Past tense formation in Williams syndrome

LANGUAGE AND COGNITIVE PROCESSES, 2001, 16 (2/3), 143–176
Past tense formation in Williams syndrome
Michael S.C. Thomas, Julia Grant, Zita Barham,
Marisa Gso¨dl, Emma Laing and Laura Lakusta
Neurocognitive Development Unit, Institute of Child Health, London, UK
Lorraine K. Tyler
Department of Experimental Psychology, University of Cambridge,
Cambridge, UK
Sarah Grice, Sarah Paterson and Annette Karmiloff-Smith
Neurocognitive Development Unit, Institute of Child Health, London, UK
It has been claimed that in the language systems of people with Williams
syndrome (WS), syntax is intact but lexical memory is impaired. Evidence
has come from past tense elicitation tasks with a small number of participants
where individuals with WS are said to have a speci c de cit in forming
irregular past tenses. However, typically developing children also show
poorer performance on irregulars than regulars in these tasks, and one of the
central features of WS language development is that it is delayed. We
compared the performance of 21 participants with WS on two past tense
elicitation tasks with that of four typically developing control groups, at ages
6, 8, 10, and adult. When verbal mental age was controlled for, participants in
the WS group displayed no selective de cit in irregular past tense
performance. However, there was evidence for lower levels of generalisation
to novel strings. This is consistent with the hypothesis that the WS language
system is delayed because it has developed under different constraints,
constraints that perhaps include atypical phonological representations. The
Requests for reprints should be addressed to Michael Thomas or Annette Karmiloff-Smith,
Neurocognitive Development Unit, Institute of Child Health, 30, Guilford Street, London
WC1N 1EH, UK. Email: M.Thomas@ich.ucl.ac.uk or a.karmiloff-smith@ich.ucl.ac.uk
We would like to express our appreciation to the Williams Syndrome Foundation, UK, for
their generous help in putting us in touch with families whom we warmly thank for their
participation in this research. Thanks also to Dorothy Bishop and Marc Joanisse for helpful
comments on an earlier draft on this paper. This research was supported by MRC Programme
Grant No. G9715642 to Annette Karmiloff-Smith.
®c 2001 Psychology Press Ltd
http://www.tandf.co.uk/journals/pp/01690965.html
DOI: 10.1080/01690960042000021

---

144
THOMAS ET AL.
results are discussed in relation to dual-mechanism and connectionist
computational models of language development, and to the possible
differential weight given to phonology versus semantics in WS development.
INTRODUCTION
Williams syndrome (WS) is a rare neurodevelopmental disorder occurring
in approximately 1 in 20,000 live births (Morris, Demsey, Leonard, Dilts, &
Blackburn, 1988). It is caused by a micro-deletion on one copy of
chromosome 7 (Tassabehji et al., 1999) and results in speci c physical,
cognitive, and behavioural abnormalities (Karmiloff-Smith, 1998; Mervis,
Morris, Bertrand & Robinson, 1999). The syndrome has been of particular
interest to cognitive scientists because individuals with WS exhibit an
uneven cognitive-linguistic pro le together with mild to moderate mental
retardation (Howlin, Davies, & Udwin, 1998; Mervis et al., 1999). Thus
Udwin and Yule (1990) found that 54% of their sample of 43 WS
participants had a full-scale intelligence quotient (IQ) of µ 50 and 42%
had an IQ between 51–70. However, in general the full-scale IQ score in
WS masks differences in speci c cognitive abilities. The syndrome is often
characterised as one where verbal abilities are superior to visuospatial
abilities (Mervis et al., 1999), although in both areas performance is below
that expected for chronological age. This pattern of uneven abilities may
be one that emerges and increases over the course of development
(Jarrold, Baddeley, & Hewes, 1998; Bellugi, Lichtenberger, Mills,
Galaburda, & Korenberg, 1999). The uneven pro le extends to other
abilities. Thus while individuals with WS often perform within the normal
range on standardised tests for face recognition (Bellugi, Wang, &
Jernigan, 1994), and show relatively good performance on theory of mind
tasks (Karmiloff-Smith, Klima, Bellugi, Grant, & Baron-Cohen, 1995),
they exhibit dif culties in numerical cognition (Karmiloff-Smith et al.,
1995), and in problem solving and planning (Bellugi, Marks, Bihrle, &
Sabo, 1988).
The uneven cognitive pro le found in WS has been of interest because it
promises to offer the potential to identify developmental fractionations in
the cognitive system. For example, given limitations in general cognition,
the largely successful acquisition of language might be taken as evidence of
the developmental independence of language from cognition (see Mervis
& Bertrand, 1997; Rossen, Bihrle, Klima, Bellugi, & Jones, 1996). A
similar argument might be made for the developmental independence of
face recognition from spatial cognition. Given the standard assumption
that the adult cognitive system has a modular structure and that WS has a
genetic origin, there is an additional temptation to link dissociations in the
cognitive abilities of adults with WS with damage to or sparing of innate
cognitive modules. This approach attempts to extend the logic of adult

