5 )
70 (6.7 )
NA

65 (2.9)
50 (1.0)
NA

The CELF Receptive Language Score comprises three subtests:

(1) linguistic concepts ( LC );
(2) sentence structure (SS); and
(3) oral directions (OD).
In keeping with prior subgroup classi cation criteria ( Evans 1996, Evans et al.
1997 ), children were designated as E-SLI if their composite receptive language
scores were less than 1 SD from the mean (x = 89; SD 6.7), but their expressive
language scores were more than 2 SD below the mean (x = 65; SD 2.9). Children
were designated as ER-SLI if their composite receptive language scores were more
than 1.5 SD below the mean (x = 70; SD 6.5), and their expressive language scores
were more than 2 SD below the mean as well (x = 50; SD 1.0).
Task
Fifty-four gram matical and semi-grammatical sentences comprised of two nouns
and a verb with animacy as a second factor were presented to the children. The
sentences were of the form NVN, NN V, and VNN, consistent with previous
research with the Com petition Model (e.g. Bates et al. 1984). Nouns were either
an imate/an imate (AA), an imate/inan imate (AI ), or inan imate/an imate ( IA). The
sam e enactm ent paradigm used in previous research with school-aged children
within the Com petition Model ( Wulfeck 1993) was used in this study. Children
were presented pictures of two nouns and asked to point to the one that was
doing the action.
To prevent possible interference of role assignment due to intonational cues,
all sentences were presented with the sam e intonational contour pattern appropriate
for well-formed sentences. Speech stimuli were digitized using SoundEditPro, and
stored in 16 bit, 22 Hz format on a Macintosh Quadra 660AV. The sentence stimuli,
the position of the correct answer, and the order of presentation were generated
random ly using PsyScope (Cohen et al. 1993). Examples of gram matical and semigram mat ical forms as well as the nouns and verbs used to construct the sentences
are shown in the Appendix.
Each child was tested individually in a quiet room. Prior to the experimental
task, each child was presented 10 training items which were similar but not identical
to the test sentences. Children were instructed to listen carefully since they would
hear each sentence only once. All of the children were able to complete the training

122

J. L. Evans and B. MacW hinney

items. After the training items were completed, the children again were rem inded
to choose the animal or object they thought was doing the action and to listen
carefully, as they would hear each sentence only once.
Results
The rst question was, how do the two SLI groups diV er from each other and their
age-m atched peers with respect to sentence comprehension strategies? A 3 3 3
mixed-model analysis of analysis of variance (AN OVA) was conducted with percentage choice of the rst noun as the dependent variable. The two within-subject
factors were word order ( NVN, NN V, VNN ) and animacy (AA, AI, IA) and the
between-subjects factor was group ( E-SLI, ER-SLI, CA). The results revealed
signi cant main e V ects for word order (F(2,36 ) = 16.15, p < 0.001), animacy
(F(2,36 ) = 23.2, p < 0.001) and group (F(2,18)= 27.6, p < 0.001). There also was a
signi cant group-by-an imacy interaction e V ect (F(4,36) = 4.43, p < 0.01). Post hoc
analysis revealed that the E-SLI and CA groups did not diV er signi cantly from
each other, but that the ER-SLI group diV ered signi cantly from both the E-SLI
and CA groups.
The results can be understood more clearly by looking at gures 1 4, which
illustrate the group diV erences in sentence comprehension strategies. The percentage
rst noun choice for the three word order conditions for the E-SLI, ER-SLI, and
CA groups is shown in gure 1. The rst noun was the predominant choice for all
three word order conditions for the children with E-SLI. For the CA group, a slight
shift in the pattern of noun choice can be seen between the NVN and the other
two conditions. In the NVN condition, rst noun choice was high for the CA
group. However, in the NN V and VNN conditions, the CA group shifted to a

Figure 1.

Word order e V ects for E -SLI, E R-SLI and CA-matched controls.

