Content uploaded by Paul W Burgess
Author content
All content in this area was uploaded by Paul W Burgess on Oct 29, 2018
Content may be subject to copyright.
Pergamon
Neuropsychologia,
Vol. 34, No. 4, pp. 263-273, 1996
Copyright © 1996. Published by Elsevier Science Ltd. All rights reserved
0028-3932(95)00104-2 Printed in Great Britain
0028-3932/96 $15.00 + 0.00
Response suppression, initiation and
strategy use following frontal lobe
lesions
PAUL
W.
BURGESS*t and TIM SHALLICEt:I:
tDepartment of Psychology, University College London, Gower Street, London WC1E 6BT, U.K.; and
:~Department of Clinical Neuropsychology, The National Hospital for Neurology and Neurosurgery, London WC1, U.K.
(Received
30
July
1993;
accepted
30
May
1995)
Abstract--Ninety-one patients with cerebral lesions were tested on a task involving two conditions. In the first condition (response
initiation) subjects were read a sentence from which the last word was omitted and were required to give a word which completed the
sentence reasonably. In the second condition (response suppression) subjects were asked to produce a word unrelated to the
sentence. Patients with frontal lobe involvement showed longer response latencies in the first condition and produced more words
which were related to the sentence in the second, in comparison to patients with lesions elsewhere. Moreover, in the second condition
patients with frontal lobe lesions produced fewer words which showed the use of a strategy during response preparation.
Performance on the initiation and suppression conditions was unrelated at the group or single case level. The relationships between
response initiation, suppression and strategy use are discussed.
Key Words: frontal lobes; executive function; strategy; response suppression; inhibition; initiation.
Introduction
It has long been known that frontal lobe lesions can
cause deficits in both response initiation and suppression
[7, 8]. For instance, on the Verbal Fluency test (VFT)
patients with frontal lobe lesions produce fewer words
than patients with lesions elsewhere [e.g. 2, 10, 12; but
see 18]. One view is that this reduced word fluency is
linked to a general aspontaneity of thought [e.g. 7, 17]
and so is a feature of frontal initiation problems. A more
specific interpretation of reduced verbal fluency in
certain patients has been given by De Lacey Costello
and Warrington [4], namely that reduced word fluency is
one characteristic of the syndrome of dynamic aphasia,
in which patients present with reduced spontaneous use
of language in the context of normal confrontation
naming, language comprehension and speech produc-
tion. They further suggest that dynamic aphasia
*To whom all correspondence should be addressed.
represents a deficit in verbal planning at a stage prior
to sentence construction.
Response suppression, too, is well known to present
problems for frontal lobe patients. Thus, Drewe [6]
found that patients with frontal lobe lesions were
significantly poorer than patients with lesions elsewhere
at a go-no go task which required the subjects to
withhold responses to one or two stimuli. More
specifically, Verfaellie and Heilman [2l] gave a patient
with medial frontal lobe lesions the task of responding
to a touch of one hand by raising the contralateral hand.
The patient consistently raised the ipsilateral hand in
response to the touch. The authors interpreted this as an
inability to inhibit an incorrect response since the
patient showed self-corrections and was able to describe
verbally the correct response when asked to do so.
The issue of whether the inability to suppress the most
salient response is related to any response initiation
difficulty is raised by the work of Perret [16] who used a
form of the Stroop test with groups of patients with
cortical lesions, and found that the patients whose lesion
invaded the left frontal lobe were significantly poorer at
suppressing the habitual response than patients with
lesions elsewhere. In addition, those patients who
263
264 P. W. Burgess and T. Shallice/Response suppression following frontal lobe lesions
performed poorly on the Stroop test were also those
who were most likely to achieve a poor Verbal Fluency
score.
Perret's favoured explanation of this finding was that
the Stroop task and the Verbal Fluency task make
similar cognitive processing demands, despite their
apparent differences in format. He argued, in reference
to the VFT, that the "usual criterion in the search for
words" (p. 323) is word meaning. By asking the patients
to perform their search according to initial letter, one is
asking them "to suppress the habit of using words
according to their meaning" (p. 324). Thus for Perret,
both the Stroop test and the VFT require the subjects to
suppress the habitual response. A possibility which
Perret does not consider, but which is a variant of this
view, is that an initial step that is required before the
generation of a new word is the suppression of the
currently active representation.