---

WILLIAMS SYNDROME AND PAST TENSE FORMATION
145
neuropsychology in which patterns of adult brain damage are taken to
reveal (under some circumstances) the functional modules comprising the
cognitive system. When extended to developmental disorders that have a
genetic basis, the implication is that de cits in the endstate behaviour of
individuals will reveal the innate modular structure of the cognitive system
(see e.g., Baron-Cohen, 1998; Temple, 1997). In this paper we will seek to
question whether the adult brain damage model is indeed appropriate for
characterising behavioural de cits found in developmental disorders. To
do so, we will examine a speci c example, that of the acquisition of past
tense formation in Williams syndrome.
WS and SLI: a double dissociation of innate
mechanisms?
Williams syndrome has been used to support the presence of innate
structure in the normal language system. This innate structure supposes the
existence of two sorts of mechanism, a computational, syntactic, rule-based
mechanism responsible for learning the abstract rules of grammar, and an
associative memory system responsible for learning information about
individual words (Pinker, 1991, 1994, 1999). We will refer to this as the
dual-mechanism account, by which we speci cally mean a model with one
rule-based mechanism and one associative mechanism. (It is of course
possible to have dual-mechanism accounts where both mechanisms are
rule-based or both are associative. Debates about the quantity of
mechanisms are orthogonal to those about the nature of those mechan-
isms.) Pinker (1991) proposed that Speci c Language Impairment (SLI)
and Williams syndrome together provide a developmental double
dissociation between these two language mechanisms. SLI is a develop-
mental disorder in which impairments are found in language in the absence
of any apparent cognitive, social, or neurological de cits. In addition, there
is a genetic component to this disorder (Bishop, North, & Donlan, 1995).
Referring to evidence from Gopnik and Crago (1991), Pinker proposed
that people with SLI have an impairment to the syntactic, rule-based
device, but that their ability to memorise words is intact. Citing evidence
from Bellugi, Bihrle, Jernigan, Trauner, and Doherty (1990), he further
proposed that in Williams syndrome, there is a ‘‘selective sparing of syntax,
and grammatical abilities are close to normal in controlled testing’’
(p. 479), but that there is an impairment to the associative memory
mechanism such that individuals ‘‘retrieve words in a deviant fashion’’
(ibid.). In short, we have the claim that the two mechanisms can be
dissociated because they can independently fail in two distinct develop-
mental disorders, forming, as Pinker describes it, a ‘‘genetic double
dissociation’’ (1999, p. 262).