Sentence processing in subgroups of SLI

123

second noun choice strategy; a pattern consistent with an OSV strategy previously
reported for children this age and adults ( MacWhinney et al. 1985, von Berger et al.
1993). For the ER-SLI group, however, the percentage rst noun choice was at or
slightly above chance across levels for all three word order conditions (4058%).
These group diV erences can be seen more clearly across the three an imacy
conditions shown in gure 2. The percentage rst noun choice for the E-SLI group
was high across all three animacy conditions, and did not diV er statist ically, nor did
the E-SLI group diV er from the CA group across the three conditions. The
percentage rst noun choice was signi cantly diV erent across the three conditions
for the ER-SLI group, however. For the AA condition in which both nouns were
an imate (e.g. the condition where animacy is e V ectively rem oved as a cue), the
percentage rst noun choice for the ER-SLI group was at chance levels. For the
AI condition, in which only the rst noun was an imate, the percentage of rst
noun choice for the ER-SLI was high (79%) and for the IA condition, in which
the second noun was animate, percentage rst noun choice was low (19%), indicating
that these children with receptive de cits were relying not on word order but on
an imacy cues instead.
The diV erences between the two subgroups are the clearest in the NVN word
order condition shown in gure 3. For both the E-SLI and CA groups, the choice
of rst noun was high across all three animacy conditions. For the ER-SLI children,
choice of rst noun was slightly above chance levels in the condition where both
nouns were animate (AA: 64%); high when the rst noun was animate (AI: 90%);
and extremely low when the rst noun was inan imate ( IA: 21%), indicat ing that
the children with good receptive language abilities were using word order to process

Figure 2.

A nim acy e V ects for E -SLI, E R-SLI and CA-matched controls.

124

J. L. Evans and B. MacW hinney

Figure 3.

A nim acy e V ects for N VN word order.

sentences, whereas the children with poor receptive language abilities were relying
exclusively on animacy cues.
Three diV erent patterns of cue competition, contrasting a low, high and null
cue con ict contexts are presented in gure 4. The NVN-AI condition represents
a low cue con ict condition due to the convergence of the animacy and word order
cues which both support rst noun as agent. The NVN-IA pattern represents a
high cue con ict condition with animacy cueing the last noun as agent and word
order cueing the rst noun. The VNN-AA pattern represents a null cue condition
where there are no cues for either word order or animacy.
One might argue that little can be learned from the use of ungrammatical
sentences, as in the VNN-AA condition. However, patterns of responses to these
ungrammatical sentences can provide valuable insights into the strategies employed
by children with language disorders when confronted with unfamiliar or novel
sentence forms. MacWhinney et al. (1985) have shown that the processing of
ungrammatical sentences provides an accurate measure of cue strength in gram matical sentences. This can be seen clearly in the diV erent processing strategies used
by the two language-impaired groups and the normal 7;0 year old peers across the
three cue con ict conditions. In the NVN-AI, the percentage rst noun choice was
high for all three groups ( E-SLI, 100%; ER-SLI, 91%; CA, 100%). In the NVN-IA,
however, the performance of the E-SLI and the CA groups diV ered distinctly from
the ER-SLI group. For the E-SLI and CA groups, rst noun choice was still high
despite the competing animacy cues. For the ER-SLI children, however, rst noun
choice was extremely low (21%), indicating a strong second animate noun
choice instead.
In the VNN-AA condition, the pattern of performance for the three groups

Sentence processing in subgroups of SLI

Figure 4.

125

Percenta ge rst noun choice across high, low and null cue con ict conditi ons for E -SLI,
E R-SLI and CA-matched groups.

diV ered in interesting ways. For the E-SLI children, choice of rst noun was still
high, but for the CA-group, choice of rst noun drop ped to chance levels (58%),
suggesting that for typical 7;0 year old children, the absence of any word order or
an imacy cues profoundly disrupted their performance and resulted in guessing. In
contrast, the ER-SLI groups choice of rst noun was still low. In the absence of
an imacy cues, however, responses in this context suggest that these children were
choosing the last word they heard, indicating a possible recency e V ect. Thus, the
children with good receptive abilities appeared to use a strict rst noun strategy
across cue conditions, seemingly unaw are of the absence of word order or an imacy
cues in the VNN-AA condition. In contrast, the children with poor receptive
language abilities consistently relied on animacy cues when these cues were made
available, but appeared unable to hold the order of words in mem ory when an imacy
cues were not available, choosing the last noun in these contexts.
Given the ER-SLI groups reliance on animacy cues, we might ask whether
their processing strategies are characteristic of younger normally developing children.
The ER-SLI childrens expressive skills, as measured by standardized expressive
language measures, were comparable to those of 3 year olds, whereas their receptive
language scores were comparable to those of 4 year olds; thus data from 4 year old
normal language controls was collected instead. A 3 3 2 mixed-model analysis
of variance comparing the percentage rst noun choice for ER-SLI and the younger
4 year normal language control group was conducted, with the sam e two withinsubject factors (word order and animacy) and the between-subject factor being

