Post on 25-Feb-2021
Cognition, 51 (1994) 29-72 OOlO-0277/94/$07.00 0 1994 - Elsevier Science B.V. All rights reserved.
29
Canonical linking rules: forward versus reverse linking in normally developing and specifically language-impaired children
Heather K.J. van der Lely Department of Psychology, Birkbeck College, University of London, Malet Street, London WClE
7HX, UK
Received September 2, 1991, final version accepted June 17, 1993
Abstract
Canonical linking rules for mapping thematic roles with syntactic functions were studied. Three experiments were undertaken to investigate the nature of productive forward linking (from semantics to syntax) and productive reverse linking (from syntax to semantics). I proposed that reverse linking, in contrast to forward linking, requires more detailed specification of the syntactic structure; that is, a syntactic representation which specifies each particular syntactic frame and all the argument positions within that frame. Six specifically language-impaired children (aged 6;l to 9;6) were matched on language abilities to 17 younger, normally developing children (language age 3;l to 6;6). In Experiment 1 -forward linking - the children were shown the meaning of a novel verb and had to describe the event using the novel verb. Experiment 2 - a comprehension task - required acting out sentences containing the newly learned verbs. In Experiment 3 - reverse linking - the children were told a sentence with a novel verb and had to act out its meaning,
I would like to thank Dorothy Bishop, Harald Clahsen, Richard Ingham, Rita Manzini, and Neil
Smith and in particular Steven Pinker and two anonymous reviewers for their comments and
interesting discussions on a previous draft of this paper. I am also grateful to the staff of Glebe Infant
school and Moorlands day nursery and to the speech and language therapists who assisted me in
finding appropriate subjects for this investigation.
The basis of this research was undertaken in partial fulfilment for a PhD degree at the University of
London, and was supported by a competition studentship from the Economic and Social Research
Council. I thank Margaret Harris and David Howard who supervised this work. A version of this
paper was presented at the 17th annual Boston University Conference on Language Development,
October 1992, Boston: Mass. The preparation of this paper was supported by a post-doctoral fellowship from the British Academy.
SSDZ 0010-0277(93)00577-T
30 H.K.J. van der Lely I Cognition 51 (1994) 29-72
assigning thematic roles on the basis of the syntactic frame. Group and individual analysis generally revealed no significant differences between the specifically language-impaired children and the language age control children in Experiments 1 and 2, but a significant difference was found for Experiment 3. The normally developing children showed a good use of productive forward and reverse linking. The specifically language-impaired children demonstrated good productive forward linking but were significantly worse at reverse linking. An interpretation of the data, showing differences in the syntactic representation required for forward versus reverse linking, can account for the findings. I propose that a deficit in the area of “government” or “locality” which underlies c-selection and specifies the syntactic relationship between constituents can account for the data from this study and the data from previous investigations of specifically language-impaired children.
Introduction
Linguists and psychologists have recently shown increased interest in a group
of children who are “specifically language impaired” (SLI). SLI children are
characterized by severe problems in the development of language comprehension
and expression, but are normal in non-linguistic cognitive and motor develop-
ment, hearing, and emotional behaviour. The nature of the linguistic disorder and
the underlying deficit in these children can provide insights into the mechanisms
of language acquisition which could not be possible from investigations into
children developing normally.
This study investigates the “canonical linking rules” which have been hypoth-
esized to play an important role in the learning of verb-argument structure (e.g.
Bowerman, 1990; Pinker, 1989). Canonical linking rules refer to the regular
relationship found in the majority of languages between thematic roles, such as
agent and patient, and syntactic functions such as subject and direct object. For
example, if a child knows that a verb involves an agent and a patient, she can
infer that those arguments are expressed as the grammatical subject and object,
respectively. The canonical or default mapping between particular thematic roles
and particular syntactic functions may apply in the absence of information to the
contrary.
Canonical linking rules can operate bidirectionally; from semantics to syntax
(forward linking) and from syntax to semantic (reverse linking). The study aims
to provide insight into the nature of the forward versus reverse linking pro-
cedures, focusing on the syntactic nature of the linking rules when they are used
in a forward as opposed to a reverse direction. By investigating linking in a
population of SLI children these two mapping mechanisms can potentially be
distinguished. Do individual SLI children differentially use forward linking and
H.K.J. van der Lely I Cognition 51 (1994) 29-72 31
reverse linking and what are the characteristics which may explain this‘differential
use?
Canonical linking rules
In this section I shall first discuss linking rules and how they may facilitate
language acquisition. I shall then consider the possible syntactic nature of the
linking rules when they are used in a forward versus reverse direction.
Linking rules are the regular ways in which thematic roles (such as agent and
patient) are mapped onto syntactic functions (e.g., subject, object) (Bowerman,
1982; Bresnan, 1982; Jackendoff, 1972; Pinker, 1984). Particular thematic roles
are associated with a verb and form part of its semantic structure. The linking
rules, or correspondences between thematic roles and syntactic functions (e.g.,
agent-subject), may be innate (Grimshaw, 1987; Pinker, 1984, 1989) or learned
early in the process of language acquisition (Bowerman, 1990; Foley & Van Valin,
1984). In either case the linking regularities facilitate language acquisition.
There has been much discussion about what constitutes the “canonical” or
default mapping. The earlier approach characterized canonical linking in terms of
two interrelated hierarchies: one for thematic roles and one for syntactic functions
(Bresnan, 1982; Fillmore, 1968; Jackendoff, 1972; Pinker, 1984). The thematic
role of a predicate, which was the highest on the hierarchy (e.g., agent), was
linked to the highest available syntactic function (e.g., subject). The next highest
thematic role (e.g., theme/patient) was linked to the next available highest
syntactic function (e.g., object) and so on. The more recent approach does not
use the notion of thematic roles being in a fixed list but defines them as argument
positions within the semantic representation of a verb’s meaning (Dowty, 1991;
Jackendoff, 1987; Levin, 1985; Pinker, 1989; Rappaport & Levin, 1988). The
theory makes explicit the properties of the event that the verb refers to; for
example, agentive, causal and temporal. Traditional thematic labels like “agent”
and “patient” can serve as mnemonics for some of the argument positions
(Gropen, Pinker, Hollander, & Goldberg, 1991b) and will be used as such in this
paper. For example, “agent” is defined as the first argument of CAUSE and
ACT, “patient” is the second argument of CAUSE and ACT, “theme” is the first
argument of GO and BE, “location” is the second argument of BE, and goal is
the second argument of TO, etc. (e.g., Pinker, 1989). For each thematic role (or
more accurately, each argument position) a linking rule maps it to a syntactic
function; for example, agent is linked to subject, patient is linked to object,
theme is linked to subject if that function has not already been assigned,
otherwise to object, and location to oblique object.
The use of a verb in a different syntactic frame causes differences in meaning.
For example, the passive is represented as a circumstantial state (BE) in which
32 H.K.J. van der Lely I Cognition 51 (1994) 29-72
the circumstance of the theme is the event or state corresponding to the active
verb form (Pinker, 1989). Therefore, the semantic representation and the
argument structure differ for corresponding active and passive verbs. If the child
can identify the appropriate semantic properties and thus the arguments of a
predicate, in the absence of any conflicting information, she can apply the
particular linking rule to map each argument onto the appropriate syntactic
function (Pinker, 1989). This paper will be based on this more recent theory of
canonical linking. All the syntactic frames in which a verb can occur reflect the
semantic representation and canonical linking. However, the term “non-canoni-
cal” linking will be used in this paper to refer to any mapping that does not
represent the default semantic-syntactic relationship: for example, an agent (first
argument of CAUSE) mapped onto the direct object in a transitive active
sentence would be a non-canonical mapping.
Linking rules may facilitate language acquisition in several ways. Pinker
(1984), Berwick (1985) and others have hypothesized that children use knowledge
of thematic roles and the linking rules to “bootstrap” themselves into syntax; that
is, “semantic bootstrapping” (Pinker, 1984), which facilitates the development of
phrase structure rules.
The semantic cues may also be used by children to express the meaning of a
new predicate in an utterance without previously hearing it in a sentence. For
example, when learning a new verb the child can determine the thematic roles of
the noun arguments on a non-linguistic basis and then, with the use of the
canonical linking rules, map the arguments onto the default syntactic functions.
This may be seen as “productive” learning of verb-argument structure (Pinker,
Lebeaux, & Frost, 1987). Previous research has shown that young children of 3-5
years of age are sensitive to the semantic-thematic roles associated with a
predicate and will canonically link these roles onto syntactic functions productive-
ly (Gropen, Pinker, Hollander, Goldberg, & Wilson, 1989; Gropen et al., 1991a;
Pinker et al., 1987). The use of semantic cues and forward linking will be
investigated in this study.
Another way in which linking rules may facilitate language acquisition is when
syntactic cues are used in conjunction with the linking rules to help learn the
meanings of verbs (Gleitman, 1989, 1990; Golinkoff, Hirsh-Pasek, Cauley, &
Gordon, 1987; Landau & Gleitman, 1985; Naigles, 1990; Naigles, Gleitman, &
Gleitman, 1992). The syntactic cues may be used in more than one way. First, by
attending to the verb’s subcategorization frame the child can predict aspects of its
meaning (Landau & Gleitman, 1985). For example, Naigles and Kako (in press)
conclude that the transitive frame specifies a sense that one character is affecting
another.
Landau and Gleitman (1985) have coined the term “syntactic bootstrapping”,
which refers to attending to sets of subcategorization frames associated with a
verb stem. This procedure may facilitate learning verbs by narrowing down the
possible meanings shared by all of the stem’s syntactic entries. It is a contentious
H.K.J. van der Lely I Cognition 51 (1994) 29-72 33
issue as to the extent syntactic bootstrapping can facilitate and is needed to learn
a verb’s meaning (see Pinker, 1993). However, hearing a verb in one syntactic
frame may give some information about the verb’s semantic structure (e.g., the
number of arguments and type of arguments), although it cannot reveal anything
about the core meaning content of the verb (e.g., the movement defined as
“hitting”). The use of syntactic cues to infer thematic roles associated with a verb
has been referred to as reverse linking (Pinker, 1989) (i.e., syntactic to semantic
mapping). Pinker has suggested that reverse linking may be particularly important
in learning verbs which undergo certain syntactic alternations such as the passive
and locative. This paper focuses on the semantic information that may be derived
from hearing a verb in one syntactic frame; that is, reverse linking.
The nature of the syntactic representation used with the linking rules is not
clear. However, I shall attempt to specify the syntactic representation which may
be required for linking rules to be used in a forward and reverse direction. In
forward linking in which semantic cues are used to map from thematic roles to
syntactic structures, Gleitman (1989) has suggested that a “one-to-one” mapping
may be used. Thus, one thematic role (e.g., agent) may be mapped onto one
noun phrase (NP) (e.g., first NP) without recourse to the other thematic roles and
syntactic functions associated with the predicate. A “basic” syntactic representa-
tion of the phrase structure may be sufficient to facilitate the appropriate
canonical linking relationship. For example, when the child is presented with a
new verb, providing the child has already worked out that the agent of a causative
event corresponds to the first NP in the sentence and the patient corresponds to
the second NP, etc., she may not require any further syntactic knowledge of the
structural syntactic relationships between the constituents. However, a syntactic
representation of the linear order of an NP in relation to the verb may not be
sufficient if the syntactic cues are to be used to infer particular thematic roles.
This is because of the many-to-few mapping between thematic roles and syntactic
functions; for example, if something is the agent of a transitive active verb it is
sure to be the subject in nominative accusative languages, but if something is the
subject of a transitive active verb it need not be an agent (it could be a recipient,
as in “receive”; a location, as in “contain”; and so on).’ Pinker (1989) indicates
that it is not completely clear how reverse linking should work at all given that
there are many sets of thematic roles that are consistent with many surface
frames. However, he suggests that reverse linking would have to involve
comparisons between the surface form of a sentence and some kind of schema or
core for common linking patterns abstracted from sets of individually acquired
verbs.
My proposal is that, in order to activate the appropriate schema and to
constrain the possibilities of which thematic role to map onto the first NP (in the
‘I am grateful to Steven Pinker for reminding me of this.