---

146
THOMAS ET AL.
Much of the behavioural evidence behind this proposal comes from
performance on forming the English past tense. The English past tense is
characterised by a rule in which the past tense of a verb is formed by
adding the suf x -ed to the verb stem (e.g., talk-talked). However, there is
also a minority of verbs which form their past tense in different ways (e.g.,
go-went, think-thought, hit-hit). These irregular or exception verbs often
fall into clusters sharing a family resemblance (e.g., sleep-slept, creep-
crept, leap-leapt). The English past tense is important for Pinker’s dual-
mechanism theory, since performance on the regular and irregular past
tense formations are taken to directly index, respectively, the functioning
of the rule-based and associative mechanisms. Pinker’s claims about SLI
and WS then translate into the following empirical predictions: (1) we
should expect individuals with SLI to show a selective de cit in forming
regular past tenses but not irregular past tenses; (2) we should expect
individuals with WS to show a selective de cit in forming irregular past
tenses but not regular past tenses. Recent work has sought to address these
claims in detail.
SLI: de® cit on regulars but not irregulars?
Van der Lely and Ullman (this issue) have examined English past tense
formation in a sample of children with ‘‘grammatical’’ SLI. SLI is a
heterogeneous disorder (Aram, Morris, & Hall, 1993) which may have a
number of underlying causes. Van der Lely and Stollwerck (e.g., van der
Lely, 1997; van der Lely & Stollwerck, 1996) have identi ed a subgroup of
children with SLI based on behavioural measures, such that their
predominant de cit is restricted to grammatical abilities. Van der Lely
claims that, at least for this subgroup, their disorder can be characterised as
a ‘‘primary de cit in the computational syntactic (grammatical) system’’
(van der Lely, 1998). Van der Lely and Ullman found that in a past tense
elicitation task, the children with SLI predominantly responded by
reproducing the stem without marking it, accounting for approximately
65% of all responses. In terms of correct performance, the children with
SLI showed no advantage of regular over irregular verbs which, compared
to controls, represented a greater de cit on regulars than irregulars. Lastly,
they found frequency effects in the performance of the SLI group on
regular verbs, an affect normally con ned to irregular verbs. On the
assumption that Pinker’s dual-mechanism model is correct, van der Lely
and Ullman took these results as supporting the view that in grammatical
SLI, the rule-based mechanism is impaired but the associative memory is
intact. Although the children with SLI provided some correct regular past
tense items, these were taken as re ecting compensatory activity of the
associative memory. Frequency effects are taken as a hallmark of such an

---

WILLIAMS SYNDROME AND PAST TENSE FORMATION
147
associative system. Van der Lely and Ullman thus interpreted the
frequency effects found in regular past tense formation as an indication
that, in the absence of a rule-based mechanism, all past tenses were being
treated as exceptions (see also Ullman & Gopnik, 1999).
Williams syndrome: de® cit on irregulars but not
regulars?
Clahsen and Almazan (1998) recently examined the performance of four
children with WS (aged 11;2 to 15;4) on a range of grammatical tasks.
These included an analysis of expressive language in story telling, a test of
comprehension of active and passive sentences, a test of the comprehen-
sion of sentences using syntactic binding in referential dependencies
between anaphoric elements, a test of in ection morphology (English past
tense formation), and a test of derivational morphology (past tense
formation for normal and denominal irregular verbs). The analysis of
expressive language showed that the performance of the WS group was
appropriate for their mental age (as measured by their overall scores on
the Wechsler Intelligence Scale for Children-III; Wechsler, 1992), and that
their language comprised complex syntactic structures and grammatical
morphemes that were almost always correct. Performance of participants
with WS on the particular tests used for passives and syntactic binding was
at ceiling.
Clahsen and Almazan used the same past tense elicitation procedure as
van der Lely and Ullman so the results are directly comparable. Their
results pointed to a selective de cit in irregular past tense formation in two
individuals with WS with mental ages (MA) of 5 years and two individuals
with WS with MA of 7 years compared to MA-matched control groups.
They concluded that the individuals with WS had an impaired associative
memory mechanism, citing as evidence the fact these participants
irregularised novel verbs which rhymed with existing irregular verbs
(e.g., crive-crove, drive-drove) at a much lower rate than their controls.
Thus participants with WS ‘‘seemed to be impaired (relative to controls) in
associating phonological patterns of novel verbs to corresponding strings
of existing irregular verbs’’ (p. 193). On the assumption that Pinker’s dual-
mechanism model is correct, Clahsen and Almazan concluded that in
Williams syndrome, the ‘‘computational system for language is selectively
spared yielding excellent performance on syntactic tasks and on regular
in ection, whereas the lexical system and/or its access mechanisms
required for irregular in ection are impaired’’ (ibid.). Their results on
in ectional morphology in WS are in line with previous unpublished data
for six participants with WS presented by Bromberg, Ullman, Coppola,
Marcus, Kelley, and Levin (1994).