126

J. L. Evans and B. MacW hinney

group ( ER-SLI, 4;0). Results revealed a signi cant main e V ect for word order
(F(2,24)= 14.1, p < 0.0001), animacy (F(2,24)= 17.4, p < 0.0001) and group
(F(1,12) = 5.89, p < 0.01), as well as a signi cant group-by-an imacy interaction e V ect
(F(4,48)= 9.44, p < 0.001). The pattern of performance for the ER-SLI and 4 year
old children is displayed in gure 5. The percentage rst noun choice for the
ER-SLI group was high in the AI condition and low in the IA condition. In
contrast, the percentage rst noun choice for the 4;0 control group was high across
an imacy conditions, similar to the performance of both the E-SLI and CA-matched
groups examined earlier. This suggests that children with ER-SLI are not simply
delayed in the comprehension strategies they use; instead, the processing strategies
they use are qualitatively diV erent from those of younger typically developing
children as well.
The results of the analysis of variance revealed that diV erences in the sentence
comprehension strategies for the two groups of children with SLI not only diV ered
from each other but from normally developing control groups as well. However,
this statistical approach does not provide information regarding the direction and
relat ive strength of the word order or an imacy cues for the two SLI groups nor
does it allow for analysis of individual patterns of performance. Maximum likelihood
estimat ion ( MLE ) techniques, in contrast, have been used successfully to investigate
individual diV erences in children with SLI in general ( Evans et al. 1997 ) and
in particular with this speci c task paradigm ( McDonald and MacWhinney
1989, Evans and MacWhinney 1995, 1996 ). MLE uses an iterative procedure in

Figure 5.

A nim acy e V ects for N VN word order for the E R-SLI and younger nor m al language control
groups.

Sentence processing in subgroups of SLI

127

conjunction with a mathem atical model to nd values for param eters in a model
which best t the data, thereby providing direct estimations of the relative strength
of cue usage for both groups and individuals. The speci c model used for this
study ( McDonald and MacWhinney 1989) predicts the probability of choosing the
rst noun as actor based on the strengths of word order and an imacy cues. The
model is presented below, where:
P (first noun choice) =

P i Si 1
S j (P i S i j )

where S i 1 is the strength with which cue i favours a noun candidate, i = 1 to I,

where I is the total number of cues available, and j = 1 to J, where J is the total
number of candidates for actor, in this case 2.
In this study, the children had to pick one of the two nouns as the actor of the
sentence. It was assumed that the degree to which either word order or animacy
resulted in the childs favouring one noun as the actor was an addit ive complement
to the degree to which the child favours the other noun. In the case of animacy
for example, if the child preferred the rst noun in an AI sentence at some given
strength S then the complementary strength of the childs favouring the second
noun was (1 S ).
Separate param eters were estimated for each level of each cue, with three levels
of animacy (AA, AI, IA) and three levels of word order ( NVN, NN V, VNN ).
Using the sam e procedure outlined by McDonald and MacWhinney (1989), parameters were estimated using the STEPIT program of Chandler (1969). An iterative
process was used to obtain a t that minimized the squared deviations between the
obser ved data points and those predicted by the model. STEPIT also provides
goodness-of- t statistics, including the root mean squared deviation ( RMSD) and
the correlation (r 2 ) between the models predictions and the data. W hen the
procedure is allowed to set its own param eter weights, the statistics are a measure
of how well the model ts the data. A high correlation between the model and the
data indicates that it has captured the ordering of the cue strengths.
The results from the ANOVAs revealed that word order was a strong cue for
the E-SLI, CA, and 4;0 year old groups and that animacy was a strong cue for the
ER-SLI group, but these analyses revealed little about the relat ive strength of these
cues for the four groups. The param eter estimates for the two language disordered
and the two normal language control groups are presented in table 2. The param eter
estimates clearly show the strong rst noun preference for the E-SLI group (estimates substantially above 0.5 for NVN, NN V, and VNN conditions). The param eter
estimates for the CA and 4;0 groups correspond to the developm ental trends
Table 2. Parameter estim ates for the three word order condition s (NVN, NNV, VN N) and
the three anim acy condition s (AA, AI, IA) for the expressive SLI, chronolo gically age matched
(CA), young er norm al (4;04;2) and expressive/receptive SLI groups