34 H.K.J. van der Lely I Cognition 51 (1994) 29-72
absence of any semantic information), the child needs to identify each and every
argument position within a particular syntactic frame. The structural relationships
between the constituents in a syntactic frame enable each NP to be uniquely
identified; for example, subject-NP versus object-NP. Therefore, the child
requires a syntactic representation of each particular syntactic frame (e.g.,
transitive active frame) in which the NPs are differentially identified. Merely
identifying one NP and the verb would not provide sufficient syntactic cues for the
child to activate the appropriate linking schema and constrain the many possible
mappings of thematic roles to the NP. A non-specified NP could as easily be
linked to a patient/theme as to an agent thematic role. I propose that the more
detailed syntactic representation of the syntactic frame and the argument
positions within that frame are needed for reverse linking but not necessarily
forward linking. This structural syntactic representation can be linguistically
characterized by the c (categorical)-selectional properties of a verb (Chomsky,
1981) and syntactic government which differentially “marks” the NPs in a
sentence by specifying their positions.
Specifically language-impaired children
The majority of SLI children previously investigated have been found to have
an expressive deficit which is most pronounced in the phonological and morpho-
grammatical aspects of language (Bishop, 1992; Clahsen, 1989; Gopnik, 1990;
Leonard, 1982, 1989, 199,l). Recent studies have highlighted consistent deficits in
SLI children’s comprehension of sentences (Bishop, 1979, 1982; Connell, 1986;
Precious & Conti-Ramsden, 1988; van der Lely, 1990; van der Lely & Dewart,
1986; van der Lely & Harris, 1990). The most significant findings to be revealed
by these data are the following:
(1) Children previously classified as suffering from an expressive disorder as
well as those with a receptive language disorder are usually impaired in their
comprehension of sentences (Bishop, 1979).
(2) The deficits have been found in interpreting complex sentences, such as
passive and embedded sentences (Bishop, 1982; van der Lely & Harris, 1990) and
also simple active transitive sentences (van der Lely and Dewart, 1986; van der
Lely and Harris, 1990).
(3) Particular difficulties have been found in comprehending semantically
reversible sentences in which word order is particularly important in signalling the
relationship between the syntactic and thematic roles. The SLI children’s
performance in these studies has been characterized by a high proportion of
“word order errors”; that is, errors in sentences where correct assignment of
H.K.J. van der Lely I Cognition 51 (1994) 29-72 35
thematic roles to syntactic functions such as subject and object is reversed,
(Bishop, 1982; van der Lely & Dewart, 1986; van der Lely & Harris, 1990).
(4) Finally, SLI children’s sentence comprehension deficits have been found in
a variety of sentences (transitive, locative and dative) (van der Lely & Harris,
1990). These studies have all included control groups of children matched on
receptive language abilities of single word vocabulary. In some studies the SLI
children were also matched to the control children on production of morphologi-
cal forms and naming vocabulary. Thus, the SLI children had a disproportionate
language deficit in the comprehension of reversible sentences. However, not
everyone agrees that SLI children do have problems assigning thematic roles.
Gopnik and Crago (1991) did not find that their group of SLI children (who were
characterized by a genetic familial aggregation) were impaired in this area of
language. This may be because of the different methodology used by Gopnik and
Crago, or because there is heterogeneity in the population. That is, subgroups of
SLI children exist which have different linguistic characteristics. The extent to
which these subgroups are related is far from clear. This study investigates the
subgroup of SLI children who show grammatical deficits in expression and
comprehension of language.
There have been few studies which have explored the processes involved in
sentence comprehension in SLI children. If we are to go beyond merely
describing their language difficulties and provide further insight into their disorder
it is evident that further investigations are needed into the processes involved in
sentence comprehension. This study aimed to establish if SLI children’s superficial
difficulty with word order in comprehending reversible sentences could be
attributed to faulty “mapping mechanisms” for learning verb-argument structure.
In this study, three experiments were conducted to investigate the use and
nature of semantic and syntactic cues and the canonical linking rules to learn the
mapping properties of novel verbs in normally developing and SLI children.
Experiment 1 investigated the productive use of forward linking (semantic to
syntactic mapping). Experiment 2 tested the children’s comprehension of the
newly learned verbs in Experiment 1, and Experiment 3 investigated the
productive use of reverse linking (syntactic to semantic mapping). The experi-
ments investigated a subgroup of six SLI children with a disproportionate
grammatical impairment in the comprehension and expression of language in
relation to their other language abilities, such as single word vocabulary. Because
of the possibility of heterogeneity in groups of SLI children, the study investigated
individual subjects’ patterns of responses as well as group data.
The questions to be addressed were: (a) Do SLI children’s responses demon-
strate the productive use of forward linking to map semantic-thematic roles onto
syntactic structures? (b) Do SLI children comprehend the newly learned novel
verbs when they are presented in active and passive sentences? (c) Do SLI
36 H.K.J. van der Lely I Cognition 51 (1994) 29-72
children demonstrate the productive use of reverse canonical linking to map
syntactic functions onto thematic roles for novel verbs presented in sentences? (d)
Does the SLI children’s ability to perform these tasks differ from that of children
matched on some measures of language abilities?
It was predicted that (a) if the SLI children’s deficit is with the linking rules
themselves (either because the lacked the appropriate (innate) semantic-syntactic
knowledge or they had not learned the canonical relationship between semantic
and syntactic roles) then they should be impaired in all three tasks; (b) if the SLI
children’s deficit lies in the use of semantic or syntactic cues to facilitate learning
the lexical properties of verbs then they should be impaired in Experiment 1 or 3,
respectively; (c) if the SLI children’s deficit lies in recalling or using a newly
learned representation of verbs, then their performance would be impaired in
Experiment 2, but not in Experiment 1 or 3. Such a deficit could be indicative of a
more general learning problem for linguistic knowledge.
EXPERIMENT 1: FORWARD LINKING
In Experiment 1 the children were shown the meaning of a novel verb using an
acting out procedure; for example, the experimenter said “This is voozing”, as
she made a toy girl jump up and down on a toy boy’s back. The children were
then asked to say what was happening using the new verb to see if they would say
“The girl is voozing the boy”, or “The boy is voozing the girl”. The experiment
was designed to see if the children identified thematic roles from two types of
events. One corresponded to a causative transitive event; that is, an event
involving an agent and a patient/theme. The second was a causative event
involving agent and patient/theme thematic roles and also a goal thematic role;
for example, a car was made to tap a lorry sideways to a train. For the purpose of
this paper I shall refer to this second event as the “locative” event or scene. As in
some previous studies (Pinker et al., 1987; Gropen et al., 1989, 1991b) I used
here invented novel verbs, for which there was no exact equivalent verb in
English. This would prevent prior semantic and syntactic knowledge from
facilitating performance in the experimental task. In comprehension tasks it is
possible that linking rules could be used in the reverse direction based on the
syntactic cues provided by the sentence. Therefore, to avoid this possibility,
expressive language was used as a measure of productive forward linking. Probe
questions, used to elicit the responses, were syntactically neutral in that no
specific role could be attached to a particular toy based on the syntactic structure
of the question; for example, “Tell me about the boy”. There were 10 verbs and
three probe questions for each verb.
H.K.J. van der Lely I Cognition 51 (1994) 29-72 37
Method
Subject details
Two subject groups participated in the experiments: a group of SLI children
and a group of younger children matched on receptive and expressive language
abilities.
Specifically language-impaired children
There were six SLI children - five boys and one girl - selected from a larger
group that had been under investigation for the previous 2 years. The children
were characterized by a disproportionate deficit in their grammatical abilities in
the expression and comprehension of language.
The original selection criterion was as follows. Firstly, all the children had been
identified by speech and language pathologists and educational psychologists as
having persistent difficulties with language comprehension; that is, their scores fell
more than 1.5 s.d. below the expected normal range of abilities on some
standardized tests - and they were undergoing their education in language units.*
The selection of the SLI children was based on the set of criteria proposed by
Stark and Tallal (1981). To provide a uniform measure of language abilities and a
criterion for inclusion in the group, two tests of comprehension and two of
expressive language abilities were administered. Only children who had a mean
comprehension age of at least 6 months lower than their chronological or mental
age (whichever was the lower) and a mean expressive language of at least 12
months below their chronological or mental age were included in the group (Stark
& Tallal, 1981).
The non-verbal abilities of all the SLI children had been assessed by education-
al psychologists using standardized IQ tests (e.g., WISC-R) and fell within the
normal range. In addition, two tests of non-verbal abilities were administered to
provide a mental age score on which to base the children’s language age. The
Block design from the British Abilities Scale (Elliott, Murray, & Pearson, 1978)
and the Draw-A-Man test (Harris, 1963) were used. Only children who attained
an IQ of above 85 on both tests were included. This original group’s language
ages ranged from 2;8 to 5;5 (yearqmonths), and their chronological ages ranged
from 4;5 to 7;lO. The raw scores from the tests for the original selection of the
group have been previously reported (van der Lely & Harris, 1990) and,
therefore, further details will not be given here. All the children met the criterion
of normal development with respect to hearing, neurological development, socio-
emotional behaviour and development as described by Stark and Tallal (1981).
*That is, a small class, specializing in language remediation
38 H.K.J. van der Lely i Cognition 51 (1994) 29-72
The six SLI children in this study were selected to be representative of the
original group. Therefore the severity of their original language impairment and
their rate of progress on standardized tests of language over the previous 2 years
varied. The SLI children were reassessed on the same four standardized tests. The
choice of these tests was determined by their previous use for research purposes
and reliability in identifying SLI children (Bishop & Adams, 1989; van der Lely &
Harris, 1990), rather than any adherence to theoretical notions on which the tests
were based. The two tests of comprehension were the Reynell Developmental
Language Scales (Reynell, 1977) (a general test of understanding) and the British
Picture Vocabulary scale (Dunn, Dunn, Whetton, & Pintillie, 1982) (a more
specific test of comprehension of single word vocabulary). The two standardized
measures of expressive language were the Grammatical Closure Subtest from the
Illinois Test of Psycholinguistic Abilities (ITPA) (Kirk, McCarthy, & Kirk, 1968)
(a test of grammatical morphology) and the Expressive Vocabulary from the
British Ability Scales (Elliott et al., 1978). Characteristically, the SLI children
performed better on the vocabulary tests than on the test of morphology. To
provide the reader with an estimate of their language abilities, an overall language
age was calculated from the four language tests. However, this language age is
merely a guide and caution is expressed in taking this measure too literally as the
SLI children perform relatively well on some tests but show severe impairments
on other tests which tap different language abilities. Their language ages ranged
from 3;7 to 6;5. Their chronological ages ranged from 6;l to 9;6. Table 1 provides
a summary of subject details; further details of the language scores for each of the
children on the four tests can be found in Appendix A.
As is characteristic of SLI children generally, some of the SLI children in this
study performed within normal limits on some of the tests. However, the
language impairment of all six SLI children was of such severity they were still
undergoing their education in language units.
Language age control group
Each child in the SLI group was individually matched to three language control
children on the basis of individual scores from three standardized language tests.
Table 1. Subject details for the SLI children and LA controli
Subject group Chronological age Language age
Mean Range Mean Range
SLI children 7;2 6;1-9;6 5;4 3;4-6;5 (N=6) LA controls 5;2 3;4-6;6 5;3 3;1.5-6;6 (N = 17)
H.K.J. van der Lely I Cognition 51 (1994) 29-72 39
The tests were the British Picture Vocabulary Scale (BPVS), the Grammatical
Closure Subtest from the ITPA, and the Expressive Vocabulary from the British
Ability Scales (BAS). The tests were administered to 55 children whose
chronological ages approximated the language ages of the SLI children. The
control children came from an infant school and private day nursery in a rural
city. Seventeen matched control children (nine boys and eight girls) whose raw
scores on at least two of the three language tests were the nearest to those of their
assigned SLI child were selected from the sample. Two of the SLI children, OC
and SJ, who had identical overall language ages, shared one control child. The
scores of all the language age (LA) controls fell within the normal limits on the
three language tests. Analysis revealed no significant differences between the LA
controls and the SLI children on the raw scores of the three standardized tests of
language abilities (for the BPVS, 421) = 0.15, for the Grammatical Closure,
ITPA subtest, 421) = 0.35, and for the Expressive Vocabulary, BAS, 421) =
0.51). Therefore the LA control group were matched to the SLI children on all
three tests of language ability. The mean language ages for the LA controls
ranged from 3;1.5 to 6;6. A summary of the LA control subject details can also be
found in Table 1, and further details of the raw scores for each test in Appendix
A. Additional information of the matching procedure can be found in van der
Lely and Howard (1993).
