Analogical transfer is effective in a serial reaction time task in

~
Pergamon
PII: S0028~3932(96)0005~4
Neuropsycholoqia, Vol. 35, No. I, pp. 1 9, 1997
Copyright ~', 1996 Elsevier Science Ltd. All rights reserved
Printed in Great Britain
0028 3932/97 $17.00+0.00
Analogical transfer is effective in a serial reaction
time task in Parkinson's disease: Evidence for a
dissociable form of sequence learning
PETER F. DOMINEY,*]~:§ JOCELYNE VENTRE-DOMINEY,*
EMANNUEL BROUSSOLLE~ and MARC JEANNEROD*t
*Vision et Motricit6, Unit6 94 INSERM, 69500 Bron, France; tlnstitut des Sciences Cognitives EP-100-CNRS. 69008 Lyon, France; ~
++Servicede Neurologie, H6pital Neurologique, 69003 Lyon, France
(Received2 November 1995; accepted 2 May 1996)
Abstract Several studies of procedural learning in Parkinson's disease (PD) have demonstrated that these patients are impaired
with respect to age-matched control subjects. In order to examine more closely the specific impairment, we considered three
dimensions along which a procedural learning task could vary. These are: (1) implicit vs explicit learning, (2) instance vs rule learning,
and (3) learning with internal vs external error correction. We consider two hypotheses that could explain the impairments observed
in PD for different types of explicit motor learning: (H1) an impairment related to the acquisition of rules vs specific instances, and
(H2) an impairment in learning when no explicit error feedback is provided. In order to examine the condition of rule learning with
external error feedback, we developed a modified version of the serial reaction time (SRT) protocol that tests analogical transfer in
sequence learning (ATSL). Reaction times are measured for responses to visual stimuli that appear in several different repeating
sequences. While these isomorphic sequences are different, they share a common rule. Verbatim learning of a sequence would result
in negative transfer from one sequence to a different one, while rule learning would result in positive transfer. Parkinson's patients
and age-matched controls demonstrate significant acquisition and positive transfer of the rule between sequences. Our results
demonstrate that PD patients are capable of learning and transferring rule or schema-based representations in an explicit learning
format, and that this form of learning may be functionally distinct from learning mechanisms that rely on representations of the
verbatim or statistical structure of sequences. Copyright @'~ 1996 Elsevier Science Ltd.
Key Words: Parkinson's disease; serial reaction time; analogical transfer; sequence learning.
Introduction
While the contribution of Parkinson's disease (PD) to
deficits in initiating and executing familiar movements
and movement sequences is fairly well established [1, 2,
3, 11, 22, 30, 38], the contribution of PD to deficits in the
procedural learning of new movement sequences appears
to be present but less clear. In several tasks that test
different aspects of procedural learning, Parkinson's patients are significantly impaired with respect to controls
[15, 23, 26, 36, 43, 44]. However, modifications of these
tasks that change the mode of learning from implicit to
explicit can produce striking improvements in performance [36], indicating that it is not general m o t o r
learning, but something more specific that is impaired. In
§ Address for correspondence: Vision et Motricit6, Unit6 94
INSERM, 69500 Bron, France; fax: 3372369760.
order to clarify the specific impairment(s) we consider
three dimensions along which a procedural learning task
can vary. These three dimensions are (1) implicit vs
explicit learning, (2) specific instance vs rule learning [45],
and (3) learning with internal vs external error correction.
Implicit learning has been extensively studied with the
serial reaction time (SRT) task, developed by Nissen and
Bullemer [32]. In their SRT task, subjects respond to the
presentation of visual stimuli in one of four locations on
a video monitor by pressing one of four associated keys.
If the stimuli are presented in a repeating sequence, the
reaction times (RTs) are reduced, reflecting a form of
learning that was not seen when random stimuli were
used. The learning can be quantified as the difference
between mean RTs for the random vs the sequential
stimuli. In the implicit condition, subjects are simply told
to respond as quickly as possible, yet they display significant learning, and are often not aware that such learn~
ing has occurred [6, 7, 32]. If subjects are warned that a
2
P.F. Dominey et al./Analogical transfer in PD
sequence may be present, the task becomes explicit, often
with significantly greater learning taking place (e.g. Curran and Keele [7]).
In several recent studies with implicit SRT learning
tasks, PD patients displayed significant learning impairments with respect to their age-matched controls [15, 26,
36]. When the task is made more explicit, however, PD
patients' performance improves to that of control subjects [36]. In one explicit form, patients were informed
that a sequence would be presented, and they were asked
to concentrate on the sequence without making any
motor responses, and to be prepared to reproduce the
sequence at the end of the test. PD patients' learning in
this explicit task, as measured by the declarative knowledge of the sequence, was equal to that of the control
subjects [36]. Pascuale-Leone et al. [36] considered that
the PD patients' improvement is due to the explicit
instructions, though the effects of the eliminated motor
effort cannot be ignored. In a related test of the effect of
explicit declarative knowledge on SRT performance, PD
subjects demonstrated learning equal to that of control
subjects after 30 repetitions of the sequence. These results
indicate that the procedural learning impairment in PD
can be reduced or eliminated if tasks are made explicit
[36].
This does not appear to be the whole story, however,
when we consider procedural learning experiments of
Saint-Cyr et al. [43]. In their Tower of Toronto (TOT)
task, subjects are faced with a set of colored discs stacked
on one of three vertical pegs, with the discs progressing
from light on top to dark on the bottom of the stack.