---

148
THOMAS ET AL.
Problems with existing WS past tense data
There are two serious problems with the current data on in ection
morphology in Williams syndrome. Firstly, typically developing children
usually show poorer performance on irregular verbs than regular verbs
(with the exception of the very early stages of language development
where vocabulary size is small) (Bybee and Slobin, 1982; see also van der
Lely & Ullman, this issue). One of the most salient characteristics of
language development in Williams syndrome is that it is delayed (Mervis
et al., 1999; Singer Harris, Bellugi, Bates, Jones, & Rossen, 1997; Thal,
Bates, & Bellugi, 1989). Therefore, to show a selective de cit in irregular
past tense formation in individuals with WS, it is not enough to
demonstrate that irregular past tense formation is poorer than regular
past tense formation. Rather, it must be shown that their level of past
tense formation is poorer than we would expect given their level of
language development. In the unpublished data of Bromberg et al. (1994),
no such comparison is possible since participants with WS were only
roughly matched to normal controls. While the comparison is possible for
the Clahsen and Almazan data, their study only comprised four
individuals with WS, and even for these, the data appear fairly noisy.
For example, for irregular verbs, the two individuals with MA of 5 scored
14% correct on irregular verbs compared to the 57% correct scored by
the two individuals with MA of 7. And when performance on irregulars
was re-tested as a control condition in the derivational morphology task,
the MA-5 individuals now scored 44% correct. On the evidence of this
study alone, one cannot be con dent that the apparent de cit on irregular
verbs in the WS group is any more than a consequence of delayed
language development.
Secondly, Clahsen and Almazan note a marked difference between the
WS and control groups in how willing they were to extend patterns of
irregular past tense formation to novel items (e.g., crive-crove). Levels of
irregularisation were much lower in the WS group and they took this as
revealing an impairment to lexical associative memory. However, the
control data Clahsen and Almazan used in this comparison look very
different to those collected by van der Lely and Ullman (2000) on exactly
the same task. Clahsen and Almazan’s two control groups irregularised
novel rhymes at rates of 68% and 75%. Van der Lely and Ullman’s groups,
of a similar age, irregularised novel rhymes at levels of 10%, 9%, and 10%.
These latter levels are much closer to the rates that Clahsen and Almazan
reported for their WS group. Thus the apparent de cit shown by the WS
group would seem to depend crucially on the true level of novel
irregularisation in the normal population at an equivalent level of
language development.

---

WILLIAMS SYNDROME AND PAST TENSE FORMATION
149
In our study, we set out to rectify these problems in order to establish
whether the performance of individuals with WS in irregular past tense
formation is indeed reliably poorer than would be expected for their level
of language development. We did this in three ways. Firstly, we examined a
much larger sample of participants with WS than the Clahsen and
Almazan study. Secondly, we sought to build a normal developmental
pro le of performance on this particular task against which we could
compare the performance of the WS group. To do so, we tested four
groups of control participants, aged 6, 8, 10, and adult. Thirdly, we
employed an additional past tense elicitation task to explore whether any
features of the Clahsen and Almazan results were due to particular
features of the task they used.
The second elicitation task was developed for use with patients with
brain damage by Lorraine Tyler and William Marslen-Wilson. This task
does not require participants to repeat sentences and instead provides
them with the initial sound of the past tense form. Consequently, it may
be seen as having a lower memory load. In addition, it employed a set of
regular and irregular verbs three times as large as that used in the
Clahsen and Almazan (1998) study. This larger set of verbs allowed us to
explore underlying factors in the elicitation task, such as the role of verb
frequency and verb imageability in past tense formation. We have already
seen that frequency effects have been taken as a hallmark of lexical
associative processing. Evidence of effects of imageability, a semantic
dimension differing across verbs, could also be taken to implicate lexical
memory in the operation of the grammatical process of past tense
formation.
Two contrasting hypotheses were tested in the current study. The rst
represents the Pinker/Clahsen and Almazan position: Individuals with
Williams syndrome show a speci c de cit in irregular past tense formation.
Thus if one controls for language ability, one should expect performance
on regulars to be the same for the WS and control groups. On the other
hand, one should expect performance on irregular verbs to be lower for the
WS group than the control group. With novel items, one should expect
performance on regularising novel words which do not rhyme with any
existing irregular verb to be the same as controls (e.g., stoff-stoffed). On
the other hand, one should expect performance on irregular rhyming novel
verbs to be different from controls. Perhaps the WS group might show less
irregularisation (e.g., crive-crove), in keeping with the hypothesis of an
impaired associative mechanism, or more regularisation, in keeping with
the hypothesis of a preserved rule-based mechanism (e.g., crive-crived).
The alternative hypothesis suggests that poor performance on irregular
past tense formation in WS is a marker of their delayed language
development. If one controls for level of language ability, performance on