Design and materials
Experiment 1 was designed to test the productive use of canonical forward
linking, mapping semantic roles to syntactic functions. Subject group (SLI
children and LA controls) constituted a between-subject variable in the experi-
ment. There was one within-subject variable; that is, the semantic structure of the
scene (transitive or locative).
Ten verbs were invented for which there was no exact equivalent meaning in
English. Five of the verbs corresponded to transitive activities. That is, the
activity involved two participants: one which could be construed semantically as
an agent; and the other a patient/theme thematic role. The remaining five verbs
corresponded to the so-called locative activities. These activities involved the
movement of two toys (an agent and patient/theme) towards a toy which could be
construed as the goal of the activity. For example, a car (agent) moving sideways
tapping a train (patient/theme) sideways to a lorry (goal). The phonological form
of the 10 verbs, which corresponded to a CVC sequence, and a description of
their meanings, is summarized in Table 2. All the verbs were potentially
semantically reversible, in that either toy participating in the activity could take
on either thematic role. Selection of the toys was carefully controlled to maintain
40 H.K.J. van der Lely I Cognition 51 (1994) 29-72
Table 2. Experiment 1. Forward linking: list of novel verbs and a description of their meaning
Transitives
1. Mooking
2. Tiving
3. Kalling
4. Geeping
5. Voozing
X slides down the back of Y.
X bounces Y on his head, holding Y upside down.
X pushes Y over, with front of vehicle.
X pushes Y, head to head while lying on floor.
X jumps up and down on Y’s back.
Locatives
6. Mipping
7. Zecking
8. Dassing
9. Fiming
10. Bozing
X backs into Y and pushes it backwards to 2.
X taps Y sideways to Z.
X carries Y on her head to Z and puts Y on Z’s head.
X carries Y sideways across her back
and bounces with Y to Z.
X pulls Y to Z with a rope looped around the wheels of
X and Y, and attaches Y to Z’s wheels.
this semantic neutrality. Therefore, for each scene the selected toys were either
all people or all transport toys (see list below).
Eight small toys were used in the Experiment. Four toys corresponded to
people (man, lady, girl, boy) and four to transport (car, lorry, train, tractor).
Procedure
The children were tested individually in a small room. The experimenter sat
opposite the child at a small table. The eight toys were placed on the table, with
the people and transport toys grouped separately. The child was asked to name
the toys. Following successful naming of the toys, it was explained to the child
that she would be told a “funny” word and then the toys would show her what it
meant. The funny word was attributed to an unusual looking puppet who was
going to watch what was happening and make sure the examiner showed the child
the correct meaning. The puppet was going to see if the child had learnt the word.
The order of presentation of the 10 verbs was randomized, but with a transitive
scene occurring first. The child was told the first verb, which was presented as a
gerund, and asked to repeat it at least twice, with the examiner repeating it at
least three times. An example of the teaching and experimental phase for one of
the verbs is described below. The example involves the verb tiving, which has the
meaning “X bounces Y on his/her head whilst holding Y upside down”.
Demonstration (a): the toy man is made to walk towards the toy woman. The
experimenter said “This is tiving” as she demonstrated the meaning of the verb;
H.K.J. van der Lely I Cognition 51 (1994) 29-72 41
that is, the man was made to hold the woman’s legs and he bounced the woman
up and down on her head.
Demonstration (b): this involved the same toys as the first demonstration,
except that the thematic role the toys were made to take on was reversed.
Therefore the woman was made to take on the agent role and the man the
theme/patient role.
Demonstration (c): two different toys were selected for this demonstration;
that is, the boy was made to perform the agent role and the girl the theme role.
The procedure was otherwise identical to the first demonstration.
Testing phase: the toys in demonstration (c) were used in the testing phase, but
the thematic roles the toys were made to take on were reversed. That is, the girl
was made to tiv the boy. Three identical demonstrations were made in the testing
phase. Each demonstration was accompanied by one of three probe questions
which were presented in the following order: (i) Tell me/the puppet what’s
happening. (ii) Tell melthe puppet about the girl (agent). (iii) Tell melthe puppet
about the boy (patient/theme). If the child responded to the first question with
only a gerund, or did not respond at all, then a further puppet was introduced to
try to make the required response pragmatically more plausible. (This puppet hid
in a bag, as he was very shy, but he wanted to know all about what was
happening!) The session was recorded on a portable tape recorder for later
transcription.
Scoring
Utterances were transcribed for each subject from the audiotape. Appendix C
gives an example of the responses for three SLI children and three LA controls to
one of the novel verbs. The three possible responses to each verb were coded in
the following way:
(a) Use of the novel verb: from the 30 possible responses, each response was
coded as to whether the novel verb or an alternative verb (which partially
described the scene) had been used for the five transitive verbs and five locative
verbs.
(b) The responses to the transitive and locative scenes were analysed separ-
ately and according to whether a novel verb or other verb had been used. The
responses were coded into the type of syntactic frame in which the verb was used.
For the transitive scenes these were: verb-only response (e.g., tiving/VP);
intransitive response (e.g., The boy is tivinglNP VP); transitive response (e.g.,
The boy is tiving the girflNP VP NP). For the locative scenes an additional
category was used in which the location-goal was also encoded in the response
42 H. K.J. van der Lely I Cognition 51 (1994) 29-72
(e.g., The car bozes the lorry to the train/NP VP NP PP). This will be referred to
as a locative response.
(c) Canonical linking: responses in which the novel verb had been used for
each question were analysed to establish the number of thematic roles that
conformed to a “canonical” or “non-canonical” mapping of semantic-thematic
roles to syntactic functions. Canonical linking, as discussed in the Introduction,
was based on the definition put forward by Levin (1985), Dowty (1991),
Jackendoff (1987), and Pinker (1989). So, for example, if the agent of the
causative event was mapped to the subject-first NP it was counted as canonical,
but if it was mapped to the object-second NP it was counted as non-canonical.
Further description of canonical linking is given below.
There were three possible argument structure types that were appropriate to
describe the transitive scenes: (1) unergative intransitive; (2) transitive active; (3)
transitive passive. An additional argument structure type was possible for the
locative scenes: (4) a transitive with oblique object containing to. Noun phases and preposition phases were coded as canonical on the basis of
word order when an active sentence was used. However, the morphology was
taken into account when a passive utterance was used. The intransitive responses
were scored as canonical if the agent was assigned to the subject. The child
obtained a score of 1 for each canonical intransitive response. The transitive
active responses were scored as canonical if the toy taking the agent role in the
demonstration was assigned to the subject/first noun phrase, and the patient/
theme to the object/second noun phrase. The transitive passive responses were
scored as canonical if the theme was assigned to the subject and the agent to the
oblique object (preposition phrase) (after Pinker et al., 1987). If a response was
ambiguous when both word order and morphology were considered then a strict
criterion was adhered to and the response was scored as non-canonical. An
example of this ambiguity is clearly shown by one of MP’s responses to question
3: The girl is kalling by the boy. On the basis of word order alone this response
would have been scored as canonical. However, on the basis of morphology, the
use of the gerund and the preposition make it ambiguous as to MP’s intention.
The child obtained a score of 2 for each canonical transitive active or passive
response. However, short passive sentences (The girl is mooked) scored only 1.’
The scoring for the transitive with oblique object responses (prepositional
locatives) followed the criterion set out above. In addition, if the goal thematic
role was assigned to the oblique object and was morphologically marked by the
preposition to or another appropriate preposition such as up to, it was taken to be
canonical. For the locative scenes, if all three possible arguments were expressed
‘This conservative criterion was used as it is possible that the short passive responses could have
been adjectival rather than verbal passives. Therefore it cannot be assumed that the second argument is “absorbed” by the passive morphology. (cf. Borer & Wexler. 1987).
H.K.J. van der Lely I Cognition 51 (1994) 29-72 43
and corresponded to canonical linking, the child scored 3. Responses for the
locative scenes which consisted of three noun phrases (e.g., The cur mips the train and the lorry) were coded as transitive utterances, providing the first two noun
phrases conformed to a canonical linking of the agent to the subject and the
theme/patient to one of the following noun phrases. The “goal” thematic role
was not differentiated and therefore did not contribute to the child’s score. Table
3 provides the details from all of the responses containing the novel verbs for the
six SLI children according to the scoring criterion.
Results
Analysis of group data
The initial analysis assessed the frequency of novel verb responses. The
percentage use of the 10 novel verbs was calculated for each child from the three
responses for each verb. A 2 x 2 (Group X Verb type) ANOVA was used to
investigate these data. It can be seen from Table 4 that both groups used the
novel verbs, rather than an alternative known verb, for the majority of their
responses. Analysis revealed no significant effects for group, F( 1,21) = 1.13,
p = .30, verb type, or the interaction. Thus, the groups did not differ in their use
of the novel verb for the transitive or locative scenes. Further analyses were
carried out on the novel verb responses.
The main analysis for forward linking was based on responses to questions 1
and 2 for the transitive and locative scenes. Responses to question 3 were not
included owing to some ambiguity in some of the responses (see scoring above).
From the maximum of 20 novel verb responses, the total number of canonically
expressed thematic roles in the intransitive, transitive and locative syntactic
frames were tallied. A canonical intransitive response scored 1, and a transitive
and locative response 2 and 3, respectively (see scoring section for further
details). It can be seen from this procedure that the total forward linking score
was sensitive to both the mapping relationship and the number of thematic roles
expressed in the sentences. The maximum score for the 10 responses for the
transitive scenes was 20, whereas for the 10 responses for the locative scenes it
was 30. The proportion of canonically expressed thematic roles was calculated for
each child from the maximum possible scores for the two scene types. Thus, a
score of 50% for the transitive scenes would represent 10 thematic roles expressed
canonically, whereas the same proportion (50%) for the locative scenes would
represent 15 thematic roles expressed canonically. The mean proportion scores
for the two subject groups for the transitive and locative scenes can be found in
Table 4. This mean proportion score is a conservative estimate of the use of
forward linking as the proportion was based on the maximum possible score.
44 H.K.J. van der Lely I Cognition 51 (1994) 29-72
Table 3. Experiment 1. Forward linking: coding of the novel verb responses for the six SLI children
Transitive scenes Locative scenes
Response type Response type
Verb Intrans. Trans. Verb Intrans. Trans. Locative
only only
Subject: MP
Question 1 Canonical 2 4
Non-canonical 2 1
Question 2 Canonical 5 4
Non-canonical 1
Question 3 Canonical 1 4
Non-canonical 4 1
Response type: total 1 14 15
Subject: OC
Question 1 Canonical 5 5 Non-canonical
Question 2 Canonical 5 3 1 Non-canonical
Question 3 Canonical 4 1 Non-canonical 5
Response type: total 15 12 2
Subject: SJ
Question 1 Canonical 2 1 2 1
Non-canonical 1
Question 2 Canonical 1 3 Non-canonical
Question 3 Canonical 1 Non-canonical 1 1
Response type: total 2 3 2 3 6
Subject: AS
Question 1 Canonical 4 1 1 2
Non-canonical
Question 2 Canonical s 3 2
Non-canonical
Question 3 Canonical 4 2
Non-canonical
Response type: total 1 13 1 6 4
H.K.J. van der Lely I Cognition 51 (1994) 29-72 45
Table 3. (Continued)
Transitive scenes Locative scenes
Response type Response type
Verb Intrans. Trans. Verb Intrans. Trans. Locative only only
Subject: AW
Question 1 Canonical 3 3 1
Non-canonical 1
Question 2 Canonical 1 1 3 3
Non-canonical 1 1
Question 3 Canonical 3 2 1
Non-canonical 1 1 2
Response type: total 5 7 9 6 1
Subject: DL
Question 1 Canonical 3 2 1 Non-canonical
Question 2 Canonical 3 2 1 2 Non-canonical
Question 3 Canonical 1 1
Non-canonical 1 1
Response type: total 1 7 3 5 3
Table 4. Experiment 1. Forward linking: mean scores for the novel verb responses for the SLI children and LA controls
SLI children LA controls
Mean % (SD) Mean % (SD)
Transitive scenes
Novel verb usea
Number of canonically linked
thematic roles (max. = 20)b
83.33 (19.02) 78.91 (19.06) 53.33 (23.38) 46.18 (32.38)
Locative scenes
Novel verb usea
Number of canonically linked
thematic roles (max. = 30)h
83.50 (13.65) 64.91 (34.41) 42.20 (16.02) 36.27 (30.03)
“Responses to questions 1, 2 and 3.
bResponses to questions 1 and 2.