Subjects are instructed to displace the entire stack from
one peg (initial state) to a different one (final state) following two rules: (1) only one disc can be moved at a
time, and (2) a darker disc can never be placed on a
lighter disc. Large color photographs of the initial and
final state are constantly in view. Solving the problem
optimally is based on an inherent rule that, once learned,
applies in all cases. Despite the fact that this is an explicit
task, PD patients require significantly more moves to
solve the four-disc problem than do age matched
controls. This suggests that the impairment may be in
learning a general procedural rule vs a specific repetitive
procedure as in the SRT task.
There is, however, another characteristic of the ToT
task that distinguishes it from tasks such as SRT with
respect to the presence or absence of error feedback. In
the SRT task, even if the subject has a tendency to generate an incorrect response, the correct response is always
provided, i.e. the SRT task has external error feedback. In
the ToT task, however, there is no such external feedback,
and incorrect moves may pass undetected. The importance of such error feedback has been clearly demonstrated by Vriezen and Moscovitch [49], who studied
the number of errors for PD patients and controls to
learn a set of stimulus-response associations between
numbers and visual stimuli (line drawings, abstract
designs and spatial locations). Subjects were presented
with the visual stimulus and had to respond with the
associated number. For the 'correction' procedure, on
error trials, the correct response was given as feedback,
and PD patients made no more errors than control
subjects. For the 'trial and error' procedure, on error
trials, subjects had to continue to guess until they made
the correct response, with only one error scored. In this
case, PD subjects made significantly more errors than
controls. This observation coincides with the conclusion
of Taylor et al. [48] that PD patients are specifically
impaired in conditions that require the ability to spontaneously generate efficient, self-directed response strategies and with several observations that PD patients are
disadvantaged in the absence of external cues [4, 18, 19,
27, 28, 31,47].
This leaves two open hypotheses consistent with the
observed PD impairments in explicit learning: HI: Rule
learning is impaired in PD while specific instance learning
is retained. H2: Self-directed learning is impaired in PD
while error feedback learning is not. Table 1 displays a
categorization of several explicit learning tasks along
these axes of internal vs external correction, and rule
vs instance learning. These data indicate that instance
learning is intact provided that external correction is
available, but they do not allow a similar conclusion
regarding rule learning, i.e. can rules be learned with
external correction? In order to respond to this question,
one should develop a protocol that requires a rule or rules
to be learned, and provides external error correction.
We have recently developed such a protocol that is
based on an explicit SRT task, with the modification that,
in each successive block of sequence trials, instead of
using the same repeating sequence, different isomorphic
sequences are used. Thus, while the sequences have
different serial or verbatim structure [46] in terms of the
spatial locations of the stimuli, they all share a common
rule that describes a repetitive structure. Examples of two
such isomorphic sequences are: (1) F C J - F C J and
(2) Q - L P - Q - L P. In both cases, the fourth to sixth
elements are fully predicted by the first to third elements.
Analogical transfer is the process of forming and exploiting such a rule or analogical schema, based on a familiar
problem in order to solve a new one [5, 20, 25, 33, 42].
We thus refer to our task as a test of analogical transfer
in sequence learning (ATSL), since it tests the ability to
acquire and transfer knowledge of an analogical schema
or rule from one problem to another in the domain of
sequence learning. We now compare PD and control
subjects in the ATSL task in order to determine if PD
patients can acquire a procedural rule in an explicit, externally guided task.
Material and methods
Patients and normal control sub/ects
Seven non-demented, non-depressed, right-handed patients,
with early- or mid-stage (duration range 1-10 years) idiopathic,
P. F. Dominey et al./Analogical transfer in PD
3
Table 1. Explicit learning of rules and instances with internal and external error correction
Rule
lnstance
Internal correction
Tower of Toronto
(PD group impaired)
Association-trial and error
(PD group impaired)
External correction
Analogical transfer in sequence
learning A T S L
(???'?)
A ssociation~orrection
SRT-explicit, declarative
(PD group unimpaired)
Rule and single instance based learning have been studied using procedures that require
internal correction, and single instance-based learning has been studied using procedures that
provide external correction. The current study addresses explicit rule learning using a procedure
that provides external correction.
levodopa-responsive Parkinson's disease, were tested. The
mean age (-+SD.) was 51.9_+11.6 years. Chronic anti-parkinsonian treatment consisted of levodopa (150-900 rag/day,
with peripheral levodopa-decarboxylase inhibitor) in all but
one case, dopamine agonists in four cases, deprenyl in one
patient, and anticholinergic drugs in two cases. One patient
(PI) was drug-free. The clinical evaluation of the motor state
of the patients was quantified on the Hoehn and Yahr scale
[24], and the Unified Parkinson's Disease Rating Scale
(UPDRS) [14]. Individual patient data are summarized in Table
2. All patients consented to participate in the experiment, and
had normal or corrected-to-normal vision.
A group of six healthy age matched volunteers was tested as a
control under the same experimental conditions for subsequent
comparison with the patients. The control subjects were all
right-handed. Their mean age (___S.D.) was 55.3 _+6.9 years.
(from target onset until subject's contact with the screen) was
recorded. The eight sequence targets used [labeled A H in Fig.
I(A)] were each 2.5 cm ~-squares, and were illuminated one at a
tirne. The letters A H were never presented to the subjects and
are only used here for describing the sequences. After a target
was touched, it was extinguished, the reaction time was recorded
and the next target was immediately displayed. It is important
A-B-C-B-C-D-C-D-E-D-E-F-E-F-G-F-G-H-G-H-A-H-A-B
A
B
A nalogica/tran.!'l~'r protocol
I-v] I-if1 m
I-if1
IT1
A-B-C
B-C-D
C-D-E
The primary task was based on the SRT protocol [32] and
involved pointing to single illuminated square targets on a
touch-sensitive screen as quickly and accurately as possible.
