A study of relative clauses in Williams syndrome

J. Child Lang.  (), –.
#  Cambridge University Press
DOI : .\SX
Printed in the United Kingdom
N O T E
A study of relative clauses in Williams syndrome*
J U L I A G R A N T
Neurocognitive Development Unit, Institute of Child Health, London
V I R G I N I A V A L I A N
Hunter College and CUNY Graduate Center, New York
A N N E T T E K A R M I L O F F - S M I T H
Neurocognitive Development Unit, Institute of Child Health, London
(Received  August . Revised  March )
       
Despite growing empirical evidence to the contrary, claims continue to
be made that the grammar of people with Williams syndrome (WS) is
intact. We show that even in a simple elicited imitation task examining
the syntax of relative clauses, older children and adults with WS (n l
, mean age l  ;  years) only reach the level of typical five-year-old
controls. When tested systematically in a number of different labora-
tories, all aspects of WS language show delay and\or deviance through-
out development. We conclude that the grammatical abilities of people
with WS should be described in terms of relative rather than absolute
proficiency, and that the syndrome should no longer be used to bolster
claims about the existence of independently functioning, innately
specified modules in the human brain.
           
The myth of normal language abilities in people with Williams syndrome
(WS) persists ; Pinker (), for example, makes reference to the ‘ excellent
language skills ’ of people with WS, claiming that their morpho-syntax is
intact. Whilst we agree that WS is a syndrome in which language is
[*] This research was supported by MRC Programme Grant No. G and a grant from
the Williams Syndrome Foundation to A. Karmiloff-Smith. We wish to thank the
Williams Syndrome Foundation for their help in putting us in touch with the participants
and their families, and the participants themselves for their time and enthusiastic co-
operation. Address for correspondence : A. Karmiloff-Smith, Neurocognitive Devel-
opment Unit, Institute of Child Health,  Guilford Street, London, WCN EH, UK.
e-mail : a.karmiloff-smith!ich.ucl.ac.uk


---

     ET AL.
surprisingly good, given the relatively poor non-verbal mental skills, it is in
our view theoretically misleading and empirically inaccurate to claim that
morpho-syntax, semantics, or pragmatics are intact in this clinical population
(Karmiloff-Smith, ). What is true about the syndrome is that older
children, adolescents and adults display in most cases considerably better
scores on language tasks than on non-verbal tasks, and better scores on
language tasks than MA-\CA-matched controls with, say, Down’s syndrome.
Note, however, that these are , not absolute advantages. This fact is
overlooked when WS findings are reported in secondary sources. Indeed,
there are frequent claims in the literature about a total dissociation between
language and cognition in WS and a slippage from ‘ relative ’ proficiencies to
absolute abilities, with the syndrome flagged as the prime example of
preserved language sub-modules in the face of mental retardation (e.g.
Bellugi, Wang & Jernigan,  ; Clahsen & Almazan,  ; Pinker, ).
First, it is worth noting that the non-verbal level of those individuals with
WS who produce particularly good language scores is in no way so low as to
suggest linguistic idiots-savants. Referring to IQ levels in a profile, rather
than MA levels, can paint a very misleading picture. Take an adult with a
verbal IQ of  and a non-verbal IQ of . This suggests a serious
dissociation between language and cognition. However, if one calculates the
same individual’s MA, the result can turn out to be a verbal MA of eleven
years and a non-verbal MA of seven years. It is then far less surprising that
a person with a developmental disorder but an MA of seven years has
relatively fluent language (Karmiloff-Smith, ), since normal children
have fluent language as of about five years of age. A second point worth
noting is the fact that each time an empirical study has actually been made
with a WS population clinically and genetically diagnosed, and of sufficient
number, their language turns out to be either delayed or deviant (Udwin &
Yule,  ; Mervis, Golinkoff & Bertrand,  ; Capirci, Sabbadini &
Volterra,  ; Stevens & Karmiloff-Smith,  ; Volterra, Capirci, Pezzini,
Sabbadini & Vicari,  ; Karmiloff-Smith, Grant, Berthoud, Davies,
Howlin & Udwin,  ; Singer Harris, Bellugi, Bates, Jones & Rossen,
 ; Jarrold, Baddeley & Hewes,  ; Karmiloff-Smith, Tyler, Voice,
Sims, Udwin, Howlin & Davies,  ; Thal, Bates & Bellugi,  ;
Thomas, Grant, Barnham, Gso$dl, Laing, Lakusta, Tyler, Grice, Paterson &
Karmiloff-Smith, ).
The strong claims made about the intactness of morpho-syntax are
sometimes supported by weak evidence. For example, the study of WS past
tense on which Pinker () based his conclusions was a conference poster
(Bromberg, Ullman, Coppola, Marcus, Kelley & Levine, ) which,
because never published, cannot be evaluated. Other work on WS past tense
performance has argued that in WS the computational system for language
is selectively spared compared to lexical look-up (Clahsen & Almazan, ).