Therefore, to score 100% all the children’s responses would have had to contain
the novel verb and the maximum number of possible thematic roles (two for each
of the two responses for the five transitive verbs and three for the two responses
for each of the five locative verbs). Approximately 80% of the children’s
46 H.K.J. van der Lely I Cognition 51 (1994) 29-72
responses contained the novel verb and, although appropriate, the children’s
responses did not always express the maximum number of thematic roles.
The effects of subject group and verb type (transitive or locative) were
investigated using a 2 x 2 ANOVA (Subject group x Verb type). It can be seen
from Table 4 that there was a lower proportional mean score for the locative
verbs than for the transitive verbs, reflecting the increased semantic-syntactic
complexity for these verbs. However, the main effect of verb type failed to reach
the significance level, F(1, 21) = 3.47, p = ,076. Surprisingly, the SLT children’s
mean proportional scores showed that they expressed more thematic roles
canonically than the LA controls for both the transitive and locative scenes.
However, the main effect of subject group was not significant, F(1, 21) = 0.27,
p = .607, nor was the Group x Verb type interaction, F(1, 21) = 0.01. The initial
analyses indicated that the SLI children and the LA controls were performing in a
similar way. The children expressed thematic roles of novel verbs demonstrated in
transitive and locative scenes canonically and productively in utterances. The two
groups also expressed a similar number of canonical thematic roles in their
responses.
Further planned analyses were undertaken to establish whether differences in
the performance of the SLI children and LA controls could be found. The first of
these investigated whether the two groups’ responses reflected a similar propor-
tion of argument structures types. From the 1.5 possible responses (i.e., the
responses to the three questions for each of the five verbs) for the transitive
scenes, the proportion of intransitive and transitive responses using the novel
verbs was calculated for each child. The proportion of locative responses for the
locative scenes was also calculated. Canonical and non-canonical assignment of
the thematic roles was not taken into account as this had been assessed in the
previous analysis. Table 5 shows the mean proportion of responses for the
transitive and locative scenes according to the sentence type used. One-way
(Group x Sentence type) ANOVAs were carried out on the data from the
Table 5. Experiment 1. Forward linking: mean proportion of sentence types for
the transitive and locative scenes for the two subject groups
Sentence
response type
SLI children
Mean % (SD)
LA controls
Mean % (SD)
Transitive scenes (max. score = 15) Intransitive 32.22
Transitive 43.33
Locative scenes (max. score = 15) Intransitive 30.00
Transitive 42.22
Locative 5.54
(35.17) 13.33 (17.64)
(39.26) 40.39 (31.93)
(28.83) 5.49 (8.24)
(30.60) 31.76 (37.38)
(10.65) 9.80 (17.81)
H.K.J. van der Lely I Cognition 51 (1994) 29-72 47
transitive and the locative scenes. Analysis of the transitive scenes revealed no
significant difference between the groups in the proportion of transitive responses
used, F(1, 21) = 0.03. The proportional scores for the intransitive responses
showed a skewed distribution and were therefore transformed by taking the
square root. The analysis was carried out on these transformed data. Although
the SLI children showed a higher mean proportion score in comparison to the LA
controls for the intransitive responses, the difference was not significant, F(1,
21) = 1.72, p = .204.
Analysis of the locative scenes revealed no significant differences in the
proportion of locative responses, F(1, 21) = 0.30, or transitive responses, F( 1,
21) = 0.38. The proportional scores for the intransitive responses were trans-
formed to control for a skewed distribution. Unlike the previous analyses, and
despite the high standard deviation, the SLI children were found to use a
significantly higher proportion of intransitive responses than the LA control
children, F (1, 21) = 9.18, p = 0.006.
Individual subject analysis
Because of the reported heterogeneity of SLI children as a group, the
individual pattern of performance for each child was analysed on the basis of a
“consistently canonical response”. For the purpose of this analysis, if the response
to each question contained at least one canonically linked thematic role and no
non-canonical linked thematic roles, it was counted as a canonical response. This
criterion was used as pragmatic reasons may have caused the children not to
express all the possible thematic roles in their responses.
The number of canonical responses were calculated for each child for the
transitive and locative scenes (see Table 3 for the SLI children’s individual novel
verb responses). Responses to question 3 were not included in this analysis owing
to the known difficulty some SLI children have with morphology and in producing
and comprehending passive sentences. Therefore, the scores were based on the 20
responses from questions 1 and 2 for the 10 verbs. The criterion for a consistently
canonical response was set at 14 or more out of 20, p = .057, cumulative
binomial. This criterion was a conservative one as the children did not always use
the novel verb, and when they did they sometimes did not express any thematic
roles.
Five of the SLI children passed the criterion. One SLI child, SJ, did not pass
the criterion, scoring 10 canonical responses out of the possible 20. This was
largely due to a greater number of real verb responses, rather than the novel
verb, and some elliptical verb-only responses. Five of the 17 LA controls also did
not pass the criterion. This was due to elliptical verb-only responses, made by
some of the younger LA control children. Selection of appropriate control
48 H.K.J. van der Lely I Cognition 51 (1994) 29-72
children in studies investigating SLI children is always a problem. The most
appropriate control group is language-matched controls for investigations of
language abilities. This has the disadvantage in that the control children are
younger and their non-linguistic cognitive abilities are less mature than those of
the SLI children. It is possible that for the LA control children the task demands
in Experiment 1 exceeded their development of cognitive abilities, and that this
accounted for the elliptical responses made by some of the youngest children.
The data were scrutinized for non-canonical responses to questions 1 and 2; for
example, where the demonstration showed a boy taking an agent role and the girl
a patient role, who were assigned to an object and subject, respectively, in an
active sentence. None of the LA controls made any non-canonical responses. Two
of the SLI children (MP and AW) made four non-canonical responses (20%) and
two SLI children (SJ and OC) made one such response (5%). This indicates that
some SLI children have a slight but non-consistent difference in response pattern
in comparison with the LA control children. It can also be seen from Table 3 that
MP made an additional five non-canonical responses to question 3. Scrutiny of the
data revealed that all of these non-canonical responses were due to ambiguity
between word order and morphology; for example, The lady is fiming by the man. Thus, they may be attributed to his difficulties with morphology and hence with
forming passive sentences. Whilst it is interesting to note these morphological
errors, further discussion of these errors falls outside the scope of this paper.
Discussion
Throughout the results, generally, the SLI children were not found to be
performing significantly differently from the younger children who were carefully
matched on the comprehension and expression of single word vocabulary and
expressive morphology.
The SLI children and LA controls were able to observe a scene acted out with
toys, identify the participants and their appropriate thematic roles, and verbally
express these relationships in a sentence. Although the mean proportion of
thematic roles expressed in the responses are lower than those found when real
verbs are used or when novel verbs are modelled in a sentence (e.g., Pinker et al.,
1987), the mean proportions are still much greater than zero. The proportions of
canonical thematic roles expressed by the children in this experiment (approxi-
mately 40-50%) are comparable to those found in previous experiments in which
unmodelled novel verbs were elicited in sentences (Gropen et al., 1989). The
results indicate that both the normally developing and SLI children were using
forward, canonical linking productively to map from semantic-thematic roles
onto syntactic structures. The findings support previous research which has
indicated that normally developing children can use canonical linking rules to
H.K.J. van der Lely I Cognition 51 (1994) 29-72 49
express events containing novel verbs without having heard the verb in a syntactic
context (Gropen et al., 1989, 1991a,b). Analysis of the different sentence types
(intransitive, transitive or locative sentences) expressed by the children, when
using the novel verbs, showed that the SLI children and LA controls did not differ
in their use of locative or transitive syntactic frames. Importantly, these data
indicate that the similar scores obtained for the canonically linked thematic roles
by the SLI and LA control children were not achieved for different reasons; for
example. the SLI children producing fewer transitive novel verb responses but
making fewer errors. The SLI children used more intransitive sentences for both
the transitive and locative scenes with the difference reaching significance for the
locative scenes. It is likely that this increased proportion of intransitive responses
can account for the SLI children’s overall slightly higher number of canonically
linked thematic roles.
The group analysis was generally supported by the individual subject analysis.
This showed that five of the SLI children used forward canonical linking
consistently; that is, 80% or more of the time. One SLI child used forward
canonical linking in only 50% of her responses. This lower forward linking score
was due, primarily, to a greater use of real verbs, rather than the novel verbs. The
few non-canonical responses to questions 1 and 2, which were used by four SLI
children, could indicate that they were performing differently from the LA
controls. (None of the LA controls showed any non-canonical responses.) Three
of the eight non-canonical responses used by two SLI children occurred for the
novel verb mooking. It is possible that the agent role was not as salient in the
demonstration for this verb as it was for some of the other verbs. In addition, the
influence of semantic argument structures of previously (conservatively?) learned
verbs which share semantic similarities may have detrimentally influenced the SLI
children to a greater extent than their younger language peers. The following
example, which has some semantic similarities with mooking, illustrates this
possible conflict: in the situation where a girl gives a boy a “piggy-back”, the girl
is seen as the agent although it may be that the boy approaches the girl and climbs
onto her back, whereas for the verb “ride” the opposite is true. In view of the
evidence that SLI children can and do consistently assign thematic roles canoni-
cally to syntactic structures this would seem to be a plausible explanation for these
non-canonical responses.
EXPERIMENT 2: COMPREHENSION OF NOVEL VERBS
To assess how the children had encoded the semantic structure of the newly
learned novel verbs, the expressive responses for each verb in Experiment 1 were
followed by a comprehension task. The 10 novel verbs from Experiment 1 were
each presented in reversible active and passive sentences, according to the verb’s
50 H.K.J. van der Lely I Cognition 51 (1994) 29-72
prescribed meaning: for example, The lorry is zeked by the car to the train (prepositional locative). The child’s task was to act out with small toys the
meaning of the sentence containing the newly learned verb. This task could tap
the previous use of productive forward linking and a current use of reverse linking
to facilitate mapping between the thematic roles and syntactic functions.
Method
Subjects
The same group of subjects who participated in Experiment 1 participated in
Experiment 2.
Design and materials
For each of the 10 verbs created in Experiment 1, six reversible sentences were
constructed for the comprehension task for Experiment 2. There were three
active and three passive sentences. The six sentences for each verb were written
onto a presentation card in a random order, except that the first sentence was
always an active one. The eight toys used in Experiment 1 were used in this
experiment.
Procedure
The six questions testing comprehension for each verb immediately followed
the presentation of the corresponding verb in Experiment 1. The acting out
procedure was explained simply to the child, who was asked to show the puppet . . for example, the car tivs the train (question 1). The six questions for
each novel verb were presented in the same random order for each child. A
description of the child’s response was recorded by the experimenter on prepared
test sheets.
Scoring
An acting out response was scored as correct if it conformed to a canonical
linking (as defined in Experiment 1) between all of the syntactic functions in the
sentence and the thematic roles the toys were made to take on. Errors were
assigned to the following categories:
(4
(b)
(4
(d)
H.K.J. van der Lely I Cognition 51 (1994) 29-72 51
Non-canonical; that is, when the correct toys were selected but the
thematic roles demonstrated did not correspond to a canonical linking; for
example, the subject in an active sentence was assigned to a patient
role.
Semantic error: (i) verb; (ii) noun; that is, either an incorrect action was
demonstrated, or the child selected an object that was not mentioned in
the sentence.