---

150
THOMAS ET AL.
regulars and irregulars should be the same in participants with WS and
controls. One should nd a similar pattern in performance on novel items.
METHOD
Participants were tested on two tasks, both of which were designed to elicit
past tense verb forms but which imposed somewhat different demands on
memory. Task 1 was adapted from Ullman (1993), Ullman et al. (1997),
and Clahsen and Almazan (1998). Task 2 was developed by Tyler and
Marslen-Wilson.
Participants
Twenty-one children and adults with WS, 12 male and 9 female, were
recruited through the Williams Syndrome Foundation UK to take part in
this and other studies. Mean chronological age was 22;8 (range 10;11–53;3).
Mean General Cognitive Ability (GCA; IQ equivalent as assessed by the
British Abilities Scale II) was 45, (range 39 ( oor)–73).
Three groups of typically developing children were also tested, with ve
boys and ve girls in each group. Their mean ages were as follows: 5–6-
year-olds ˆ 6;0 (range 5;5–6;40), 7–8-year-olds ˆ 8;1 (range 7;8–8;5), 9–10-
year-olds ˆ 9;10 (range 9;6–10;6). These children attended a North London
primary school. A group of 16 normal adult controls were recruited by
means of notices placed at Great Ormond Street Hospital and in a local
community centre. Ten males and six females took part in the study, with a
mean age of 30;5 (range 17;3–45;0). Participants in all groups were drawn
from a range of socio-economic classes.
Materials
Task 1. Fifty-six sentence pairs were constructed according to the form
illustrated in the following two examples:
(1) Every day I slam a door
Just like every day, yesterday I .............. a door
(2) Every day I swim in the pool
Just like every day, yesterday I .............. in the pool
The verbs in these sentences were those used by Clahsen and Almazan
(1998). Existing regular and irregular verbs were matched for frequency
and familiarity (see van der Lely & Ullman, this issue). The stimulus set is
shown in Table 1. It included 16 existing regular verbs, 14 existing irregular
verbs, 12 novel verbs with stems which did not rhyme with any existing
irregular verbs, and 14 novel verbs which rhymed with existing irregulars.