However, Perret also considered an alternative
explanation, namely that the two processes might be
functionally independent, but that the cerebral struc-
tures which subserve them could be in close proximity in
the brain so that both processes might tend to be
affected together in patients with cerebral damage. Yet a
third explanation is possible for his result, namely that a
high-level "supervisory system" [18] is damaged in
frontal lobe patients, thus both the initiation of novel
responses and response suppression abilities are im-
paired. The question of the relationship between
response initiation and inhibition therefore remains
open.
A difficulty in the interpretation of the performance
of frontal patients across different frontal tests such as
the Stroop and word fluency arises from the differing
characteristics of the specific tests. The present study,
therefore, utilises a task in which the differing
components of initiation and inhibition can be
examined with minimal changes in the background
characteristics of the task. Sentences in which the last
word is omitted are presented to the patient, the
sentences being chosen such that there is a high
probability of a particular response occurring. The
patient's task is to complete the sentence, in one
condition with any word and in the other with a word
which makes no sense in the context; in this second
condition the rational response has to be inhibited.
Method
Subjects
Right-hand dominant (as assessed by the admitting
neurologist) and aged between 18 and 75, with a unilateral
lesion restricted to one or two lobes (with the exception of
bilateral frontal lesions with no posterior involvement, who
were considered as a separate group; no history of psychiatric
problems, alcohol or substance abuse, or any previous
neurological conditions; no hydrocephalus or long-standing
epilepsy; first language as English; having a CT brain scan
available for their current condition; and of course being
capable of, and willing to undertake the experimental
procedure in addition to a full neuropsychological assess-
ment.
Seventeen of the original 108 patients failed to meet these
criteria and were excluded, leaving 91 patients. The remaining
patients were classified according to site of lesion based on
the radiographer's report of the CT scan. Any patient who
had involvement of the frontal lobes was classified as
'anterior', and those patients who had involvement elsewhere
in the cortex, but not involving the frontal lobes, were
classified as 'posterior'. Patients who had bilateral frontal
lobe lesions with no posterior (i.e. non-frontal) involvement
were considered separately as a 'bifrontal' group. The ages of
these patients, together with their pre-morbid IQ estimates
based on the NART [14], and current levels of intellectual
functioning [22] are shown in Table 1. Of the 91 cases, 44
were confirmed as suffering from primary tumours, the
majority of which were forms of glioma with 15 cases of
menigioma and six with cystic lesions. Sixteen cases had
space-occupying lesions, presumed neoplastic, the precise
nature of which could not be confirmed at the time of testing.
Twenty-one cases were suffering from lesions which were
vascular (infarct and haemorrhage) in origin and six cases
had a subdural haematoma. Additionally there were two
cases with a metastatic lesion and two cases with a cerebral
abscess. Overall 60.9% of the anterior group and 77.8% of
the posterior group were confirmed as having, or were
presumed (on the basis of CT scan) to be suffering from, a
tumour. Of the anterior cases 28.1% had vascular lesions,
compared with 11.1% of the posteriors. Of the posterior
group 57.9% had lesions confined to one lobe, compared
with 61.8% of the anterior group. Details of the patients
along with a group of 20 normal subjects matched for age
and IQ are given in Table 1. The controls were mainly
relatives accompanying patients. None had experience of
psychology or related disciplines.
Materials
The Hayling test consists of 30 sentences in which the final
word was omitted. They were selected from those given by
Bloom and Fischler [3] who present norms for completion of
the sentences; the 30 chosen were those for which there was a
particularly high probability of one specific response. The
American terms in the sentences were translated into their
British equivalents (see Appendix 1 for examples). The 30
sentences were pseudo-randomly assigned to one of two
groups of 15 sentences to be used in either the first or second
sections of the test. In addition four more sentences from
Bloom and Fischler were selected to be used as examples for
the subjects.
A consecutive series of 108 patients referred for routine
neuropsychological assessment to the Department of Clinical
Neuropsychology at the National Hospital for Neurology and
Neurosurgery were seen as part of the present study. Patients
were then excluded if they did not pass all the following
criteria:
Method and task instructions
The task consists of two sections (A and B), each containing
15 sentences. Subects were tested with only the examiner
present, in a quiet standard hospital office.