E -SLI
CA
4;0 4;2
E R-SLI

N VN

NNV

V NN

AA

AI

IA

R M SD

r2

0.93
0.98
0.85
0.62

0.79
0.59
0.55
0.41

0.72
0.55
0.40
0.38

0.62
0.55
0.43
0.47

0.53
0.67
0.59
0.80

0.38
0.32
0.43
0.18

0.02
0.03
0.05
0.04

0.95
0.97
0.92
0.96

128

J. L. Evans and B. MacW hinney

reported in previous studies ( MacDonald and MacWhinney 1989), with a strong

preference for rst noun in the NVN word order, no real preference in the NN V,
and VNN word orders and no preference across any of the animacy conditions. In
contrast, the param eter estimates for ER-SLI children were high in the AI cue
condition and extremely low in the IA cue condition ( i.e. low percentage rst noun
choice), indicating a strong preference for an imacy and no preference for any word
order conditions (estimates 0.300.50). The low RMSD values and the high r 2
values indicate a good t between the model and the data, as well as the ordering
of these cue strengths for all of the groups.
The real advan tage of the MLE technique, however, is its ability to model
ind ivid ual diV erences in performance. The param eter estimates for each SLI child
are presented in table 3 for the NVN and AI cues. Inspection of the individual cue
preferences reveals some interesting patterns. For the majority of the E-SLI children,
word order alone, again, clearly was the cue being used to assign the actor in a
sentence. For the majority of the ER-SLI children, the use of animacy was clearly
the predominant strategy instead.
Spearm an rank order correlation coe Y cients revealed that the correlations
between the param eter estimates for the word order cue and several of the subtests
of receptive language abilities from the CELF-R were signi cant. The correlation
of the level of word order cues with the subtests assessing mem ory for directions
(Oral Directions), mem ory for spoken language ( Recalling Sentences) and knowledge of morphology ( Word Structure) were signi cant (OD, rs= 0.73, p < 0.03; RS,
rs= 0.79, p < 0.01; W S, rs= 0.81, p < 0.01), but not for expressive language abilities
( MLU ), or knowledge of gram matical structures (SS).
Closer inspection of param eter estimates for subjects E-SLI 7, and ER-SLI 1,
3 and 7 also indicate individual diV erences as well. The percentage rst noun choice
Table 3. Parameter estim ates for word order (NVN) and anim acy (AI) condition s, standardized scores for receptive language subtests of the CELF-R (oral directions , word structure,
sentence structure, recalling sentences), and mean length of utterance (M LU ) for children
with expressive and expressive/receptive SLI
N VN

AI

OD

W S

SS

R S

M LU

E -SLI
1
2
3
4
5
6
7

0.99
0.88
0.86
0.99
0.99
0.99
0.97

0.62
0.58
0.64
0.66
0.70
0.65
0.95

9
9
5
9
8
10
9

6
5
5
5
5
5
5

14
9
9
9
7
6
6

6
6
6
4
4
5
5

3.0
4.68
5.38
4.4
4.13
3.68
4.70

E R-SLI
1
2
3
4
5
6
7

0.86
0.62
0.85
0.67
0.68
0.22
0.50

0.71
0.97
0.87
0.99
0.99
0.77
0.48

7
4
6
5
5
4
4

3
4
3
4
3
3
3

6
9
6
6
5
4
5

3
3
3
3
3
3
3

2.47
3.07
3.34
3.86
2.7
2.58
2.79

Oral directio ns, word structure, sentence structure, recallin g sentences.