Subjects sat in front of the touch sensitive computer screen on
which the sequence targets were displayed, and response time
C
"n-2, n-2, u"
D
n-2
Table 2. Details of the individual patients with Parkinson's
disease
Patients Sex
Age
(years)
Duration of
illness
Hoehn and
(years)
Yahr stage
Pl
P2
F
M
38
62
I
10
P3
F
58
9
P4
M
45
8
P5
F
38
4
P6
P7
F
M
55
67
~
8
1.0 (off)
2.0 (on)
2.5 (off)
2.0 (on)
3.0 (off)
1.5 (on)
2.5 (off)
1.5 (on)
1.5 (on)
2.0 (on)
2.5 (off)
Treatment
None
250rag l-d,
d-a, D
900 mg l-d,
d-a, a
200 mg l-d, a
150rag l-d,
d-a
200 mg 1-d
750 mg l-d.
d-a
Medication abbreviations: l-d, levodopa/day; d-a, dopamine
agonist; a, anticholinergic; D, deprenyl. The Hoehn and
Yahr[I 5] stage of disease is indicated in all patients while chronically treated ('on'), and for the fluctuating patients in the "on"
and in the ~off" medication motor state.
n-2
Analogical Schema
['F"] [-E"] FD]
Fig. 1. Analogical transfer in sequence learning protocol.
Above-24 element sequence. (A) One mapping of letters A tt
to target locations on the touch-sensitive screen. Targets are
presented one at a time. Note that the letters themselves are
never displayed. (B) Breakdown of sequence A B C B C D
C D-E into three-element chunks. Note that each shaded
element is predictable from the element two positions behind,
labeled qa-2". For example, the first two elements "B-('" in
"B-C D' are predicted by the last two elements B C" in the
preceding chunk "A-B C'. (C) Analogical Schema. The first
two elements are predicted from the previous two elements
('n 2"), and the third element is unpredictable Cu'). These
elements are referred to, respectively, by their position in the
analogical schema, P1, P2 and P3. P1 and P2 are predictable
and P3 is unpredictable. This analogical schema is the unit that
describes the isomorphism between SEQI, SEQ2 and SEQ3
(see text). (1)) An alternative mapping of A H to the eight
target Ioc~ttions used to generate an isomorphic sequence.
P . F . Dominey et al./Analogical transfer in PD
4
to recall that, since the correct response is always presented,
this task provides explicit external correction of errors.
Targets could appear in blocks of two types--random and
sequence. In random blocks, 120 targets were successively presented in random order. In sequence blocks, 120 targets were
successively presented in five repetitions of 24 element sequences
of the form A B-C--B-C--D-_C-D-E D - E - F - E - F - G - F - G H G - H - A - H - A - B (Fig. 1). To appreciate the overall complexity of this sequence, note that each of the eight (A-H)
elements repeats three times, with two different successors, thus
yielding a complex, or ambiguous, sequence (see C, for example). A form of repetitive, internal organization becomes evident
if it is noted, for example, that G and H are repeats of the
elements two places behind them ( n - 2 ) , while A is unpredictable (u). As shown in Fig. I(B), this pattern ' n - 2 , n - 2 , u '
repeats throughout the sequence, forming the basis of the analogical schema displayed in Fig. I(C). In this notation, G and
H are said to be in positions 1 and 2 (both predictable) and A_
in position 3 (unpredictable) of the analogical schema
' n - 2 , n - 2,u'. We can observe that, in its first appearance, C is
in position 3 (unpredictable). It then appears in position 2 and
finally again in position 1 (both predictable). Finally, we see
that the sequence 'wraps around' so the repeating pattern is
never interrupted as the sequence repeats.
Since our goal is to study the transfer of knowledge between
different, isomorphic sequences, we must construct several
sequences that meet these requirements. Three such sequences
were generated by using the 24-element pattern described above
with three different mappings of A - H to the eight locations on
the touch sensitive screen [e.g. Figure I(A and D)]. Thus, the
three resulting 24-element sequences differ completely in their
serial or verbatim ordering of the spatial targets. However, they
are isomorphic in that they all share the analogical schema
' n - 2,n - 2,u' that describes their different specific serial orderings in a common abstract form.
To summarize, in each of three sequence blocks, one of these
sequences was repeated five times for a total of 120 targets•
Each individual target presentation and touch is referred to as
a trial. An experimental session started with 20 random trials to
familiarize the subject with the touch-screen. The data recording
started with a random block of 120 trials (RAND1), followed
by the three sequence blocks of 120 trials each (SEQ1, SEQ2,
SEQ3), and a final random block of 120 trials (RAND2).
Subjects were shown a diagram visually depicting the analogical schema ' n - 2, n - 2, u', and told before and once during
the examination that such a rule-like structure might be found
in the subsequent testing and that searching for and finding
such a structure could aid their performances. In this sense,
ATSL is an explicit learning task.
Data analysis
Results
Overall learning
F i g u r e 2 shows the m e a n RTi values for the P D and
c o n t r o l groups. The t h r e e - w a y A N O V A revealed a significant m a i n
effect for g r o u p
[F(1,105)-- 18.6,
P < 0.0001], with the m e a n RTi values o f - 1 1 2 msec for
the c o n t r o l g r o u p a n d - 38 msec for the P D group. T h e r e
was also a significant m a i n effect for b l o c k
[F(2,105)=13.76, P < 0 . 0 0 0 1 ] , with RTi values o f
- 14msec, - 109msec, a n d - 9 4 m s e c for SEQ1, SEQ2
a n d SEQ3, respectively. There was no g r o u p x b l o c k
i n t e r a c t i o n [F(2,105) = 0.52, P = 0.59].