P. W. Burgess and T. Shallice/Response suppression following frontal lobe lesions
Table 1. Subject characteristics (S.D. in brackets)
265
Group n Age WAIS-IQ*~ NARTt
L anterior 21 40.8 (13.2) 100.1 (13.8) 106.3 (10.5)
R anterior 26 48.5 (13.3) 109.6 (13.4) 114.3 (8.8)
L posterior 12 37.8 (14.3) 102.1 (15.8) 111.2 (11.6)
R posterior 15 48.1 (14.0) 108.6 (9.5) 111.5 (9.8)
Bifrontal 17 51.4 (13.0) 100.6 (11.0) 107.2 (9.6)
Collapsed to give:
Anteriors 47 45.1 (13.7) 105.3 (14.3) 110.7 (10.3)
Posteriors 27 43.5 (14.8) 105.7 (12.8) 111.4 (10.4)
Bifrontal 17 51.4 (13.0) 100.6 (11.0) 107.2 (9.6)
Controls 20 49.7 (13.7) ~ 112.0 (12.9)
*Certain subjects were given the WAIS--R; a correction factor
~-National Adult Reading Test [16].
:~Wechsler Adult Intelligence Scale [22].
§Not administered.
of + 8 points was then used [5].
Section A: Initiation
The subject was given spoken instructions of which the
critical part is "I want you to listen carefully to each sentence
and when I have finished reading it, your job is to give me a
word you think could fit at the end of the sentence as quickly
as possible". Two practice sentences are read to the subject and
s/he was required to give a reasonable response. If the subject
was still unsure about the task, further explanation was given
(in practice this was extremely rare). Sentences were then read
to the subject at a normal reading pace. Timing of response
latency was by stop-watch, started as soon as it was judged the
last word of the sentence had been read, with timing being
stopped as soon as the subjects began their response.
Section B: Response suppression
Section B was conducted immediately after section A and
used the same type of sentence stem. Between the two tasks the
subjects were given instructions with the following critical
section "this time I want you to give me a word which makes
no sense at all in the context of the sentence. I want the word
you give me to be unrelated to the sentence in every way". The
two examples were then given to the subject. If the subjects was
unable to think of an unrelated word, the example word
'banana' (which is unrelated to either of the two practice items)
was suggested to the subject. If at any time during this stage of
the test, the subject gave a sentence completion rather than an
unrelated word, s/he was told that the word was too related to
the sentence, and the task instructions were repeated. In
practice, with a few of the subjects who were very poor at the
task, this resulted in their being reminded about the task
instruction after each of the 15 trials. If the patient had not
produced a word within 60 sec, that trial was terminated and a
response latency of 60 sec was recorded.
Results
No significant laterality effects appeared in any of the
measures (see Table 2), so the unilateral patients were
collapsed to give four groups, unilateral anterior or
posterior, a bi-frontal group and controls.
In the statistical analysis we will principally consider
the contrast between the control group and the two
unilateral lesion groups since the control group and
particularly the unilateral posterior group are the critical
contrast groups for drawing theoretical conclusions
about any putative unilateral anterior group deficit. To
include the bifrontal group would be inappropriate since
any selective deficit the group shows is potentially
explicable in terms of lesion size effects and moreover
their inclusion would contaminate analyses of lesion
extent effects within the unilateral lesion patients. The
bifrontal anterior patients are therefore contrasted with
the unilateral anterior patients to see if any selective
unilateral anterior deficit is exacerbated in the bilateral
anterior group.
Table 2. Response latencies across groups (sec; S.D. in brackets)
Group Section A Section B B-A
Controls 12.6 (6.5) 26.4 (19.5) 13.8 (16.3)
Posteriors 17.4(14.1) 40.6 (38.9) 23.2 (28.9)
Anteriors 28.6 (34.5) 83.1 (106.8) 54.4 (79.3)
Bifrontals 35.8 (23.1) 135.4 (119.3) 99.6 (118.9)
LH (n=33) 26.2 (22.3) 59.6 (68.2) 33.3 (56.0)
RH (n=41) 23.2 (33.9) 74.1 (105.1) 50.9 (74.5)
LH = all patients with (unilateral) left hemisphere lesions considered together.