Sentence processing in subgroups of SLI

129

for the three animacy and word order conditions are presented in table 4 for these
children. In cases where animacy was not available as a cue (e.g. AA), subject E-SLI
7 relied on word order and chose the rst noun as the agent ( NVN 67%, NN V
100%, VNN 100%). However, when an imacy cue was available (e.g. AI, IA), this
child chose the animate noun. Thus, in conditions where no cues were available,
E-SLI 7 used a rst noun choice strategy, but when word order and animacy cues
were combined, this appeared to confuse this child, causing him to revert back to
an an imacy strategy.
Interestingly, for the two children in the ER-SLI group with the highest scores
on the Oral Directions subtest of the CELF-R ( ER-SLI 1 and 3), it appeared that
in fam iliar word order conditions ( NVN ), these children were able to use word
order cues, choosing the rst noun, but in less typical ( NN V ) or unfamiliar ( VNN )
word order conditions, these children reverted to the use of an imacy. Finally, the
param eter estimates for the child with the most severe expressive and receptive
de cits ( ER-SLI 7) indicate the absence of any consistent sentence comprehension
strategy, suggesting that this child was possibly guessing.
Discussion
The two questions posed in this study were: (1) Do the processing strategies
employed by children with SLI vary with the severity of receptive language de cits?
and (2) W ith changes in cue con ict levels, do subgroups of children with SLI
diV er in their comprehension strategies? W ith regard to the rst question, this study
shows that children with better receptive language abilities used word order as the
predominant comprehension strategy, whereas children with more severe receptive
de cits relied on animacy cues instead. This nding is consistent with prior research
indicat ing that children w ith more severe receptive de cits rely heavily on semantic
cues during sentence comprehension tasks (van der Lely and Dewart 1986, van der
Lely and Harris 1990).
The E-SLI and ER-SLI groups in this study diV ered not only in the degree of
receptive de cits, but in expressive de cits as well. One might argue that the
obser ved diV erences could be the result of overall severity, not diV erences in
receptive abilities alone. The correlations between the childs use of word order
cues and the receptive language subtests were signi cant, but were not signi cant for
the expressive measures. Analysis of individual diV erences did reveal, however, that
the two children in the ER-SLI group with the highest receptive language scores
were able to use word order cues in the NVN conditions but reverted back to
reliance on animacy cues in the NN V and VNN, and that the ER-SLI child with
Table 4.

Percentage rst noun choice for all three word order (NVN, NNV, VN N) and
anim acy condition s (AA, AI, IA) for a subset of the children
N VN

E -SLI 7
E R-SLI 1
E R-SLI 3
E R-SLI 7

NNV

V NN

AA

AI

IA

AA

AI

IA

AA

AI

IA

67
83
83
50

100
83
100
33

50
83
33
67

100
33
33
50

100
67
83
50

17
33
17
50

100
33
33
40

67
67
83
50

0
33
0
17

130

J. L. Evans and B. MacW hinney

the most severe expressive and receptive de cits exhibited performance that was
predominantly at chance levels. Previous research has demonstrated diV erences in
processing de cits that vary along a continuum of severity of receptive language
de cits ( Evans 1996, Evans et al. 1997). The ndings from this study re ect this
pattern as well.
In a review of processing de cits in SLI, Bishop (1992) has argued that one
would predict that children with SLI will have particular diY culties in contexts
where critical information is transient, or where transient representations must be
held in mem ory while addit ional information is processed. The children in this
study with lower scores on language tests which tap auditory mem ory skills had
more diY culties using word order cues to process sentences. Although auditory
mem ory skills were not assessed directly in this study, recently, researchers have
proposed de cits in working mem ory as a possible account of SLI (Gathercole and
Baddeley 1990). The ndings from this study suggest that de cits in auditory
mem ory might result in the child having problems retaining the sequential order
of the words in a sentence long enough to use syntactic (e.g. word order) information
during comprehension.
W ith respect to the second question of whether the subgroups diV ered across
low versus high cue con ict conditions, the children with E-SLI in this study
diV ered qualitatively from the children with ER-SLI, as well as from their normal
peers. The E-SLI children exhibited a strong rst noun preference across the low
and high cue con ict conditions as did their age-m atched peers. On the other hand,
the children with ER-SLI relied on an imacy cues regardless of competing word
order cues. In the VNN-AA condition, however, the pattern for both groups of
SLI diV ered from the normal 7 year olds performance in interesting ways. The 7
year olds were clearly sensitive to the absence of word order or animacy cues in
this condition, whereas the E-SLI children continued to use a rst noun strategy and
in the absence of word order cues, the ER-SLI chose the last noun heard .
The model of language processing employed in this study is one where the
acquisition of morphosyntactic knowledge is viewed as the result of the interaction
between the pragmatic and semantic demands of the communication interaction
coupled with the competing constrai nts on the speech channel. Speakers must rely
on a small set of forms ( lexical items, word order, morphological markers, and
intonation) to map all possible competing mean ings and speaker intentions on to
the linear speech stream . As Kirchner and Skarak is-Doyle (1983) have suggested,
the child with language disorders must contend with these sam e communicative
task demands, but due to the incomplete acquisition of surface forms, tenuous
control over these forms, and possibly limited processing capacity, the communicative system of these children may be qualitatively diV erent, and children with SLI
may adapt by employing compensatory strategies or alternative means of processing
information. In this study, it is seen that the children with E-SLI appeared to use
an adapt ive strict rst noun strategy regardless of sentence type. This strategy diV ered
from that of 7 year old normal children, who switched to an OSV strategy in the
NN V context and who appeared to be thrown by the absence of cues in the
VNN-AA context. On the other hand, the ER-SLI children used semantic knowledge as opposed to syntactic information, a pattern that diV ered not only from the
CA matches but from the younger normal language children as well.
Leonard (1995) has suggested that it is not the gram matical knowledge that is
impaired in these children, but their ability to process linguistic information. For a