W e verified t h a t the a c c u m u l a t e d learning d e m o n s t r a t e d in SEQ3 differed significantly f r o m the r a n d o m
responses in R A N D 2 by c o m p a r i n g a b s o l u t e R T values
for SEQ3 a n d R A N D 2 in p a i r e d t-tests. The c o n t r o l
g r o u p showed a significant difference between SEQ3 a n d
R A N D 2 [151 msec, t = 3.6, P < 0.01], as did the P D g r o u p
[48 msec, t = 4.9, P < 0.005]. This learning effect was verified by p e r f o r m i n g the same analysis on each i n d i v i d u a l
2 -5
For each subject, mean RTs were calculated for each block
of trials (RAND1, SEQ1, SEQ2, SEQ3, RAND2). Mean RT
values for two groups are presented in Table 3. RTs in RAND2
were used as a reference for simple sensory-motor coordination
improvement, independent of sequence-related learning, that
could occur during the five blocks. We thus calculated RT
Table 3. Mean RT values for the PD and control groups (in
milliseconds)
Control
PD
improvement (RTi) by subtracting the mean RT for RAND2
from the mean RT for each sequence block (RTi--mean SEQ
R T - mean RAND2 RT). This provided a quantification of the
sequence-learnin9 specific 1LT improvement.
Data were analysed using multifactor ANOVA and Scheffe's
post-hoe comparisons. For a more specific analysis, we used a
Student's t-test. All statistical analyses were performed by the
STATISTICA software package. For ANOVA, the within-subject factors were 'block' (SEQI, SEQ2, SEQ3), and 'position'
('predictable', 'unpredictable'). P1 and P2 were collapsed to
form the predictable case, and P3 formed the unpredictable
case. 'Group' (Parkinson, control) was the between-subjects
factor. The dependent variable was the RT improvement (RTi)
measure as described above. Note that an increasingly negative
RTi value corresponds to a greater learning effect.
RAND 1
SEQ 1
SEQ2
SEQ3
RAND2
502
643
444
589
336
505
340
531
491
579
0 Control
RAND1
SEQ1
SEQ2
SEQ3
RAND2
Fig. 2. Mean reaction time improvement (RTi) values by block
for the control and Parkinson groups. RTi values are calculated
by subtracting the mean RT for RAND2 block from the other
SEQ and R A N D blocks. Negative values indicate improvement
over Rand2 performance. First, the significant difference in
RT between SEQ3 and RAND2 indicates significant overall
learning for both groups (see text). Second, a cumulative
reduction in RTi for SEQ3 vs SEQ1 is seen in the control
and in Parkinson groups. This reflects analogical transfer since
SEQl-SEQ3 are all different isomorphic sequences.
P. F. Dominey et al./Analogical transfer in PD
which revealed significant SEQ3 R A N D 2 differences
[P<0.01] for all control subjects and for six of the seven
PD subjects. All of the control subjects reported an
awareness of a repeating structure in the three sequence
blocks, and five of the six were able to sketch a directed
graph figure that reflected the ' n - 2 , n - 2 , u' structure.
Likewise, all of the PD subjects reported an awareness of
a repeating structure in the three sequence blocks, and five
of the seven were able to sketch a pattern that reflected the
' n - 2 , n - 2 , u' structure. The sketches were considered
to reflect reasonable awareness if they included a directed
graph representing the pattern A B - A - B C.
In addition, the PD group showed an overall m o t o r
facilitation, independent of their sequence-related learning, as revealed by a significant RT improvement for
Rand2 over R a n d l [63msec, t=2.3, P<0.05]. The control group similarly displayed a small but non-significant
Rand2 vs Rand 1 improvement [ 11 msec, t = 1.58, P > 0.1 ].
Acquisition of analogical schema
The analogical schema is characterized by the properties that sequence elements in positions P1 and P2 are
predictable, whereas elements in position P3 are not predictable by the anological schema [Fig. I(C)]; thus, the
degree of schema acquisition can be quantified by this
(non-predictable-predictable) difference. Figure 3 displays the mean RTi values for the three sequences by
element position (predictable and non-predictable). The
three-way A N O V A revealed a significant main effect for
position [F(1,105)=63.1, P<0.0001], with mean RTi
Mean RTi for PD and Control Subjects
Predictable and Non-Predictable Elements
50 C
5
values of - 1 0 7 m s e c for predictable and - 2 m s e c for
unpredictable positions. There was a significant interaction between group and position [F(1,105)=10.9,
P < 0.005]. Post-hoc analysis by Scheffe's test revealed, in
the control group, a significant difference between predictable ( - 1 6 3 m s e c )
and non-predictable
RTis
( - 9 m s e c ) [P<0.0001]. The same effect was observed
for the PD group with a significant difference between
predictable ( - 6 0 m s e c ) and non-predictable RTis
(4 msec) [P < 0.02]•
Tran,}'['er ~/ the analogical schema
If positive transfer occurs between the three sequences,
then the difference between non-predictable vs predictable responses should become increasingly significant
across the three sequence blocks as a result of that transfer. In the three-way ANOVA, the significant interaction
between position and block [F(2,105)=5.23, P<0.01]
indicates a change in the relation between non-predictable and predictable positions during the course of
the three sequences, i.e. transfer of the analogical schema.
There was no significant group difference in this transfer
as revealed by the non-significant three-way interaction
[F(2,105)=0.81, P=0.45]. We analysed this transfer
effect more closely by two-way block (SEQ1, SEQ2.
SEQ3) x position
(predictable,
non-predictable)
A N O V A s for the PD and control groups independently
(Fig. 4).