Omission: one or more thematic roles were omitted in the acting out of the
sentence.
Ambiguous;
on were not
that is, where the thematic roles the toys were made to take
clear.
Results
Group analysis
The total number of responses that corresponded to canonical linking was
calculated for each child. Table 6 shows the mean scores for these data for the
two subject groups. A mixed design 2 x 2 X 2 ANOVA was used to investigate the
data. There was one between-subject variable: subject group (SLI children, LA
controls); and two repeated measures: sentence type (transitive and locative) and
word order (active, passive).
The main effects for sentence type and word order were significant, F( 1,
21) = 17.72, p < ,001, and F(1, 21) = 12.76, p = .002, respectively. As can be
seen from Table 6, this reflected the higher scores for the transitive than for the
locative sentences, and for the active than for the passive sentences.
Although the SLI children generally performed worse than the LA controls the
main effect of subject group did not reach the significance level, F(1, 21) = 3.30,
p = .084. None of the interactions were significant, F(1, 21) <2.74, p > .113, in
all analyses. The SLI children and the LA controls performed at a low level on
the passive locative sentences, and the SLI children also performed at a low level
Table 6. Experiment 2. Comprehension: mean canonical acting out responses to the sentences by the SLI children and LA controls
Sentence type SLI children (max. = 15) LA controls (max. = 15)
Mean (SD) Mean (SD)
Transitive actives 11.33 (1.03) 14.29 (1.53) Transitive passives 10.16 (3.97) 12.53 (3.77)
Locative actives 7.33 (5.60) 10.47 (4.89)
Locative passives 6.67 (5.32) 7.34 (5.37)
52 H.K.J. van der Lely I Cognition 51 (1994) 29-72
on the active locative sentences (see Table 6). To clarify the results and establish
whether the children were performing above chance level, and to take account of
possible heterogeneity in the SLI group, an individual subject analysis was carried
out.
Individual subject analysis
To investigate individual subjects’ pattern of responses the data from each
child were recoded to derive a pass/fail score (i.e., an above/below chance
performance) for each sentence type. The criterion for a consistent response was
set at 12/15 for the transitive sentences, p = .035. For the locative sentences the
criterion was set at 6/15, p < .05. The criterion for the locatives was based on the
possibility that the three nouns in the sentence could potentially take on any of
the thematic roles. Therefore, the probability of a single correct response for the
locative sentences was estimated to be p = .167. (In the event, if the stricter
criterion of 12/15 had been used it would not have made a difference to the
results.) Table 7 summarizes these data. Analysis of the numbers of children from
each subject group passing or failing the criterion for each set of sentences was
carried out using the Fisher exact test. No significant differences between the two
groups were found, confirming the findings from the group analysis.
Although on each of the individual pass/fail analyses there were no significant
differences between the groups, the percentages of SLI children passing the
criterion was lower than that of the LA controls for each sentence type. It was
possible that across the four sentence types the SLI children were performing
below the LA controls. A meta-analysis was carried out to test for this possibility.
The exact probability from each of these tests was combined using the
following formula: ~22~ = -2~ log,; Pi will be distributed on x2(2 d.f.) (Howell,
Table 7. Experiment 2. Comprehension: numbers of SLI and LA control children reaching the criterion for the active and passive sentences for the two verb types
Sentence type SLI children LA controls
Fail (%) Pass (%) Fail (%) Pass (%) p
Transitive active 2 (34) 4 (66) 2 (12) 15 (88) ,231
Transitive passive 3 (50) 3 (50) 4 (24) 13 (76) .226 Locative active 2 (34) 4 (66) 4 (24) 13 (76) ,638
Locative passive 3 (50) 3 (50) 7 (41) 10 (59) ,708
H.K.J. van der Lely I Cognition 51 (1994) 29-72 53
1987).4 The result from this combined test was still not significant (xi = 7.5
p > .25).
Discussion
The results of the test of the comprehension of active and passive sentences
containing the novel verbs showed that the SLI children’s performance did not
differ significantly from the LA controls’ performance. This finding was confirmed
by the analysis of the pass/fail data based on individual children’s above/below
chance performance. The increased syntactic and semantic complexity of the
passive and locative sentences can account for the overall worse performance on
these sentences by the SLI children and the LA controls. Only one SLI child (AS)
showed a consistently correct response to all four sentence types. One of the SLI
children (DL) did not show a consistently correct interpretation on any of the sets
of sentences. Two out of three of DL’s LA-matched controls also failed to
respond correctly on any of the sets of sentences. Thus, DL’s failure may be
attributed to his general level of language abilities.
The results of these data indicate that at least five of the SLI children were able
to recall the semantic lexical representation of the newly learned verb. It is likely
that this lexical representation of the verb was based on the previous use of
forward linking. The findings further indicate that the children were able to learn
the syntactic properties of a verb by using the canonical linking rules in a forward
direction, and could retrieve the semantic properties associated with a lexical
entry. The pass/fail data revealed that only some of the SLI children can use
productive rules for mapping with more complex structures such as the passive or
locative sentences. The SLI children who were able to interpret the passive
sentences may still have been able to rely largely on their previous representation
of the newly learned novel verb: that is, if the SLI children processed the passive
participle and know something about its stative qualities this could have been
sufficient to assign the first NP to the patient/theme role, without further syntactic
processing of the sentence.
It is possible that the children used the syntactic cues from each sentence and
applied the linking rules in the reverse direction to help them assign the syntactic
functions to thematic roles. It is not clear whether the SLI children were relying
solely on their previous representation of the verb, based on forward linking, or
whether they were also using reverse linking to facilitate mapping. A reliance only
on the newly formed representation of the verb could account for the poorer
‘Strictly speaking this assumes that the samples are independent.
54 H.K.J. van der Lely I Cognition 51 (1994) 29-72
performance of the SLI children in Experiment 2. Experiment 3 aimed to clarify
whether SLI children could use reverse linking productively.
EXPERIMENT 3: REVERSE LINKING
For Experiment 3, involving syntactic to semantic mapping, a new set of novel
verbs was presented in a total of 24 reversible active and passive transitive
sentences and active locative sentences. Passive prepositional locative sentences
were not included in this experiment owing to the low (chance) level of
performance on these sentences in Experiment 2. The children’s task was to
assign appropriate thematic roles to the grammatical functions; to make up an
event that went with the novel verb in the sentence; and to act out with toys a
possible meaning. For example, on hearing the sentence The girl voozes the boy, the child was to make the girl (agent) do something to the boy (patient). A
productive use of reverse linking was required to complete the task. This task
investigated the use of only syntactic cues to assign thematic roles in the absence
of an observable event. Therefore, in contrast to the previous two experiments,
there were no semantic cues available to constrain the possible meanings of the
verbs in this task.
Method
Subjects
The same subjects participated in Experiment 3 as in the previous two
experiments.
Design and materials
A further 16 novel CVC verbs were invented. For eight of the novel verbs
reversible active sentences were created; for example, The girl kals the boy. The
same eight verbs were used to create eight reversible passive sentences. These
sentences consisted of the same subject and object for each verb, but with passive
morphology, thereby changing the thematic roles of the grammatical functions
(e.g., The girl is kalled by the boy). The remaining eight novel verbs were used to
construct eight reversible active prepositional locative sentences with a final to
H.K.J. van der Lely I Cognition 51 (1994) 29-72 55
prepositional phrase (e.g., The cur rits the train to the lorry). A full list of the
sentences can be found in Appendix B.
The nouns used in the sentences corresponded to seven of the eight toys used
in Experiment 1. The tractor was omitted, as it may not have been so familiar to
some of the children. The nouns were counterbalanced across the sentences, so
that they occurred with approximately equal frequency in subject, object and
oblique object positions.
The eight toys used in the previous two experiments were also used in this
experiment.
Procedure
The experimenter sat opposite the child at a small table in a quiet room. The
eight toys were placed in front of the child. Prior to presentation of the test
sentences, the child was encouraged to play with the toys and to demonstrate
novel actions that the toys could be made to perform. Examples of possible
actions were demonstrated by the experimenter. For instance, the toy girl was
made to bend a sitting person over; or the car was made to repeatedly tap the top
of the lorry. The child was then told that it was her turn to make up or guess what
she thought the sentence with the “funny word” could mean and to show the
puppet by acting it out with the toys. It was stressed to the child that the
sentences could mean lots of things and that there was not a strict “right” or
“wrong” answer. However, they were to try very hard to make up a possible
meaning. Sentence presentation was random, except that a transitive active
sentence was presented first. All 24 sentences were presented in one session. A
description of the acted out responses was recorded at the time of testing. If the
experimenter was unsure of the child’s intentions as to which toy was (for
example) the agent, the child was asked to show the experimenter the meaning of
the sentence again and the stimulus sentence was repeated. If the thematic roles
assigned to the toys were still unclear the response was scored as ambiguous. In
the event, there were few ambiguous responses which required a repetition and of
those which did very few remained ambiguous. For the majority of responses
there was little possibility for misinterpreting the child’s intention as to which
thematic role the toys were taking on. The children tended to make the agent
very active and the patient/theme very passive; for example, making the boy
jump up and down on the back of the girl who was lying on the table, or making
the car hit the lorry off the table. An example of a set of responses for three SLI
children and three LA controls to one of the sentences can be found in Appendix
D.
56 H.K.J. van der Lely I Cognition 51 (1994) 29-72
The child’s acting out responses were scored as correct when the thematic roles
the toys were made to take on conformed to a canonical reverse linking. The
criterion used for canonical linking was the same as that set out in Experiment 1;
for example, for the active sentences the subject was assigned to an agent role,
the object to a patient/theme role, and for the locative sentences the oblique
object was assigned to a goal role. For a correct response all the syntactic
functions mentioned in the sentences had to be assigned canonically. Other
responses were coded into the same categories as those in Experiment 2: (a)
non-canonical, (b) semantic, (c) omission, (d) ambiguous. For further details see
scoring section, Experiment 2.
Results
Group analysis
The mean correct scores (i.e., acting out responses which corresponded to
canonical reverse linking) are presented in Table 8. A 2 x 3, Subject group x
Sentence type (active transitive, passive transitive and active locative sentences)
ANOVA design was used to investigate these data.
The main effect of subject group was significant, F(1, 21) = 9.57 p = .006,
which reflected the worse scores for the SLI children in comparison to the LA
controls. The main effect of sentence type was also significant, F(2, 21) = 12.39
p < .OOl. The Group X Sentence type interaction was not significant.
Planned comparisons provided further insight into the performances of the two
groups to the different sentence types. For the LA controls no significant
difference was found between their performance on the transitive and locative
sentences. However, a significant difference was found between the active and
passive transitive sentences, F(1, 16) = 6.75 p = ,019, which reflected the higher
scores for the active sentences.
For the SLI children, both comparisons were significant. Their performance on
Table 8. Experiment 3. Reverse linking: mean canonical acting out responses to the sentences by the SLI and LA control children
Sentence type SLI children (max. = 8) LA controls (max. = 8)
Mean (SD) Mean (SD)
Transitive active 5.50 (0.84) 7.29 (1.44)
Transitive passive 3.67 (1.50) 5.16 (2.79)
Locative active 2.33 (2.25) 5.53 (3.04)
H.K.J. van der Lely I Cognition 51 (1994) 29-72 57
the locative sentences was found to be significantly below that of the transitive
sentences, F(1, 5) = 13.97, p = .013. Their performance on the passive transitive
sentences was also significantly below that of the active transitive sentences, F(1,
5) = 7.86 p = .038.
Individual subject analysis
As in Experiment 2, to clarify the results and consider individual subjects’
pattern of responses and possible heterogeneity in the groups, the data were
recoded into a pass/fail score (above/below chance performance). The children
were credited with consistently producing a canonical response if they scored at
least 7/8 canonical responses (p < .035) for the active and passive transitive
sentences. The probability of a canonical response for the locative sentences was
again estimated to be p = .167. Therefore, the criterion was set at 418 (p = .031).