Sentence processing in subgroups of SLI

131

subset of the children in this study, changes in comprehension strategies were

obser ved with increased processing demands from low to high cue competing
contexts. In the low cue competing contexts, these children used word order cues,
but in the high cue con ict conditions, these children switched to using animacy
cues. This suggests, that for at least some children with SLI, when processing
demands do not exceed their available resources they may be able to use gram matical
information such as word order to interpret sentences correctly, but when processing
demands exceed resources in high cue con ict conditions, they may switch and rely
on earlier developm ental strategies (e.g. whole world knowledge) instead.
The ndings from this study have several important clinical assessment implications. First, the ndings from this study indicate that children with SLI are not
merely delayed in sentence comprehension abilities, but that by the age of 7 have
developed adapt ive strategies to processing language in real time. The within-group
diV erences in sentence comprehension styles observed in this study also indicate
that SLI is not an homogeneous population but one that appears to vary with
severity of receptive language abilities. Detailed analysis of novel sentence forms
(e.g. VNN-AA) revealed that the E-SLI groups comprehension strategies diV ered
qualitatively from those of age-m atched peers by making a more rigid use of the
word order strategy. If the null cue condition had not been included, the E-SLI
groups sentence comprehension strategies would have appeared unimpaired. This
suggests that for children with good receptive language skills, assessment of comprehension skills using highly fam iliar (e.g. SVO) sentence forms may indicate the
apparent absence of comprehension de cits, yet these children may still be experiencing comprehension diY culties in the real-time language processing of novel or
unfamiliar sentence forms.
A second important clinical implication was the correlation between childrens
use of word order cues and their performance on standardized tests requiring good
auditory mem ory skills. For SLI children with poor receptive language abilities,
impairments in auditory working mem ory may prevent them from attend ing to word
order cues, cues that require the child to hold in auditory mem ory the serial order
of the components of the sentence in real time while simultaneously processing the
semantic content of the sentence. For these children with ER-SLI, clinical interventions are needed which focus on developing comprehension strategies that use
word order as opposed to whole world information. In contrast, for children with
E-SLI, clinical interventions may be needed that target word order patterns where
an arbitrary rst noun strategy will result in comprehension errors, speci cally in
those instances where cues are placed in direct competition (e.g. OSV form such
as the homework the 5th graders will complete is in the yellow folder).
The results from this study provide preliminary support for a processing based
account of de cits in subgroups of children with SLI, indicat ing that in contexts
where processing demands are low, E-SLI children may appear unimpaired but in
contexts where cues are in con ict or where sentence forms are novel, their de cits
become evident when compared to typically developing children. In contrast,
semantic whole world knowledge-based strategies characterize the language processing skills of children with more severe receptive language de cits regardless of
changes in language processing demands. To further clarify these issues, however,
future research is needed to address not only the relat ionship between possible
processing capacity limitations and comprehension de cits in subgroups of SLI,
but also to investigate the possible role of auditory mem ory de cits on SLI childrens
comprehension processing abilities.