The control group displayed significant main effects
for block [F(2,48)=5.4, P<0.01], and for position
[F(1,48)=36.5, P<0.0001]. The interaction between
block and position [F(2,48)= 2.8, P = 0.067] was not significant. However, post-hoc analysis by Scheffe's test
showed that the non-predictable-predictable ( N - P )
-
O
-50
'.
,~ -100
~Z
•
D""
.....
,..C]
[]
'¢ -150
-200
-250
--o- Non-Predictable
-.o-. Predictable
Seql
Seq2
Parkinson
Seq3
"D-. .....
Seql
Seq2
Seq3
Control
Fig. 3. Mean RT improvement (RTi) by position. Solid lines:
non-predictable elements (position 3), Dashed lines: predictable
elements (position 1 and 2). (Right) Control subjects. Predictable elements show a progressive significant RT reduction
in the three sequence blocks SEQ1-3 when compared to nonpredictable elements. This indicates that it is the analogical
schema, and not the sequences themselves, that is being learned•
Note in SEQ2, the slight reduction in RTs for non-predictable
elements, due to an increase in the number of spatially adjacent
target pairs in SEQ2, that yields a non-transferable RT
reduction for that sequence. (Left) Parkinson subjects display
a similar predictable vs non-predictable profile. See the text for
analysis.
SEQ1
SEQ2
SEQ3
Fig. 4. Analogical schema transfer. The level of analogical
schema transfer is displayed as the progressive change during the
three sequence blocks of the level of analogical schema acquisition, as indexed by difference between predictable minus nonpredictable RTis. For both the control and PD groups, this
measure becomes increasingly significant in the progression
from SEQ1 to SEQ3, indicating a significant level of analogical
transfer in both groups.
6
P.F. Dominey et al./Analogical transfer in PD
difference is non-significant in SEQ1 [ N - P = 7 5 m s e c ,
P--0.71], more significant in SEQ2 [ N - P = 162msec,
P<0.05], and highly significant in SEQ3 [ N - P =
220 msec, P < 0.001].
Likewise, the PD group displayed significant main
effects for block [F(2,57) = l 1.1, P < 0.0001], and for position [F(1,57)=23.9, P<0.0001], and as in the control
group, the interaction between block and position was
not significant [F(1,57) = 2.4, P = 0.10]. Furthermore, as
seen in the control group, post-hoc analysis by Scheffe's
test showed that the non-predictable predictable difference is non-significant in SEQI [N - P = 24 msec,
P = 0.95], and then significant in SEQ2 [ N - P = 80 msec,
P<0.02], and in SEQ3 [ N - P = 88 msec, P<0.02]. Thus,
for both groups, there was a significant positive transfer
between the successive sequences, with no group interaction in this measure of transfer.
Rule vs specific sequence learning
If subjects are learning specific sequences, then any
given sequence must be presented at least once in its
entirety before the effects of learning can be observed.
Thus, if learning is already evident in a new 24-element
sequence before that sequence has been completely presented for the first time, the learning can only be due to
a rule acquired and transferred from previous training.
To address directly whether a transferable rule or individual sequences were being learned, we performed a twoway ANOVA on group (PD and control) and position
(predictable, non-predictable) for the first 24 elements of
sequence block SEQ3, i.e. the first presentation of the
previously unseen sequence of block SEQ3. The dependent variable was the mean of the RTis for the predictable
and non-predictable elements, respectively, of the first
repetition (i.e. the first 24 elements) of the sequence in
SEQ3, calculated for each subject. The significant main
effect for position [F(1,22) = 33.6, P < 0.0001 )] indicates
that, even during the very first presentation of this
sequence, its predictable structure was already being
exploited, with RTi values of 35 and - 108 msec for the
non-predictable and predictable elements, respectively.
The lack of significant group or group x position interaction effects [F(1,22)=3.7, P=0.07); and F(1,22)=2.6,
P=0.12), respectively] indicates that there is no significant difference in PD vs control groups in the use of
the learned rule during this first repetition of the 24element sequence in SEQ3.
Discussion
These results demonstrate that, under conditions of
explicit learning with external error feedback, patients
with Parkinson's disease are capable of a simple form of
analogical transfer that involves learning a generalized
rule that can apply in different but isomorphic situations.
This allows us to reject our Hypothesis H l which stated
that "Rule learning is impaired in PD while specific
instance learning is retained". In contrast, these results
support Hypothesis H2: "Self-directed learning is impaired in PD while error feedback learning is not", in
agreement with the conclusions of Taylor et al. [48] and
Vriezen and Moscovitch [49] and the general observation
that these patients rely heavily on external cues [4, 18, 19,
27, 28, 31, 47].
In addition, these results help to clarify the distinction
between rule and instance learning advocated by Shanks
and St. John [45], demonstrating how, from several perspectives, the explicit rule learning in ATSL is functionally dissociable from that of specific sequencelearning, either implicit or explicit. Particular differences
can be seen for both PD and control groups in terms of
(1) the distribution of reaction times for elements within
sequence blocks, (2) the dependence of learning on
sequence length, and (3) characteristics of performance
transfer to new sequences.
(1) Distribution of reaction times within the sequence
blocks. If a sequence is learned "verbatim" [46], the
RT reductions should be observed for all sequence
elements with a relatively uniform distribution. In the
ATSL task, reduced RTs are seen for predictable but
not for non-predictable elements, even though both
are contained in sequence blocks. This supports the
view that a rule, that is not specific to any one of these
sequences, is being learned, and not the sequences
themselves. Curran and Keele [7] performed a similar
analysis of RT distribution in an SRT task using a
repeating sequence of the form C-A B-C B D.