---

                       
But that claim was based on a very small sample (N l ), with wide
individual differences. By contrast, a very different picture emerges from the
results of a subsequent study in our laboratory of a much larger population
of children, adolescents and adults with WS (Thomas et al., ). Using the
same past tense task as Clahsen & Almazan, together with a second past tense
elicitation task, we specifically controlled for language level in our WS
population and found no unusual pattern of dissociation between regular and
irregular past tense marking. Our participants looked like much younger,
normal control children. Yet, clear evidence of a dissociation between
computational processes and simple lexical look-up is crucial for the theor-
etical claims being made about Williams syndrome in the scientific and
popular literature.
Our past tense study challenged the notion that aspects of morphology are
differentially spared in WS. However, it remains possible that another aspect
of grammar – syntax – is spared. There are several findings to doubt this,
however. From previous work (Karmiloff-Smith et al., ), we had
ascertained that people with WS scored significantly below their vocabulary
age (and well below their chronological age) on a standardized test of
grammar : the Test of Reception of Grammar (TROG), (Bishop, ). Only
two subjects with WS were at ceiling (and note that ceiling is only  years
on this test), and they turned out to be our most able subjects on non-verbal
tasks also, with chronological ages of  and  years, respectively. However,
one could argue that tasks such as the TROG, in which participants are asked
to select a picture from an array of four that best fits a sentence, have a
substantial metalinguistic component (Tyler, Karmiloff-Smith, Voice, Ste-
vens, Grant, Udwin, Davies & Howlin,  ; Karmiloff-Smith et al., ).
Syntactic capacities might therefore be underestimated by such procedures,
particularly if testing involves atypically developing participants for whom
meta-levels of processing are difficult. Acting out paradigms also present
difficulties to people with learning difficulties. Thus, despite our clear
previous results, it remains possible that people with WS do have intact
syntax, and that the TROG failed to reveal this competence, for reasons of
task demands rather than deficient grammar.
It was to evaluate this hypothesis that the present study was designed.
First, we analysed the error data on the TROG, and picked out a linguistic
structure on which all our WS participants had difficulty : relative clauses.
Note that a testee can be ‘ at ceiling ’ on the TROG but still have failed one
syntactic block, so our two participants at ceiling also displayed difficulties
with relative clauses. We then devised a simple imitation experiment whose
task demands were significantly reduced compared to the picture pointing
task. In this way, we hoped to assess more directly the syntactic competence
of people with Williams syndrome and ascertain whether claims of intact
syntax are correct.


---

     ET AL.
     
Participants
Fourteen children and adults ( female,  male) with Williams syndrome
were recruited through the Williams Syndrome Foundation (UK) as part of
a larger study. Their mean chronological age was  ;  (range  ; – ; ).
Thirty-two typically developing children were also tested. They attended a
North London primary school and were drawn from three age groups, which
will be referred to as the five-, six- and seven-year-old groups. These were
 four- to five-year-olds ( male,  female ; mean age l  ; , range l
 ; – ;  ;  five- to six-year-olds ( male,  female ; mean age l  ; , range
l ;–;); six- to seven-year-olds ( male,  female; mean age l ;,
range l  ; –; ).
All participants were monolingual English speakers and all came from a
similar range of mixed socio-economic backgrounds. Basic data on each
group are provided in Table .
     . Chronological age (CA), and BPVS scores for WS and typically
developing groups
Mean
Mean BPVS
BPVS
vocabulary
Mean
standard
mental
Group
CA (..)
score (..)
age (..)
WS
 ;  ( ; )
n (n)
 ;  ( ; )
(n l )