The number of subjects in each group who were able to interpret consistently
each set of sentences canonically was calculated (see Table 9). None of the SLI
children reached the criterion for the two sets of transitive sentences. Fisher exact
tests revealed significant differences between the groups for the active sentences,
p < .OOOl, and for the passive sentences, p = ,022. A marginally significant
difference was found for the locative sentences, in which two SLI children passed
this less stringent criterion (p = .056). These results showed that significantly
fewer SLI children than LA control children consistently used reverse linking,
mapping from the syntactic structure to a semantic structure, on the three
sentence types.
The analysis of errors made by the SLI children revealed that nearly all of the
errors fell into the non-canonical category. An appropriate event was invented by
the children involving the toys mentioned in the sentence; the thematic roles were
compatible with the syntactic frame; but the assignment of which thematic role
corresponded to which toy did not correspond to the canonical linking.
In Experiment 2, although there was no significant difference between the two
groups’ performance, the SLI children’s performance was lower than that of the
Table 9. Experiment 3. Reverse linking: numbers of SLI and LA control children whose acting out responses reached the canonical response criterion Sentence type SLI chiidren LA controls
Fail (%) Pass (%) Fail (%) Pass (%) p
Transitive active 6 (100) 0 (0) 2 (22) 15 (88) .OOOl
Transitive passive 6 (100) 0 (0) 8 (47) 9 (53) ,022
Locative active 4 (66) 2 (33) 4 (24) 13 (76) .056
58 H.K.J. van der Lely I Cognition 51 (1994) 29-72
LA controls. It is important to establish whether the SLI children and LA control
children were performing differently across Experiments 2 and 3. Both experi-
ments used the same acting out comprehension paradigms. However, only reverse
linking could be used in Experiment 3 and no semantic constraints were available,
whereas forward and reverse linking could be used in Experiment 2 and semantic
constraints were available from the newly learned lexical representation. Further
analysis of Experiments 2 and 3 together is warranted to help rule out alternative
explanations based on the task demands.
Analysis of Experiments 2 and 3
Based on the percentage of correct scores for each experiment a further 2 x 2
(Subject group x Experiment) ANOVA was carried out to investigate these data
(see Table 10).
The main effect of subject group was significant, F(1, 21) = 4.46, p = ,047,
which reflected the lower scores for the SLI children. The main effect of
experiment was not significant, F(1, 21) = 0.89. However, the Group X
Experiment interaction approached significance, F(1, 21) = 3.87, p = .062. Simple
effects analyses revealed that the two group’s performance in Experiment 2 was
not significantly different, F(1, 21) = 1.65, p = .214, whereas in Experiment 3 a
significant difference was found, F(1, 27)5 = 7.13, p = .013. This confirmed that
the SLI children’s performance was significantly worse than the LA controls’
performance in Experiment 3, but not in Experiment 2.
Analysis of the LA controls’ performance across the two experiments revealed
no significant difference, F(1, 16) = 0.52, p = .23, one-tailed, indicating that they
had a similar level of performance on Experiments 2 and 3. In contrast, analysis
of the SLI children’s performance across the two experiments revealed a
significant difference, F(1, 5) = 4.23, p = .026, one-tailed. This reflected the
lower scores obtained on Experiment 3.
The result of these analyses together all indicate that the SLI children were
Table 10. Mean proportion of canonical responses for Experiments 2 and 3 for the SLI and LA control children
SLI children
Mean % (SD)
LA controls
Mean % (SD)
Experiment 2 59.15 (24.33) 73.22 (22.11)
Experiment 3 47.90 (16.62) 77.18 (25.31)
‘Because a pooled error term is used the degrees of freedom associated with the pooled error term
were adjusted to remove bias in the test (Satterthwaite, 1946).
H.K.J. van der Lely I Cognition 51 (1994) 29-72 59
performing significantly worse on Experiment 3 in comparison to both their own
performance in Experiment 2 and in comparison to the LA controls’ performance
in Experiment 3.
Discussion
The results of the performance of the SLI children in Experiment 3 clearly
contrasted with their performance in Experiment 1. Analysis of the mean correct
scores showed that the SLI children were significantly worse at using the syntactic
cues to derive thematic roles based on reverse linking. Both groups of children
found assigning thematic roles to passive transitive sentences more difficult than
to active transitive sentences. This finding concurs with previous research into
normal language development which has shown the particular difficulty of
comprehending known verbs in reversible passive sentences (Bever, 1970; de
Villiers & de Villiers, 1973) and the lower performance with production and
comprehension of passive sentences with novel verbs (Pinker et al., 1987). The
SLI children, unlike the LA controls, performed significantly worse on the
locative sentences than the transitive sentences. This suggests that the SLI
children were affected to a greater extent by the syntactic complexity of the
locative sentences.
The pass/fail data supported and clarified the findings of the group analysis.
This analysis showed that the significantly worse performance of the SLI children
in comparison to the LA controls was evident for all three sentence types. The
data were surprising in that nane of the SLI children reached the criterion for a
consistently correct response on the active or passive transitive sentences, and
only two SLI children (AS and AW) reached the criterion on the locative
sentences. It can be recalled that the criterion for the locative sentences was set at
4/8. However, it is possible that parsing the locative sentences into phrases would
enable the “goal” thematic role to be assigned separately from the verb.
Jackendoff (1989) argued that nouns in prepositional phrases are assigned their
thematic role-argument structure by the preposition and that they are merely
“licensed” to co-occur with a particular verb. Therefore, the assignment of
thematic roles for the so-called locative verb would be a binary choice between
the subject and object. If this were so, the appropriate criterion for a consistently
correct response should have been 7/8 for the locative sentences as well as the
transitive sentences. The two subjects from the SLI group, AS and AW (who
scored 4/8 and 6/8, respectively), would not have reached this more stringent
criterion, although 8/17 (47%) of the LA controls would have reached the
criterion. It is also interesting to note that the LA controls who failed to interpret
the locative sentences generally interpreted the oblique object as an additional
theme or omitted this third argument. However, they assigned the subject and
60 H.K.J. van der Lely I Cognition 51 (1994) 29-72
object noun phrases canonically to thematic roles. No such pattern emerged for
the majority of SLI children. These children’s errors appeared to be unsystematic
with any of the grammatical functions being assigned any thematic role.
Analysis of the Group x Experiment (Experiments 2 and 3) interaction
approached significance. Analysis of the simple effects revealed that the SLI
children performed significantly worse on Experiment 3 than on Experiment 2,
whereas the LA control children showed a similar level of performance across the
two experiments. The analysis confirmed that the SLI children’s performance was
significantly worse than the LA controls’ performance on Experiment 3 but there
was no significant difference in their performance on Experiment 2. The
difference in the performance of the two groups in these two acting out tasks is
useful in helping to identify the underlying reasons for the SLI children’s deficit in
reverse linking revealed by this study, and will be taken up in the general
discussion below.
The SLI children’s chance level of performance as measured by the individual
analysis (see Table 9) in a task involving canonical reverse linking is particularly
striking when considered in relation to previous research investigating the use of
syntactic cues and canonical linking rules to infer aspects of the meaning of a
verb. Gleitman (1989), using a looking preference task, found that young children
of 3 and 4 years were heavily guided by the syntactic frame when presented with
novel sentences, with 85% of all responses showing canonical mapping. The
individual analysis of the LA control’s responses in this experiment, in which 88%
of the children reached the consistent canonical mapping criterion for active
transitive sentences, concur with Gleitman’s findings. These data support the view
that for children developing language normally, the syntax may guide the child’s
choice of interpretive options, particularly in ambiguous circumstances. In
addition, the data from this study provide evidence that the syntactic cues from a
single syntactic frame may be used in the absence of an observable scene and any
possible semantic cues, to help assign thematic roles to a verb for which they do
not have a semantic-lexical representation. In this way the syntactic cues may
constrain the possible meaning of a verb in that the verb’s meaning must allow or
“license” the particular thematic roles (as determined by the syntactic frame) in
its semantic-lexical representation. The data also indicate that this mechanism
may not be available for the SLI children.
The implications of the findings to the underlying cause of SLI in children and
the nature of forward and reverse linking will now be explored in the general
discussion.
GENERAL DISCUSSION
Two different mechanisms for mapping thematic roles with syntactic function
were investigated: the use of productive forward linking involving semantic cues
H.K.J. van der Lely I Cognition 51 (1994) 29-72 61
to map from semantics to syntax; and the use of productive reverse linking
involving syntactic cues to map from syntax to semantics.
Surprisingly, the use of productive forward linking and reverse linking has not
previously been investigated in a single population. It was not clear whether
individual differences would be apparent in the use of forward linking and reverse
linking. For example, would the children be able to use both mechanisms when
the context demanded their use, or alternatively, would they rely on one
particular mechanism? A good use of both mechanisms was shown by the
majority of the LA controls. This was not so for the SLI children, who showed an
impairment in the reverse linking task despite their good use of productive
forward linking. The subgroup of SLI children investigated in this study appeared
to be homogeneous in that all the children conformed to this pattern.
The data from this study showing differences between Experiments 1 and 2
versus Experiment 3 suggest that there are differences between using linking rules
in a forward versus reverse direction. These differences may account for the
particular difficulties the SLI children have with reverse linking. The data suggest
that the differences between using linking rules in a forward versus reverse
direction may be linguistic. Before I explore these issues it is important to
discount alternative explanations which arise from differences in the tasks used to
investigate forward versus reverse linking. The task differences are, first, the
demands of the production task in Experiment 1 and the comprehension task in
Experiment 3, and secondly, the non-linguistic cognitive demands of the tasks in
Experiments 2 and 3.
Experiments 1 and 3: task differences
Can the difference between the forward and reverse linking tasks be explained
by the different processes involved with producing versus comprehending a
sentence? Clark and Hecht (1983) pointed out that in language production the
child has to choose the words and structures to convey his or her own intention,
whereas in comprehension the child has to infer another person’s intentions.
Thus, one may look to find a deficit in inferential processing or, alternatively,
other processing demands incurred in comprehension but not production. How-
ever, a search for identifying processing differences between comprehension and
production is not warranted by these data and could not explain the SLI children’s
pattern of performance in this study: a different pattern of performance was
found for the SLI children in comparison to the LA controls in Experiment 2
versus Experiment 3, both of which required comprehension of sentences. In
addition, the SLI children performed significantly worse on Experiment 3 than
Experiment 2, whereas the LA controls had a similar level of performance in the
two experiments.
62 H.K.J. van der Lely I Cognition 51 (1994) 29-72
The role of familiarity with verbs
Another consideration is that the comprehension task in Experiment 2 was in
some ways “easier” than that in Experiment 3. It is possible that increased
familiarity and experience with the verbs in Experiment 2 is the main reason why
the SLI children did better on Experiment 2 than Experiment 3. However, the
verb was presented as a gerund in the demonstrations in Experiment 1. Thus, no
matter how many times the child was exposed to this situation the increased
familiarity would not give any indication of the assignment of the verb’s thematic
roles to particular syntactic structures. Merely being familiar with the verb and its
meaning will not provide the vital clues to improve mapping performance.
Furthermore, evidence from the analysis of the LA controls’ performance across
Experiments 2 and 3 indicates that Experiment 2 was not easier than Experiment
3. The LA controls showed a similar level of performance on both experiments
(see Table lo), with a slightly higher mean score on Experiment 3, which clearly
goes against this explanation.
Non-linguistic cognitive processing
It is difficult to see how non-linguistic demands could account for the overall
pattern of performance of the children found in this study. First the LA controls
did not generally show an inability to carry out Experiment 3, and their
performance on Experiments 2 and 3 was similar. Also, it was the SLI children
who were impaired on Experiment 3 and not the younger, and cognitively less
mature, LA control children. SLI children by definition have normal age-
appropriate non-linguistic cognitive abilities and show normal abilities on, for
example, non-linguistic hypothesis testing tasks (Kamhi, Catts, Koenig, & Lewis,
1984). Pre-experimental testing showed that all the SLI children in this study
performed within normal limits on non-linguistic performance subtests of stan-
dardized IQ tests (e.g., WISC-R); achieved an equivalent IQ score of 90 or above
on two further tests of non-linguistic abilities (van der Lely & Harris, 1990); and
the same two subject groups did not differ on extensive testing of auditory-verbal
short-term memory (van der Lely & Howard, 1993).