They compared RTs for three types of positions:
unique (A and D), after unique (B following A and
C following D), and ambiguous (C following B, and
B following C). The ambiguous elements (underlined)
are in fact repetitions of the elements three and two
places behind them and thus resemble our predictable
elements, while the unique elements resemble our
non-predictable elements. Curran and Keele's analysis confirmed, in fact, that RTs for all three of these
position types are significantly reduced with respect
to those for random elements, supporting their contention that the entire structure of the sequence was
being learned. We can consider that, in general, if a
single given sequence is successively repeated then the
sequence itself can be learned, whereas if a number
of isomorphic sequences are presented then only the
rule common to all can be learned.
(2) The lack of dependence on sequence length. Motor
sequence performance is dependent upon several parameters of the sequence, including its length [1, 2,
15, 26, 36]. Length dependence in the SRT task was
demonstrated by Pascuale-Leone et al. in both normal controls and PD subjects, using repeating
sequences of length 8, 10 and 12 elements [36]. RT
reduction was inversely related to sequence length in
P. F. Dominey et al./Analogical transfer in PD
control and PD subjects, and for the longer
sequences, PD patients were particularly impaired.
For the 12-element sequence, the RT improvements
( R A N D - SEQ) were 145msec for controls and
45 msec for PD patients. Ferraro et al. [15] observed
RT improvements of 88 and 51 msec for control and
PD patients, respectively, using a 10-element
sequence, and for sequences of length 11, Jackson el
al. [26] found RT improvements of 74 msec for control and 7 msec for PD patients. In contrast, in the
last of three different 24-element sequences, we
observed RT improvements of 151 msec and 48 msec
for control and PD subjects, respectively, a notable
deviation from the length dependencies observed in
non-rule-based SRT tasks in PD [15, 26, 36].
(3) Tran,~l~'r to isomorphic sequences. The most profound
difference between the rule learning mechanisms that
we observe vs specific sequence learning is the significant positive transfer to different but isomorphic
sequences for both PD and control groups. This is in
contrast to the negative transfer results of Robertson
and Flowers [41] who demonstrated that, while PD
subjects were able to demonstrate explicit sequence
learning, when asked to change from one sequence to
the next, these patients displayed significantly more
negative transfer than that observed in controls. That
is, after instructions to shift to a different sequence,
PD subjects tended to continue with the old sequence.
A related form of negative transfer was observed by
Benecke et al. [3] in making the transition from one
movement to another in a sequence, and is related to
the more general impairment in set-shifting observed
in PD [16, 34, 35].
In contrast, our observation of no negative transfer
likely results from the fact that it was not a set of
different sequences that was learned, but instead, a
single rule that is common to all of these sequences.
The observed transfer was, in fact, positive, since,
from the perspective of the rule in question, all three
sequences are the same. Carrying this line of reasoning to its logical end, we considered that, if subjects
are learning a verbatim representation, then there
can be no performance improvement during the first
exposure to the new sequence. Learning can only be
seen on the second and subsequent repetitions of the
sequence. We observed, however, for both PD and
control subjects, significant learning effects during
the first presentation of the new 24-element sequence
in block SEQ3, with no group interaction. In terms
of the learned rule, SEQ3 is equivalent to the previous
sequences SEQI and SEQ2 and thus it is not surprising that learning effects are seen immediately, via
transfer of the rule,
Sequential structure andJbrms o f learning
Stadler [46] pointed out that performance improvements can result from learning a verbatim representation
7
of the sequence, event by event, or from learning an
aggregate representation of the sequence in terms of
probabilities of elements, or element transitions, etc. [26,
29, 40, 46]. Here, we consider a third representation that
is rule-based. In a rule, the unit of representation is not
the set of specific sequence elements nor statistical
relations between pairs or groups of elements. Instead, it
is in terms of the relative positions of repeating elements.
For example, two sequences like A B - A - B C and G
D G - D - H can be represented in a common by the rule
'u. u, n 2, n - 2 , u' that describes the relations between
repeating elements. This type of rule-based representation has been demonstrated in tests of judging
the grammaticality of letter strings [21, 37, 39]. Subjects
learned a sel of grammatical rules by studying strings or
letter pairs generated with one set of letters, and were
then able to correctly judge the grammaticality of new
strings generated with the same letter set. In tests o1"
transfer of this rule-based knowledge to a set of strings
generated with new set of letters, only subjects trained on
strings were capable of this transfer [21]. This suggests
a dissociation between one system that learns elementspecific associations avaiable in the pairs training data,
vs another that learns more abstract relations between
repeated elements available in the strings training data.
In conclusion, we have demonstrated that, in an
explicit procedural learning context, PD subjects, like
normal controls subjects, are capable of developing a
rule-based representation of sequential knowledge that
serves as the basis for analogical transfer to new, isomorphic sequences. This suggests that PD patients are
not intrinsically impaired in learning procedural rules, as
required in the Tower of Toronto task [43], but rather,
that they are impaired in the ability to generate internally
an appropriate evaluation or response in error conditions
[48, 49].
At the same time, based on the differences in ATSL
and SRT learning characteristics, including differences in
the distribution of RT reductions, length dependencies
and transfer, these results suggest that the rule-based
representation that supports ATSL is functionally distinct Dora that involved in SRT learning. In support of
this idea, our recent simulation results have demonstrated
that, while a recurrent neural network can learn verbatim
and aggregate structures in SRT task, it failed in the
ATSL task which requires an additional capacity to represent positional relations between repeating elements [8
10, 12, 13], see also [17] for comments on fronto-cortical
hierarchical dissociations].