132

J. L. Evans and B. MacW hinney

Acknowledgem ents

This work is supported by Grant 5K08-D C900068 (Computational Models of

Language Disorders) from the National Institute of Deafness and Other
Com munication Disorders National Institutes of Health, to Dr Evans. Special
thanks is extended to all the children who participated in this study.

References
A dams, C., 1990, Syntact ic com prehension in children w ith expressive language im pairm ent. British
Journal of D isord ers of Com munica tion, 25 , 149 171.
A ram , D . and N ation, J., 1975, Patterns of language behavior in children with developm ental language
disorde rs. Journal of Speech and H earing R esearch, 18 , 229 241.
Bates, E . and M ac W hinney, B., 1987, Com pet ition variation and language learn ing. In B. M acW hinney
( E d.) M echanism s of language acquisition ( H illsda le N J: Lawrence E rlbaum ).
Bates, E . and M ac W hinney, B., 1989, Functionalism and the com petitio n m odel. In B. M acW hinney
and E . Bates ( E ds) The Cross-L inguistic Stud y of Sentence Processing (C ambridge: Cambridge U niversity
Press), pp. 1 73.
Bates, E ., M ac W hinney, B., Caselli, C., D evescovi, A ., N atale, F. and V enza, V ., 1984, A crosslinguistic study of the developm ent of sentence interpr etation strategies. Child D evelopment,
55 , 341 354.
Bever, T. G ., 1970, The cognit ive basis for linguistic structures. In J. R . H ayes ( E d.) Cognition and the
d evelopment of language ( N ew York: W iley).
Bishop, D . V . M ., 1979, Com prehensio n in developm ental language disorde rs. D evelopmental M ed icine
and Child N eurolog y, 21 , 225 238.
Bishop, D . V . M ., 1982, Com prehensio n of spoken written and signed sentences in childho od language
disorde rs. Journal of Child Psychia try, 23 , 1 20.
Bishop, D . V . M ., 1992, The underlyin g nature of speci c language im pairm ent. Journal of Child
Psycholog y and Psychia try and Allied D isciplines, 33 , 3 66.
Burgemeister, B., Blum, L. and Lorge, I., 1972, Colum bia M ental M aturity Scale ( N ew York: H arcourt
Brace Jovanovich).
Chand ler, J., 1969, Subrout ine STE PIT nds loca l m in im a of a sm ooth function of several parameters.
Behavioral Science, 14 , 81 82.
Cohen, J., M ac W hinney, B., Flatt, M . and Provost, J., 1993, PsySco pe. A n interactive graphical
system for design ing and controlling experim ents in the Psychology Laboratory using M acintosh
com puters. Behavioral R esearch M ethod s Instrum entation and Com puters, 25 , 257 271.
Craig, H . and E vans, J., 1989, Turn exchange characteristics of SLI children s sim ultaneous and non
sim ultaneous speech. Journal of Speech and H earing R esearch, 54 , 334 347.
Craig, H . and E vans, J., 1993, Pragmatics and SLI: W ithin-g roup variation in discourse behaviors.
Journal of Speech and H earing R esearch, 36 , 777 789.
E vans, J., 1996, E xpressive and expressive/recep tive SLI subgroups: interaction betw een discourse
constraints and syntact ic produ ctiv ity. Journal of Speech and H earing R esearch, 39 , 655 660.
E vans, J. and Craig, H ., 1991, SLI Subgroups: Similarities and D iVerence in Turn-Taking Behaviors. A merican
Speech and H earing A ssociation A nnual Convention: A tlanta G A.
E vans, J. and M ac W hinney , B., 1995, M axim um likelih ood m odels: syntactic processing skills in
children with expressive and expressive-recept ive language im pairm ents. Symposium on
R esearch in Child Language D isorders , M adison W I.
E vans, J. and M ac W hinney, B., 1996, R elationship betw een processing dem ands and sentence
com prehension strategies : subgroup diV erences in SLI. Symposium on R esearch in Child
Language D isorders , M adison W I.
E vans, J., K ass, R . and V iele, K ., 1997, R esponse t im e and verbal com plexity: Stochast ic m odels of
indiv idu al diV erences in children with SLI. Journal of Speech and H earing R esearch, 40 , 754 764.
Fudala, J. and R eynolds, W ., 1986, Arizona Articulatory Pro ciency Scale ( Los A ngeles: Western
Psychologica l Ser vices).
G atherc ole, S. and Baddeley, A ., 1990, Phonological m emory de cits in language disorde red
children : is there a causal connec tions? Journal of M emory and Language, 29 , 336 360.