 ;  ( ; )
n (n)
 ;  ( ; )
(n l )

 ; ( ; )
n (n)
 ;  ( ; )
(n l )

 ;  ( ; )
n (n)
 ;  ( ; )
(n l )
All participants were tested on the British Picture Vocabulary Scale
(BPVS) (Dunn, Dunn & Whetton, ), which is the British equivalent of
the Peabody Picture Vocabulary Test – Revised (Dunn & Dunn, ). The
BPVS standard score is a derived measure that expresses the extent to which
a participant’s score is above or below the mean score for people in the same
chronological age group. Like some IQ tests (e.g. Wechsler Adult In-
telligence Scale, ), the BPVS uses a mean of  and a standard deviation
of . BPVS vocabulary mental ages are also presented in the table. As can
be seen, the WS participants’ mean vocabulary mental age was higher (M l
 ; ) than that of the oldest control children, the seven-year-olds (M l  ; ).


---

                       
This allowed us to be confident that any sub-par performance by WS
participants was not a function of a reduced vocabulary. The WS participants
were also tested on the Test of Reception of Grammar (TROG) (Bishop,
). Their mean grammatical age was  ;  (.. l  ; ).
Materials
We created two sets of test materials : Set A and Set B.
Set A. Set A was composed of four exemplars of each of four types of
complex sentence containing a relative clause. The four types – described
below – were subject–subject (SS), subject–object (SO), object–subject (OS),
and object–object (OO). We have used this terminology because of its long-
standing use in the literature, even though most current linguistic analyses
would not represent the relative clause as containing either the subject or
object of the main clause. Instead, they would view the subject or object
position as containing PRO linked co-referentially with the main clause
subject or object.
Because we wished to make direct comparisons with our previous study of
the TROG, the items for the first three types of relative clause were taken
directly from that test (Bishop, ). We maintained the exact form and
content of those items because, as mentioned above, our previous work
(Karmiloff-Smith et al., ) had indicated that these were among the most
difficult for individuals with WS to comprehend. However, there are certain
linguistic limitations in the design of the TROG items. For example, in the
SS items, the relative clause is non-finite, whereas in each of the other types
the relative clause is finite. There is also a repetition across sentences of some
lexical items and of the copula ‘ to be ’. Nonetheless, for comparative
purposes it was essential to keep as close as possible to the original SS, SO,
and OS items (see examples below). Since the TROG lacks OO items, we
created OO exemplars similar in vocabulary and form to the other TROG
items.
There were four sentence types, with four items per sentence type,
examples of which are given below :
SS : The subject of the main clause is modified ; the subject of the relative
clause is co-referential with the subject of the main clause, e.g. The boy
chasing the horse is fat.
SO : The subject of the main clause is modified ; the direct object or
prepositional object of the relative clause is co-referential with the subject of
the main clause. An example of the former is The cat the cow chases is black ;
an example of the latter is The book the pencil is on is red.
OS : The object (direct or prepositional) of the main clause is modified ; the
subject of the relative clause is co-referential with the object of the main
clause. An example of the former is The dog chases the horse that is brown ; an
example of the latter is The square is in the star that is blue.