Acknowh, dqements The authors acknowledge the constructive
comments of two anonymous reviewers and the Editor. We also
gratefully acknowledge discussions with Keith Holyoak and
Jean Saint-Cyr regarding aspects of analogical transfer and
learning. PFD was supported by a Post-Doctoral Fellowship
from the Fyssen Foundation (Paris).
8
P.F. Dominey et al./Analogical transfer in PD
References
1. Agostino, R., Berardelli, A., Formica, A., Accornero, N. and Manfredi, M. Sequential arm movements in patients with Parkinson's disease,
Huntington's disease and Dystonia. Brain 115, 1481
1495, 1992.
2. Agostino, R.,.Berardelli, A., Formica, A., Fabrizio,
Stocchi, Accornero, N. and Manfredi, M. Analysis
of repetitive and nonrepetitive sequential arm movements in patients with Parkinson's disease. Movement Disorders 9(3), 311-314, 1994.
3. Benecke, R., Rothwell, J. C., Dick, J. P. R., Day, B.
L. and Marsden, C. D. Disturbance of sequential
movements in patients with Parkinson's disease.
Brain 110, 361-379, 1987.
4. Brown, R. G. and Marsden, C. D. Internal versus
external cues and the control of attention in Parkinson's disease. Brain 111, 323-345, 1988.
5. Catrambone, R. and Holyoak, K. J. Overcoming
contextual limitations on problem-solving transfer.
Journal of Experimental Psychology: Learning Memory and Cognition 15, 1047 1056, 1989.
6. Cohen, A., Ivry, R. I. and Keele, S. W. Attention
and structure in sequence learning. Journal of Experimental Psychology: Learning Memory and Cognition
16, 17-30, 1990.
7. Curran, T. and Keele, S. W. Attentional and nonattentional forms of sequence learning. Journal of
Experimental Psychology: Learning Memory and
Cognition 19(1), 189-202, 1993.
8. Dominey, P. F. Complex sensory motor sequence
learning based on recurrent state-representation and
reinforcement learning. Biological Cybernetics 73,
265-274, 1995.
9. Dominey, P. F. (1996) An anatomically structured
sensory-motor sequence learning system displays
some general linguistic capacities. Brain and Language, in press.
10. Dominey, P. F., Arbib, M. A. and Joseph, J. P.
A model of cortico-striatal plasticity for learning
oculomotor associations and sequences. Journal of
Cognitive Neuroscience 7(3), 310-336, 1995.
11. Dominey, P. F., Decety, J., Broussolle, E., Chazot,
G. and Jeannerod, M. Motor imagery ofa lateralized
sequential task is asymmetrically slowed in hemiParkinson's patients. Neuropsychologia 33(6), 727741, 1995.
12. Dominey, P. F., Ventre-Dominey, J., Broussolle, E.
and Jeannerod, M. Analogical transfer in sequence
learning: human and neural-network models of
fronto-striatal function. Annals of the New York
Academy of Science "/69, 369-373, 1995.
13. Dominey, P. F., Ventre-Dominey, J., Broussolle, E.
and Jeannerod, M. Representation and computation
for analogical transfer in sequence learning (ATSL):
human and neural models of cortico-striatal function. In Computational Neuroscience: Trends in
Research 1995, J. M. Bower (Editor), pp. 335 341.
Academic Press, San Diego, CA, 1996.
14. Fahn, S., Elton, R. L. and the Unified Rating Scale
Development Committee. In: Recent Development in
Parkinson's Disease, S. Fahn, C. D. Marsden and
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
D. Calne (Editors), Vol. 2, pp. 153-163. Macmillan,
New York, 1987.
Ferraro, F. R., Balota, D. A. and Connor, L. T.
Implicit memory and the formation of new associations in nondemented Parkinson's disease individuals and individuals with senile dementia of the
Alzheimer type: a serial reaction time (RT) investigation. Brain and Cognition 21, 163-180, 1993.
Flowers, K. A. and Robertson, C. The effect of Parkinson's disease on the ability to maintain a mental
set. Journal of Neurology, Neurosurgery and Psychiatry 48, 517-529, 1985.
Fuster, J. M. Up and down the frontal hierarchies;
whither Broca's area? Commentary on Greenfield.
Behavioral Brain Sciences 14, 558, 1991.
Georgiou, N., Bradshaw, J. L., Iansek, R., Phillips,
J. G., Mattingley, J. B. and Bradshaw, J. A.
Reduction in external cues and movement sequencing in Parkinson's disease. Journal of Neurology,
Neurosurgery and Psychiatry 57, 368-370, 1994.
Georgiou, N., Iansek, R., Bradshaw, J. L., Phillips,
J. G., Mattingley, J. B. and Bradshaw, J. A. An
evaluation of the role of internal cues in the pathogenesis of Parkinson hypokinesia. Brain 116, 15751587, 1993.
Gick, M. L. and Holyoak, K. J. Schema induction
and analogical transfer. Cognitive Psychology 15, 138, 1983.
Gomez, R. L. and Schaveneveldt, R. W. What is
learned from artificial grammars? Transfer tests of
simple association. Journal of Experimental Psychology: Learning Memory and Cognition 20(2), 396410, 1994.
Harrington, D. L. and Haaland, K. Y. Sequencing in
Parkinson's Disease: abnormalities in programming
and controlling movement. Brain 114, 99-115, 1991.
Harrington, D. L., Haaland, K. Y., Yeo, R. A. and
Marder, E. Procedural memory in Parkinson's
Disease: impaired motor but not visuoperceptual
learning. Journal of Clinical and Experimental Neuropsychology 12(2), 323-339, 1990.