Sentence processing in subgroups of SLI

133

John ston, J., 1995, Cognitiv e abilities of children w ith language im pairm ents. In R . Watkins and M .
R ice ( E ds) Speci c language impairm ents in child ren ( Baltim ore: Paul H . Brookes), pp. 91 106.
K ail, R . and Leonard, L., 1986, Word- nding abilities in language-impaired children . ASH A
M onogra phs, 25 , 1 56.
K irchner, D . and Skarakis-Doyle, B., 1983, A discourse approach to language disorde rs: invest igating
com plex sentence produ ct ion. In T. G allagher ( E d.) Pragm atics of Language (San D iego, CA:
Singular Publishing G roup), pp. 283 306.
Leonard, L., 1995, Som e problem s facing accounts of m orph ological de cits in children w ith speci c
language im pairm ents. In R . Watk ins and M . R ice ( E ds) Speci c language impairm ents in child ren
( Baltim ore: Paul H . Brookes), pp. 91 106.
Leonard, L., N ippld, M ., K ail, R . and H ale, C., 1983, Pictu re nam ing in language-impaired children .
Journal of Speech and H earing R esearch, 26 , 609 615.
M ac W hinney, B., Pleh, G . and Bates, E ., 1985, The developm ent of sentence interpr etation in
H ungarian. Cognitive Psycholog y, 17 , 178 209.
M cDonald, J. and M ac W hinney, B., 1989, M axim um likelih ood m odels for sentence processing. In
B. M acW hinney and E . Bates ( E ds) The cross-linguistic stud y of sentence processin g (C ambridge:
Cambridge U niversity Press), pp. 397 421.
N elson, K ., 1974, Variations in children s concep ts by age and categor y. Child D evelopment, 45 ,
577 584.
Semel, E ., W iig, E . and Secord, W ., 1987, The Clinical Evaluation of Language Function Revised ( Texas:
The Psychologica l Co./H arcourt Brace Jovanovich Inc.).
Stark, R . and Tallal, P., 1981, Selectio n of children with speci c language de cits. Journal of Speech
and H earing D isord ers, 46 , 114 122.
Stark, R . and Tallal , P., 1988, Language speech and read ing d isord ers in child ren (San D iego: Little
Brown Co.).
van der Lely, H . and D ewart , H ., 1986, Sentenc e com prehension strategies in speci cally language
im paired children . British Journal of D isord ers of Com munica tion, 21 , 291 306.
van der Lely , H . and H arris, M ., 1990, Com prehensio n of reversible sentences in speci cally
language im paired children . Journal of Speech and H earing D isord ers, 55 , 101 117.
von Berger, E ., W ulfeck , B., Bates, E . and Fink, N ., 1993, D evelopmental Change in R eal-tim e Sentence
Processing. Techn ical R eport N o 9303, Project in Cognitiv e N eurodevelopm ent (San D iego, CA:
Cen ter for R esearch in Language, U niversity of Californ ia).
W echsler, D ., 1974, Wechsler Intelligence Scale for Child ren Revised ( N ew York: The Psychology
Corporation).
W olfus, B., M oscovitch, M . and K insbourne, M ., 1980, Subgroups of developm ental language
im pairm ent. Brain and Language, 10 , 152 171.
W ulfeck , B., 1993, A reaction tim e study of gramm aticality judgm ents in children . Journal of Speech
and H earing R esearch, 36 , 1208 1215.

Appendix
Nouns: animate

Nouns: inanimate

Verbs

cow
camel
cat
horse
pig
lamb
monkey
lion
zebra

cup
spoon
chair
comb
ball
block

kisses
hugs
pets
chases
wants

134

Sentence processing in subroups of SLI

Exam ples of sentence types

NVN AA
NVN IA
NVN AI
NN V AA
NN V IA
NN V AI
VNN AA
VNN IA
VNN AI

The zebra pets the lamb

The block hugs the lion
The pig kisses the cup
The pig the cow pets
The ball the horse hugs
The cow the block chases
kisses the pig the cow
Hugs the comb the monkey
wants the cow the cup