---

     ET AL.
OO : The object (direct or prepositional) of the main clause is modified ; the
object of the relative clause is co-referential with the object of the main
clause. An example of the former is The dog is chasing the cow the boy sees ; an
example of the latter is The pencil is on the shoe the girl has.
Nine simple sentences, six words in length, were included as fillers in Set
A (e.g. The ring is on the finger). A single pseudorandom order of filler () and
experimental sentences () was created, with all participants tested on all 
items.
Set B. Early results with Set A suggested that SO items were particularly
difficult to imitate. We thus created a supplementary set (B) of four SO
sentences with the relative pronoun that inserted, as in The cat that the cow
chases is black. A single pseudorandom order of presentation of these four
sentences was used. For Set B, there were no fillers. Eight of the  WS
participants and all the controls were tested with Set B.
Procedure
For Set A, a mixed design was used, with group (four levels) as a between
subjects variable, and a sentence type (four levels) as a within subjects
variable. A mixed design was also used for Set B, with group (four levels) as
a between subjects variable and sentence type (two levels) as a within subjects
variable.
All participants were seen individually in a quiet room. The typically
developing children were tested at their school. Eleven of the  WS
participants were tested in our laboratory ; of the remainder, two were tested
at their school and one at home.
For the typically developing children there was a single test session. Set A
(including filler sentences) was presented first, followed by the BPVS, and
then by Set B. Testing of the WS participants took place over two to four
sessions. The general pattern was for the BPVS and TROG to be admini-
stered during the first one or two sessions as part of a comprehensive
cognitive test battery. Set A of the imitation task was presented at least a
month later. Set B was presented either on a subsequent occasion, or after an
interpolated task between Set A and Set B, as with the typical controls.
Administration of the elicited imitation task followed the same pattern for
both groups. The experimenter said something like ‘ We’re going to play a
game where you have to say what I say. Just copy what I say. Can you say … .
I like ice-cream ? ’ Most participants repeated the practice sentence accurately
but a few of the five-year-olds needed several practice trials to understand the
task. The experimenter then presented Set A in a single pseudorandom
order. Following each response, the experimenter noted whether the par-
ticipant had repeated the sentence correctly and what changes, if any, had
been made. All responses were audiotaped and checked later against the on-
line scoring. Set B was administered in the same way.


---

                       
     . Numbers of OK responses to Set A sentences in each category
Response category
V
Syn
Lex
Mor
Comb
Total OK
Mean (..) OK
SS
WS






n (n)







n (n)







n (n)







n (n)
SO
WS






n (n)







n (n)







n (n)







n (n)
OS
WS






n (n)







n (n)







n (n)







n (n)
OO
WS






n (n)







n (n)







n (n)







n (n)
Sentence types SS, SO, OS and OO are explained in the text. For participants in each group,
N l  for WS, N l  for five-year-olds, N l  for six-year-olds, and N l  for seven-
year-olds. Response category V l verbatim ; Syn l addition of relative pronoun in SO and
OO, addition of relative pronoun plus tensed auxiliary in SS, and substitution of relative
pronoun in OS ; Lex l change of a lexical item ; Mor l change in morphology ; Comb l
combination of any categories except V ; Total OK l total OK responses ; Mean OK l mean
number of OK responses out of  possible.
Scoring
Each response was categorized as OK or as an error. OK responses were
verbatim repetitions of the target sentence, or responses that left the meaning
and essential structure of the sentence unaltered and that remained gram-
matical. There were five subcategories of OK responses.
V. Verbatim responses were word-for-word repetitions of the target.
Syntactic change. Additions or substitutions of a relative pronoun occurred
if the participants added who, which, or that or substituted (in OS sentences)
who or which for that. In SS sentences, the grammatical addition of a relative
pronoun also required the addition of tense, for example changing The boy
chasing the horse is fat to The boy who is chasing the horse is fat. Omission of
that from Set B sentences was also categorized as a syntactic change response.
All of these additions, substitutions and omissions were only categorised as
syntactic if the sentence remained grammatical.


---

     ET AL.
Morphological change. Minor morphological changes, such as pluralising a
single noun, contracting is, or changing the tense or aspect were permitted.
Lexical change. Minor lexical substitutions such as a change of determiner
from the to a, or for instance a noun change from lady to woman were
permitted. Permutating the terms for subject and object or for the two
subjects of the main and relative clauses was not permitted.
Combination. Repetitions which included more than one of the above
permissible changes were placed into this category.
      
To determine whether the four groups differed in their overall ability to
imitate sentences, we compared their performance on the nine filler items.
All participants found the fillers easy to imitate ; most responses were
verbatim or had only minor changes. There were no significant differences
among the groups. Mean OK scores for the nine filler sentences ranged from
n (.. l n) for the WS participants to  (.. l ) for the seven-year-
olds.
In contrast, the groups did differ significantly in their performance on the
test sentences of Set A. Table  shows the distribution of the different
categories of OK responses for each sentence type. The mean OK responses
are graphed in Fig. . A mixed-design MANOVA revealed significant main
4
3·5
3
2·5
SS
SO
2
OS
1·5
OO
Mean OK responses
1
0·5
0
WS
5
6
7
Group
Fig. . Mean number of OK responses by group and sentence type (Set A).
effects of group (F (,) l n, p
n), and sentence type (F (,) l
n, p
n), but no interaction (F (,) l n, ns). Group differences
were further examined using Tukey-HSD tests. Tests for group differences
on individual sentence types showed that the WS group and the six-year-olds