Forward versus reverse linking
The most likely explanation for the performance of the SLI children may be
found in the linguistic differences between using linking rules in a forward versus
reverse direction. I shall now discuss the implications of the findings for the
characterization of this subgroup of SLI children and the differences between
H.K.J. van der Lely I Cognition 51 (1994) 29-72 63
forward and reverse linking. The findings will then be considered in relation to
the previous findings from investigations of sentence comprehension and mor-
phology in SLI children.
First, although the SLI children were considerably worse at using reverse
linking, their relatively good performance with forward linking indicates that they
have the innate ability or have learned the canonical relationship, that is, the
linking rules, between specific thematic roles and syntactic functions. Whilst this
may be sufficient to carry out the forward linking task it does not appear that it
was adequate for the reverse linking task. At this point it may be useful to
reiterate the inherent difference between forward and reverse linking. The
difference is due to the many-to-few mapping between thematic roles and
grammatical functions. The semantic information available in forward linking
provides the necessary constraints for a one-to-one mapping between semantic
roles and syntactic functions to be sufficient to complete the task without analysis
of the overall syntactic frame and the relationship between the constituents. The
many-to-few mapping between thematic roles and syntactic functions makes this
problematic without semantic constraints for reverse linking. Crucially, for
reverse linking, I proposed that the structural relationship, established through
c-selection and involving syntactic government, is needed to identify a particular
syntactic frame, specify each argument position within that frame, and activate a
schema or core of linking patterns. Therefore, for example, the first NP in an
utterance would be identified within an SVO, active, transitive syntactic frame.
This would activate the core of linking patterns which would assign the subject-
NP to an agent thematic role only, rather than any of the other possible thematic
roles which could be assigned to a first NP in an utterance. Alternatively, in
forward linking the semantic cues provide the necessary constraints which enable
an agent thematic role to be assigned to the first NP in the utterance.
From the proposed characterization of reverse linking there are two loci (at
least) in which a deficit may cause a child to fail. First, the child may activate the
wrong schema, or may not yet have developed a schema for common linking
patterns. Alternatively, the child’s syntactic cues may contain insufficient in-
formation for identification of the syntactic frame which enables assignment of the
thematic roles canonically.
Analysis of the errors made by the SLI children in the reverse linking
experiment provide clues as to where the SLI children’s deficit may lie. If the SLI
children’s deficit was with the schema for common linking patterns, it could be
predicted that semantic errors would occur. Semantic errors involving incorrect
thematic roles being demonstrated by the SLI children rarely occurred. Such
errors would have been made if, for example, all three mentioned nouns in the
locative syntactic frame were assigned a “theme” role. This response would be
appropriate for a sentence such as The girl, the boy and the man vode, but not for
any of the syntactic structures in this experiment. Nearly all the errors made by
64 H.K.J. van der Lely I Cognition 51 (1994) 29-72
the SLI children were associated with a non-canonical linking. An appropriate
event was invented by the children involving the toys mentioned in the sentence.
The thematic roles were compatible with the syntactic frame. However, the
assignment of which thematic role corresponded to which toy did not correspond
to the canonical linking between the syntactic functions and thematic roles. These
data suggest that the SLI children know something about a syntactic frame
involving two NPs and a verb. This was insufficient for an appropriate schema or
core of common linking patterns to be activated; for example, the argument
structure type, “transitive”6 and thematic core “X acts on Y” (Pinker, 1989) fit
these data. However, the SLI children appear unable to analyse the particular
syntactic argument positions in the sentence and reason back from these syntactic
arguments the likely thematic roles underlying them based on the linking schema.
Thus, the SLI children’s deficit is not with the schema per se, but with the
specification of the syntactic cues which would “motivate” the canonical mapping
of thematic roles. I conclude that the SLI children have the required syntactic
representations to enable them to carry out the forward linking task but that this
is insufficient (in the absence of any semantic constraints) for the reverse linking
task.
In sum: an interpretation of the data, showing differences in the syntactic
representations required for using linking rules in a forward versus reverse
direction, can account for the performance of the SLI children. A “basic”
syntactic representation, involving analysis of the individual phrases and their
linear position in relation to the verb or overall utterance, is required for
productive forward linking. A syntactic representation which specifies each
particular syntactic frame and all the argument positions within that frame (and
thus, the relationships between the constituents) is required for productive
reverse linking. I propose that SLI children have a deficit with the syntactic
representation which specifies the relationship between the verb and the argument
positions.
There are many components which could underlie the reverse linking process
and could account for the deficit found for the SLI children. I have already argued
that non-linguistic processes cannot account for the data. The possibilities are also
constrained by previous findings characterizing SLI children’s language abilities.
One possibility is a deficit in the area of government or locality (Chomsky, 1981,
1986a; Wexler & Manzini, 1987) which underlies c-selection and specifies the
asymmetrical relationship between constituents. This proposal makes predictions
across a huge range of syntactic modules and processes in which government
provides the necessary principles and constraints. It could be predicted that a
.deficit in government would cause problems with case assignment, the distribution
“‘transitive” is used here merely to identify a NP V NP sequence, which may or may not be hierarchically organized.
H.K.J. van der Lely I Cognition 51 (1994) 29-72 65
of the empty category PRO, spec-head agreement, binding theory, as well as with
theta role assignment and c-selection. Other syntactic modules and principles may
be intact; for example, the X-bar module which would be able to provide
elementary hierarchical structure through lexical categorization, phrasal projec-
tions (although the governing relationship between sister elements and the
direction of it may not be syntactically defined), and s-selection and the
generalized projection principles would ensure that all lexical information was
syntactically represented. The data from this study and previous findings from
investigations into the comprehension and the expressive language abilities of SLI
children are consistent with a deficit in government. SLI children have been found
to have particular difficulties with case, gender, tense, spec-head agreement, and
c-selection (e.g., see Clahsen, 1989; Clahsen, Rothweiler, Woest, & Marcus 1992;
Fletcher, Ingham & Kirby, 1992; Leonard, 1989; Leonard, Bortolini, Caselli, &
McGregor, in press); and with theta role assignment in comprehension of
reversible sentences (e.g., Bishop, 1982; van der Lely & Harris, 1990).
The particular strengths and weaknesses in forward and reverse linking
revealed for the SLI children may explain other findings characterizing the
linguistic abilities of SLI children. The data from this study indicate that SLI
children will be particularly impaired when language learning may be facilitated
by reverse linking. SLI children may not be able to learn verb meaning
productively from the syntactic structural cues in the input utterances. A deficit in
the use of reverse linking can explain the particular difficulties SLI children have
in comprehending sentences with “known” verbs (Bishop, 1982; van der Lely &
Harris, 1990). Pinker claims that reverse linking is mandatory for learning certain
syntactic alternations such as the active-passive alternation (Pinker, 1989). This is
an area of particular impairment in many SLI children (Bishop, 1979; Precious &
Conti-Ramsden, 1988; van der Lely & Dewart, 1986; van der Lely & Harris,
1990). If SLI children have to rely on conservative learning (i.e., direct learning
from positive evidence; Pinker, 1984) or forward linking for learning the syntactic
properties of verbs, conflict may occur between the semantic representation of the
verb, the linking rules, and the various syntactic frames the verb has occurred in.
Thus, for the SLI child hearing a verb in a set of syntactic frames, without the use
of reverse linking, multiple lexical entries may be formed, or at the very least,
conflicting information may be acquired. Therefore, the increased experience of a
verb in varying syntactic frames may only serve to confuse him rather than serving
to refine the semantic representation of that verb. This may explain why the SLI
children succeeded on the forward linking task and the comprehension of these
newly learned verbs, and yet, have previously shown an impaired performance in
the comprehension of usually early acquired verbs.
The implications for the linguistic performance of SLI children extend beyond
verb-argument structure. It could be predicted that other modules, such as
morphology, will be differentially impaired. For example, where semantic
66 H.K.J. van der Lely I Cognition 51 (1994) 29-72
bootstrapping (which relies on forward linking) is sufficient for acquisition of morphological forms, SLI children should not show a significant impairment. Clahsen’s (1989) data from German SLI children show exactly this pattern. For example, inherent features of articles such as definiteness and indefiniteness, which can be acquired through semantic bootstrapping (Pinker, 1984), are appropriate for the language abilities of the SLI children investigated by Clahsen. The data indicate that the SLI children have the DET (determiner) functional category. However, other features of the determiner phase, which do not have inherent semantic content, such as case features and gender, appear to be absent in SLI children. In German gender, which is not a primary feature of articles but assigned locally by the noun, may not be acquired through semantic bootstrap- ping. Clahsen (1989) found that gender marking on articles was specifically impaired in SLI children. He characterized the expressive language of his German SLI children as being consistent with a deficit in grammatical agreement within phrase structure. The findings from this study at one level concur with Clahsen’s claim but they suggest that this characterization is too narrow and that the SLI children’s deficit also extends to the asymmetrical syntactic relationships between constituents.
To conclude, a deficit with the syntactic structural relationships can account for both the morphological characterization of SLI children and the semantic- syntactic mapping abilities shown in this and previous studies. Further research is needed to substantiate and refine the proposed linguistic characterization of this subgroup of SLI children and to establish the scope of the proposed deficit; for example, whether the deficit is “domain specific” to just grammatical representa- tions, restricted to specific modules within grammar, or whether it is a more general language deficit for dependent structural relationships between repre- sentations. Binding theory is currently being tested as a first test of the predictions.
Conclusion
This study has shown that SLI children are able to use forward canonical linking productively to learn the syntactic properties - verb argument structures - for novel verbs. In contrast to this finding, the SLI children appear to be significantly worse in their ability to use reverse canonical linking productively to assign thematic roles to syntactic functions of a novel verb. Differences in the syntactic representations required for forward versus reverse linking can account for the findings. I proposed that reverse linking requires a more detailed representation of the syntactic structure than forward linking owing to the many-to-few mapping between thematic roles and syntactic functions. I have argued that a syntactic representation which specifies each particular syntactic
H.K.J. van der Lely I Cognition 51 (1994) 29-72 67
frame and all the argument positions within that frame is required for reverse
linking (in other words, the overall structural relationship between the verb and
its arguments). The SLI children appear unable to analyse the syntactic argument
positions within a particular syntactic frame. A linguistic characterization of this
deficit is constrained by previous investigations of SLI in children, as any
hypothesis must also account for these data. I proposed that a deficit in the area
of government or locality which underlies c-selection and specifies the syntactic
relationship between constituents can account for the data from this study and
data from previous investigations of SLI children. The hypothesized deficit of
government in SLI children makes clear predictions, but these involve a very
broad range of syntactic processes. Further research is required to explore the
claims, both in relation to the differences in the syntactic representation required
for using linking rules in a forward versus reverse direction, and in relation to the
underlying deficit in SLI children. Testing the predictions raised by this study may
not only help to identify more precisely the underlying cause of SLI but may also
help to identify modular components in universal grammar.
Finally, the study of SLI children can provide insight into language acquisition
mechanisms which could not be provided from investigations into normally
developing children.
References
Berwick, R. (1985). The acquisition of syntactic knowledge. Cambridge, MA: MIT Press.
Bever, T. (1970). The cognitive basis for linguistic structures. In J.R. Hayes (Ed.). Cognition and the development of language (pp. 279-362). New York: Wiley.
Bishop, D.V.M. (1979). Comprehension in developmental language disorders. Developmental Medi-
cine and Child Neurology, 21, 225-238. Bishop, D.V.M. (1982). Comprehension of spoken, written and signed sentences in childhood
language disorders. Journal of Child Psychology and Psychiatry, 23, l-20. Bishop, D.V.M. (1992, January). Morphological deficits in language-impaired children. Paper pre-
sented at the Experimental Pyschology Society, University College, London.
Bishop, D.V.M., & Adams, C. (1989). Conversational characteristics of children with semantic-
pragmatic disorder: II. What features lead to a judgement of inappropriacy? British Journal of Disorders of Communication, 24, 241-263.
Borer, H., & Wexler, K. (1987). The maturation of syntax. In T. Roeper & E. Williams (Eds.),
Parameter-setting and language acquisition. Dordrecht: Reidel.