---

WILLIAMS SYNDROME AND PAST TENSE FORMATION
151
TABLE 1
Stimulus sets for the two past tense elicitation tasks
Task 1
Task 2
Irregular-
Regular
Irregular
Non-rhyme
rhyme novel
verbs
verbs
novel items
items*
Regular verbs
Irregular verbs
scowl
swim
spuff
strink (strunk)
kick
laugh
stick
shrink
tug
dig
dotch
frink (frunk)
croak
help
creep
sing
ush
swing
stoff
strise (strose)
climb
mix
mislead
draw
mar
wring
cug
crive (crove)
stay
shave
shake
learn
chop
bend
trab
shrell (shrelt)
balance
agree
deal
keep
ap
bite
crog
vurn (vurnt)
dance
drag
begin
meet
stalk
feed
vask
steeze (stoze)
trim
leak
bleed
come
scour
make
brop
shrim (shram)
chase
stop
choose
grow
slam
give
satch
cleed (cled)
graze
call
leap
ring
cross
think
grush
sheel (shelt)
share
raise
cling
dream
rush
stand
plam
blide (blid)
walk
move
sting
shine
rob
keep
scur
prend (prent)
x
shove
hang
lose
drop
drive
shreep (shrept)
bless
save
weep
drink
look
send
drite (drote)
feed
stir
soar
* possible irregularisation shown in parentheses.
In order to optimise the enunciation and audibility of past tense endings
produced by participants, each verb was followed by a noun phrase or
prepositional phrase whose rst word began with a vowel (the past tense
verb ending in a sentence such as Yesterday I robbed a bank, whose verb is
followed by a word starting with a vowel, is often more full articulated and/
or more audible than in a sentence such as Yesterday I robbed the bank,
where a consonant follows the verb ending).
Fourteen pairs of practice sentences were also constructed, using the
same format as the test sentences. These incorporated six irregular verbs,
four regular verbs, and four novel verbs with stems which did not rhyme
with any existing irregular verbs.
Task 2. In this task, participants received the initial phoneme of the
past tense form as a cue. The stimuli consisted of 53 sentences, each paired
with an incomplete sentence, for example:
(1) The bull sometimes kicks.
Yesterday, it k..........
(2) Maggie always hangs the pictures.
Last time, she h.........

---

152
THOMAS ET AL.
Twenty-six regular verbs and 27 irregular verbs were used in the test
sentences. These were matched for frequency and imageability. Two
practice sentences were constructed using irregular verbs.
Standardised tests
The participants with WS were also tested on the British Picture
Vocabulary Scale (BPVS; Dunn et al., 1982) and seven subtests of the
British Abilities Scales II (BAS–II; Elliott, 1996), namely Recall of
Designs, Pattern Construction, Word De nitions, Verbal Similarities,
Matrices, Quantitative Reasoning and Recall of Digits Forward. Table 2
shows the individual participant scores for chronological age, BPVS,
General Cognitive Ability (GCA) from BAS (a composite score based on
TABLE 2
Ages and standardised test results of participants with Williams syndrome
BAS General
Verbal MA Spatial MA Non-verbal
Chronological
Cognitive
BPVS test
(BAS
(BAS
MA (BAS
Subject
Age
Ability
age
subtests)
subtests)
subtests)
(Floor 1;8,
(Floor 5;0,
(Floor 5;0,
(Floor 5;0,
(Floor 39)
ceiling 19;6) ceiling 18;0) ceiling 18;0) ceiling 18;0)
1*
10;11
44
5;2
5;10
5;0
5;4
2
11;1
48
6;8
6;9
5;0
6;2
3
11;3
46
9;3
7;11
5;10
8;2
4
11;5
44
5;3
5;7
5;0
6;2
5
11;7
47
9;0
7;0
5;1
7;0
6*
12;6
39
5;2
5;0
5;0
5;0
7
12;9
54
8;1
8;6
5;4
7;3
8
13;11
46
5;5
7;3
5;1
7;4
9
14;4
41
8;0
7;9
5;4
6;4
10
15;6
39
8;1
6;9
5;4
5;6
11
18;7
40
8;9
6;6
6;9
6;9
12*
19;3
39
5;0
5;0
5;0
5;0
13
20;10
39
8;4
6;10
5;0
5;2
14
21;8
39
7;1
6;12
5;0
5;0
15
27;6
39
8;8
6;10
5;6
5;0
16
30;3
39
15;4
7;4
5;2
6;7
17
30;8
51
15;7
13;9
6;7
7;2
18
34;9
39
7;10
6;4
5;0
5;0
19
42;9
73
19;6
16;5
8;9
10;6
20
50;11
50
13;11
13;0
5;4
7;11
21
53;3
39
14;3
9;0
5;2
6;1
Mean
22;8
44
9;7
7;11
6;5
5;6
* Starred participants were unable to complete the past tense elicitation tasks.
BAS, British Abilities Scale; BPVS, British Picture Vocabulary Scale; MA, mental age

---