---

                       
4
3·5
3
2·5
SO – that
2
SO + that
1·5
Mean OK responses
1
0·5
0
WS
5
6
7
Group
Fig. . Mean number of OK responses for sentence type SO, without the relative pronoun
(Set A) and with the addition of the relative pronoun (Set B).
differed on SO and OO sentences. The six-year-old participants’ scores for
OS sentences were also significantly higher than those of the five-year-olds
( p
n). No other group differences between scores for individual sentence
types reached significance.
The pattern of responses across the different sentence types was similar in
the WS and the typically developing groups of participants. As Fig.  shows,
the WS group most resembled the five-year-old typical controls. These two
groups were similar in their overall level of performance as well as in their
pattern of responding. This was despite the fact that the WS group had a
BPVS test age ( ; ) and a TROG test age ( ; ) greater than the chronological
age of the five-year-olds.
The SS sentences were easiest for the WS and five-year-old participants,
followed by OO sentences. Both OS sentences and SO sentences were
difficult for both groups, with SO sentences slightly harder. Length did not
account for difficulty. The difficulty order (SO
OS
OO
SS from
hardest to easiest) did not match the length order (OO
OS
SO
SS
from longest to shortest). With the exception of SS sentences, which were
both the easiest and shortest, the orders were reversed. The order of difficulty
of SO, OS and SS sentences for WS participants mirrored that found in their
TROG comprehension data. The percentages of incorrect comprehension
responses were  % for SO sentences,  % for OS sentences and  % for
SS sentences.
A second analysis compared imitations on Set B - SO sentences with the
relative pronoun that – to imitations of SO sentences in Set A that did not
contain a relative pronoun. A mixed design MANOVA had group as a
between-subjects variable and presence or absence of that as a within-


---

     ET AL.
subjects variable. Both main effects and the interaction showed clear trends
although they did not reach conventional levels of significance. The presence
of that tended to help imitation (F(,) l n, p l n). The groups
performed differently overall (F (,) l n, p l n). There was a slight
tendency for an interaction between group and presence of that (F (,) l
n, p l n). As Fig.  shows, the presence of that improved performance,
particularly for the WS group. Tables  and  illustrate that the WS group
     . Numbers of OK responses to SO sentences kthat ( from Set A)
and to SO sentences jthat (Set B) in each category
Response category
SO kthat
V
jRP
Lex
Mor
Total OK
Mean (..) OK
WS





n (n)






n (n)






n (n)






n (n)
SO jthat
V
kRP
Subst.RP
Lex
Mor
Total OK
Mean (..) OK
WS






n (n)







n (n)







n (n)







n (n)
The two SO sentence types are explained in the text. The numbers of participants in each
group are as in Table  except that in the WS group n l . Response category V l verbatim ;
jRP l addition of relative pronoun; kRP l omission of relative pronoun; Subst.RP l
substitution of relative pronoun (what for that) ; Lex l change of a lexical item ; Mor l
change in morphology ; Total OK l total OK responses ; Mean OK l mean number of OK
responses out of  possible.
both added pronouns more frequently for Set A and deleted them less
frequently for Set B than any of the typically developing groups tested. Table
 shows that only one OK response to the Set B sentences in the WS group
involved omission of that ( % of their OK responses), compared with  %
of the responses of the five-year-olds,  % of the six-year-olds and  % of
the seven-year-olds.
Error analysis
An analysis of errors across all the sentence types was then carried out. This
revealed that the WS participants rarely added a relative clause inap-
propriately, although this error was quite common among the six- and seven-
year-olds (n % of the total number of errors made by WS participants,  %
of those made by five-year-olds,  % for six-year-olds and  % for seven-
year-olds). The WS participants also rarely omitted a relative clause
inappropriately (only \ of their errors to OS sentences compared to \


---