Hoehn, N. M. and Yahr, M. D. (1967) Parkinsonism:
Onset, progression, and mortality. Neurology, Minneapolis, 17, 427~442.
Holyoak, K. J., Novick, L. R. and Melz, E. R. (1994)
Component processes in analogical transfer:
Mapping, pattern completion, and adaptation. In
Advances in Connectionist and Neural Computation
Theory, Vol. 2: Analogical connections, K. Holyoak
and J. Barnden (Editors), pp. 113 180. Ablex, Norwood, NJ.
Jackson, G. M., Jackson, S. R., Harrison, J., Henderson, L. and Kennard, C. Serial reaction time learning and Parkinson's disease: Evidence for a
procedural learning deficit. Neuropsychologia 33(5),
577-593, 1995.
Jahanshahi, M., Brown, R. G. and Marsden, C. D.
Simple and choice reaction time and the use of
advance information for motor preparation in Parkinson's disease. Brain 105, 539 564, 1992.
Klockgether, T. and Dichgans, J. Visual control of
arm movements in Parkinson's disease. Movement
Disorders 9(1), 48-56, 1994.
P. F. Dominey et al./Analogical transfer in PD
29. Lewicki, P., Czyzewka, M. and Hoffman, H. Unconscious acquisition of complex procedural knowledge.
Journal of Experimental Psychology: Learning Memory and Cognition 13(4), 523-530, 1987.
30. Marsden, C. D. (1982) The mysterious motor function of the basal ganglia: The Robert Wartenberg
Lecture. Neurology 514~539.
31. Morris, R. G., Downes, J. J., Sahakian, B. J., Evenden, J. L., Heald, A. and Robbins, T. Planning and
spatial working memory in Parkinson's disease. Jour-
nal of Neierology, Neurosurgery and Psychiatry 51,
757-766, 1988.
32. Nissen, M. J. and Bullemer, P. Attentional requirements of learning: Evidence from performance measures. Cognitive Psychology 19, 1-32, 1987.
33. Novick, L. R. Analogical transfer, problem similarity, and expertise. Journal of Experimental PO'chology: Learning Memory and Cognition 14(3), 510
520, 1988.
34. Owen, A. M., James, M., Leigh, P. N., summer, B.
A., Marsden, C. D., Quinn, N. P., Lange, K. W. and
Robbins, T. W. Fronto-striatal cognitive deficits at
different stages of Parkinson's disease. Brain 115,
1727-1751, 1992.
35. Owen, A. M., Roberts, A. C., Hodges, J. R.,
Summers, B. A., Polkey, C. E. and Robbins, T. W.
Contrasting mechanisms of impaired attentional setshifting in patients with frontal lobe damage or Parkinson's disease. Brain 116, 1159 1175, 1993.
36. Pascual-Leone, A., Grafman, J., Clark, K., Stewart,
M., Massaquoi, S., Lou, J.-S. and Hallett, M. Procedural learning in Parkinson's disease and cerebellar
degeneration. Annals of Neurology 34, 594~602, 1993.
37. Perruchet, P. and Pacteau, C. Implicit acquisition of
abstract knowledge about artificial grammar: some
methodological and conceptual issues. Journal of
Experimental Psychology: General 120(i), 102-106,
1991.
38. Rafal, R. D., lnhoff, A. W., Friedman, J. H. and
Bernstein, E. Programming and execution of sequential movements in Parkinson's disease. Journal of
Neurology, Neurosurgery and Psychiatry 50, 12671273, 1987.
9
39. Reber, A. S. Transfer of syntactic structure in synthetic languages. Journal of Experimental Psychology
81, 115 119, 1969.
40. Reed, J. and Johnson, P. Assessing implicit learning
with indirect tests: Determining what is learned
about sequence structure. Journal o[" Experimental
Psychology: Learning Memoo' and Cognition 20(3),
585 594, 1994.
41. Robertson, C., Flowers, K. A. Motor Set in Parkinson's disease. Journal of Neurology, Neurosurgery
and Po~chiatr)' 53, 538-592, 1990.
42. Ross, B. H. This is like that: The use of earlier problems and the separation of similarity effects. Journal
of Experimental Psycholog),: Learning Memory and
Cognition 13(4), 629-639, 1987.
43. Saint-Cyr, J. A., Taylor, A. E. and Lang, A. E. Procedural learning and neostriatal dysfunction in man.
Brain 101,941 959, 1988.
44. Saint-Cyr, J. A., Taylor, A. E. and Nicholson, K.
(1995) Behavior and the basal ganglia. In Behavioral
Neurology (?/ Movement Disorders, W. J. Weiner and
A. E. Lang (Editors), Advances in Neurology, Vol.
65, pp. 1 28. Raven Press, New York.
45. Shanks, D. R. and St. John, M. F. Characteristics of
dissociable learning systems. Behavioral and Brain
Sciences 17, 367-447, 1994.
46. Stadler, M. A. Statistical structure and implicit learning. Journal oJ" Experimental Psychology: Learning
Memo O' and Cognition 18, 318 327, 1992.
47. Stelmach, G. E., Worringham, C. J. and Strand, E.
A. Movement preparation in Parkinson's disease:
The use of advance information. Brain 109, 1079
1094, 1986.
48. Taylor, A. E., Saint-Cyr, J. A. and Lang, A. E. Frontal lobe dysfunction in Parkinson's disease: The cortical focus of neostriatal outflow. Brain 109, 845 883,
1987.
49. Vriezen, E. R. and Moscovitch, M. Memory for temporal order and conditional associative learning in
patients with Parkinson's disease. Neuropsychologia
28(4), 1283 1293, 1990.