Bowerman, M. (1982). Reorganizational processes in lexical and semantic development. In E. Wanner
& L. Gleitman (Eds.), Language acquisition: the state of the art (pp. 319-346). Cambridge, UK:
Cambridge University Press. Bowerman, M. (1990). Mapping thematic roles onto syntactic functions: are children helped by innate
linking rules? Linguistics, 28, 1253-1289.
Bresnan, J. (1982). The mental representation of grammatical relations. Cambridge, MA: MIT Press.
Chomsky, N. (1981). Lectures on government and binding. Dordrecht: Foris. Chomsky, N. (1986a). Knowledge of language: its nature, origin and use. New York: Praeger.
Clahsen, H. (1989). The grammatical characterization of developmental dysphasia. Linguistics, 27,
897-920.
68 H.K.J. van der Lely I Cognition 51 (1994) 29-72
Clahsen, H., Rothweiler, M., Woes& A., & Marcus, G. (1992). Regular and irregular inflection in the acquisition of German noun plurals. Cognition, 4.5, 225-255.
Clark, E., & Hecht, B. (1983). Comprehension and production and language acquisition. Annual Review of Psychology, 34, 325-349.
Connell, P. (1986). Acquisition of semantic role by language disordered children: differences between production and comprehension. Journal of Speech and Hearing Research, 29, 366-374.
De Villiers, .I., & De Villiers, P. (1973). Development of the use of word order in comprehension. Journal of Psycholinguistic Research, 2, 331-342.
Dowty, D. (1991). Thematic proto-roles and argument structure. Language, 67, 547-619. Dunn, L., Dunn, L., Whetton, C., & Pintillie, D. (1982). The British Picture Vocabulary Scales.
Windsor: NFER-Nelson.
Elliott, C., Murray, D., & Pearson, L. (1978). British Ability Scales. Windsor: NFER-Nelson.
Fillmore, J. (1968). The case for case. In E. Bach & R. Harms (Eds.), Universals in linguistic theory. New York: Rinehart & Winston.
Fletcher, P., Ingham, R., & Kirby, G. (1992). The grammatical characterisation of specific language
impairment: theoretical and methodological issues. In I. Warburton & R. Ingham (Eds.), Working papers in general and applied linguistics (pp. 112-131). University of Reading.
Foley, W., & Van Valin, R. (1984). Functional syntax and universal grammar. Cambridge, UK:
Cambridge University Press. Gleitman, L. (1989). The structural sources of verb meaning. Papers and Reports on Child Language
Development, 28, l-48. Gleitman, L. (1990). The structural sources of verb meaning. Language Acquisition, I, l-55. Golinkoff, R., Hirsh-Pasek, K., Cauley, K., & Gordon, L. (1987). The eyes have it: lexical and
syntactic comprehension in a new paradigm. Journal of Child Language, 14, 23-45. Gopnik, M. (1990). Feature blindness: a case study. Language Acquisition, 1, 139-164. Gopnik, M., & Crago, M. (1991). F amilial aggregation of a developmental language disorder.
Cognition, 39, l-50.
Grimshaw, J. (1981). Form, function and the language acquisition device. In C. Baker & J. McCarthy
(Eds.), The logical problem of language acquisition. Cambridge, MA: MIT Press.
Grimshaw, J. (1987). Unaccusatives: an overview. Proceedings of the North-East Linguistics Society, 17.
Gropen, J., Pinker, S., Hollander, M., Goldberg, R., & Wilson, R. (1989). The learnability and
acquisition of the dative alternation in English. Language, 65, 203-257. Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991a). Syntax and semantics in the
acquisition of locative verbs. Journal of Child Language, 18, 115-1.51. Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991b). Affectedness and direct objects: the
role of lexical semantics in the acquisition of verb argument structure. Cognition, 41, 153-196. Harris, D. (1963). The Goodenough Harris Drawing Test. New York: Harcourt Brace Javanovich.
Howell, D. (1987). Statistical methods for psychology. (2nd ed.). Boston, MA: Duxbury Press.
Jackendoff, R.S. (1972). Semantic interpretation in generative grammar. Cambridge, MA: MIT Press. Jackendoff, J. (1987). The status of thematic relations in linguistic theory. Linguistic Inquiry, 18,
369-411. Jackendoff, R.S. (1989). The relationship of syntax and conceptual structure. Nijmegen lectures,
Max-Planck Institute of Psycholinguistics, Nijmegen, December 1989.
Kamhi, A., Catts, H., Koenig, L., & Lewis, B. (1984). Hypothesis testing and non-linguistic symbolic
abilities in language impaired children. Journal of Speech and Hearing Disorders, 49, 169-176. Kirk, S., McCarthy, J., & Kirk, W.C. (1968). Illinois rest of psycholinguistic abilities. Urbana:
University of Illinois Press. Landau, B., & Gleitman, L. (1985). Language and Experience. Cambridge, MA: Harvard University
Press.
Leonard, L.B. (1982). Phonological deficits in children with developmental language impairment.
Brain and Language, 16, 73-86. Leonard, L.B. (1989). Language learnability and specific language impairment in children. Applied
Psycholinguistics, 10, 179-202. Leonard, L. (1991). Specific language impairment in three languages: some cross-linguistic evidence.
H.K.J. van der Lely I Cognition 51 (1994) 29-72 69
Paper presented at The second international symposium: Specific speech and language disorders in children. Harrogate, May 1991. London: Afasic.
Leonard, L., Bortolini, U., Caselli, M., & McGregor, K. (in press). Morphological deficits in children
with specific language impairment: the status of features in the underlying grammar. Language Acquisition.
Levin, B. (1985). Lexical semantics in review: an introduction. In B. Levin (Ed.), Lexical semantics in review. Lexicon Project Wording Papers, 1. Cambridge, MA: MIT Press.
Naigles, L. (1990). Children use syntax to learn verb meanings. Journal of Child Language, 17, 357-374.
Naigles, L., Gleitman, H., & Gleitman, L. (1992). Syntactic bootstrapping in verb acquisition:
evidence from comprehension. In E. Dromi (Ed.), Language and cognition: a developmental perspective. NJ: Ablex.
Naigles, L., & Kako, E. (in press). First contact in verb acquisition: defining a role for syntax. Child Development.
Pinker, S. (1984). Language learnability and language development. Cambridge, MA: Harvard
University Press. Pinker, S. (in press). How could a child use verb syntax to learn verb semantics? Lingua. Cambridge,
MA: MIT Press.
Pinker, S., Lebeaux, D.S., & Frost, L. (1987). Productivity and constraints in the acquisition of the
passive. Cognition, 26, 195-267. Precious, A., & Conti-Ramsden, G. (1988). Language-impaired children’s comprehension of active
versus passive sentences. British Journal of Disorders of Communication, 23, 229-244. Rappaport, M., & Levitt, B. (1988). What to do with theta roles. In W. Wilkins (Ed.), Thematic
relations. New York: Academic Press.
Reynell, J. (1977). Reynell Developmental Language Scales. (revised ed.). Windsor: NFER-Nelson.
Satterthwaite, F.E. (1946). An approximate distribution of estimates of variance components.
Biometrics Bulletin, 2, 110-l 14.
Stark, R., & Tallal, P. (1981). Selection of children with specific language deficits. Journal of Speech and Hearing Disorders, 46, 114-122.
van der Lely, H.K.J. (1990). Sentence comprehension processes in specifically language impaired children. Unpublished Ph.D. thesis, University of London.
van der Lely, H.K.J., & Dewart, M.H. (1986). Sentence comprehension strategies in specifically
language impaired children. British Journal of Disorders of Communication, 21, 291-306. van der Lely, H.K.J., & Harris, M. (1990). Comprehension of reversible sentences in specifically
language-impaired children. Journal of Speech and Hearing Disorders, 55, 101-117. van der Lely, H.K.J., & Howard, D. (1993). Specifically language impaired children: linguistic
impairment or short term memory deficit. Journal of Speech and Hearing Research, 36 (in press).
Wexler, K., & Manzini, R. (1987). Parameters and learnability in binding. In T. Roeper & E.
Williams (Eds.), Parameter setting. Riedel: Dordrecht.
70 H.K.J. van der Lely I Cognition 51 (1994) 29-72
Appendix A: Raw scores for the language tests for the SLI children and LA
controls
Chronological Mean Language tests
age (months) language
RDLS BPVS ITPA BAS -
age
SLI children MP 114 77
oc 78 70
SJ 84 70
AS 80 67
AW 73 56
DL 88 43
Mean
SD
LA controls MPl 77 75 49 18 17
MP2 76 75 57 19 16
MP3 66 78 68 26 16
OCl 66 70 55 22 14
oc2 65 70 54 20 15
OC3/SJl 76 67 53 16 17
SJ2 71 69 53 14 16
SJ3 66 73 56 18 16
AS1 62 64 43 13 16
AS2 67 67 51 19 15
AS3 66 63 50 17 14
Awl 56 56 36 16 15
AW2 56 52 36 11 13
AW3 63 59 41 13 15
DLl 40 37.5 25 6 9
DL2 40 47.5 25 11 13
DL3 41 43.5 28 8 10
Mean 45.88 15.71 14.52
SD 12.34 5.08 2.24
62 68
64 49
64 51
63 49
62 41
46 23
60.17 46.83
6.98 14.68
22 15
15 16
10 16
21 13
13 13
8 11
14.83
5.71
14
2.0
Note: RDLS = Reynell Developmental Language Scales. BPVS = British Picture
Vocabulary Scale. ITPA = Grammatical Closure subtest from the Illinois Test of
Psycholinguistic Abilities. BAS = Expressive vocabulary from the British Ability
Scale.
H.K.J. van der Lely I Cognition 51 (1994) 29-72 71
Appendix B: Test sentences used in Experiment 3
Active 1. The man yols the lady.
2. The girl wooks the boy.
3. The boy voozes the girl.
4. The lady faps the man.
5. The car lats the lorry.
6. The car neeks the train.
7. The lorry gams the car.
8. The car sarts the lorry.
Passive
9. The man is yolled by the lady.
10. The girl is wooked by the boy.
11. The boy is voozed by the girl.
12. The lady is fapped by the man.
13. The car is latted by the lorry.
14. The car is neeked by the train.
15. The lorry is gammed by the car.
16. The car is sarted by the lorry.
Locative
17. The man jids the girl to the lady.
18. The girl mips the boy to the man.
19. The lady pons the man to the girl.
20. The boy wids the girl to the man.
21. The train shogs the car to the lorry.
22. The car rits the train to the lorry.
23. The lorry tugs the car to the train.
24. The train foons the lorry to the car.
Appendix C: Experiment 1: Forward linking. Examples of the responses for
three SLI children and three LA control children
Verb: Tiving
Demonstration by experimenter: The girl bounced the boy on his head whilst
holding him upside down.
Question 1. Tell melthe puppet what’s happening
Child Response
MP Tiving
AW They’re both tiving
72 H.K.J. van der Lely I Cognition 51 (1994) 29-72
DL She’s tiving
OCl The girl’s tivined the boy
SJl This is . . . it’s tiving
AS1 The boy’s been tived by the girl.
Question 2. Tell melthe puppet about the girl (agent)
Child Response MP The little girl is tiving . . . the boy
AW The girl is carrying the boy
DL Tiving him
OCl The girl’s holding the boy
SJl She’s tiving the boy
AS1 The girl’s been tiving the boy
Question 3. Tell melthe puppet about the boy (theme)
Child Response MP The boy is tiving the girl
AW The boy’s got tiving by the girl
DL The boys’ hurting his head
OCl The girl’s lifting up the boy . . . upside down
SJl The boy’s being tived by the girl
AS1 The boy’s been tived by the girl
Appendix D: Experiment 3: Reverse linking. Examples of the acting out
responses to one transitive active test sentence by three SLI children and three
LA control children
Test sentence: The man yols the lady
Child MP
AW
DL
OCl
SJl
AS1
Response The man and lady were picked up. The lady was made to push
the man over.
The lady was picked up and made to hit the man.
The man was picked up, followed by the lady. The man was
made to stand beside the lady. The lady fell over.
The man is made to turn the lady over, and then he jumps up
and down on her back.
The man is made to carry the lady.
The man and lady are picked up. The lady’s feet are attached
around the man’s neck. The man is made to walk around the
table carrying the woman.