RH = all patients with (unliateral) right hemisphere lesions.
266 P.W. Burgess and T. Shallice/Response suppression following frontal lobe lesions
The study used neuropsychological group study
methodology in which selection of unilateral patients
was determined as far as the lesion was concerned only by
no more than two lobes being involved. Thus patients
with a number of different aetiologies and differing lesion
sizes are included. As a control check we therefore
followed significant findings by a multiple regression
analysis which included aetiology and number of lobes
involved.
Section A of the task (straightforward completion)
The quality of the answers was satisfactory for all
groups since less than 1%0 of responses in this section
were not one of those predictable from the results of
Bloom and Fischler [3] on the quality of the replies. The
response latencies (sum of latencies across 15 trials) for
all groups are shown in Table 2. There was a significant
difference between the unilateral anterior, posterior and
control groups [ANOVA following log transformation
to eliminate skewness d.f. 2, 91; F=7.80, P<0.002].
Tukey-Kramer
post-hoc
comparisons showed that the
anterior group were significantly slower than the
controls at the P<0.01 level. None of the other
comparisons were significant for P< 0.05.
Analysis of the patients' times by multiple linear
regression indicated both age [t-ratio 3.27, P<0.005]
and FSIQ [t-ratio -3.12, P<0.005] to be significant
predictors of performance on this measure. An analysis
of covariance (following log transformation) was there-
fore performed using age and FSIQ as covariates. Under
these conditions the anterior-posterior difference in the
section A times was significant at the 0.02 level. The
bifrontal group were poorer still than the unilateral
anteriors [t = -2.12, d.f.=36, P<0.05].
In order to investigate the effect of lesion aetiology and
number of lobes involved the patient's lesions were
classified into three broad categories: vascular, tumour
(or presumed tumour) or 'other'. Aetiology of lesion,
together with number of lobes affected and whether the
lesion was anterior or posterior in location were used as
predictors of response latencies using multiple linear
regression (log transformed data). When all predictors
were considered together, only the anterior-posterior
classification were significantly predictive of performance
[t-ratio 2.14, P < 0.05]. Neither the unilateral posterior or
anterior single-lobe involvement cases were significantly
different in their pattern of performance from their
unilateral two-lobe counterparts [posteriors: one lobe =
mean 17.8 (16.1), two lobes= 17.2 (8.9); anteriors: one
lobe = mean 22.0 (22.0), two lobes = 20.6 (11.0)].
Section B of the task (anomalous completion condition)
Error classification.
In order to discover whether the
groups differed in their ability to produce words
unrelated to the stimulus sentence in section B, three
raters, blind to the purpose of the study, rated each of
the 1665 subject reponses (111 subjects x 15 sentences).
They were asked to classify each of the responses into
one of the eight possible categories shown in Fig. 1,
following the instructions presented in Appendix 1.
These eight categories can be subdivided into three
main groups. Firstly, there are responses which are
sensible completions of the sentence, thus clearly
violating the task instructions (category C in Fig. 1).
Secondly, there are responses that are semantically
connected to the sentence in some way (categories SO,
SA, SB and SC in Fig. 1). This second set of responses
may be semantically related to either the expected
completion of the sentence (including opposites of what
might constitute a sentence completion), or the subject
of the sentence, or it may be a word, which whilst not
entirely completing the sentence in a reasonable fashion,
nevertheless gives the sentence some meaning (socially
inappropriate responses were included in this category).
Thirdly, there are those responses which are unrelated to
the sentence, as required by the task instructions (these
subcategories will be differentiated later).
Two or more raters agreed on 94.1% of occasions,
which is most satisfactory given that each response can
be assigned to one of eight classifications. In the
remaining 5.9% of responses, a fourth rater, blind to
the purpose of the study, was asked to decide which one
of the three raters was correct.
Error scoring.
An error score was devised which is a
measure of the overall semantic relatedness of each
response to its stimulus sentence (see Fig. 1). Three
points were given if the word was a straightforward
completion of the sentence (category C), and I point for
a word that was semantically related to the sentence in
some way (categories SO, SA, SB and SC). Categories
U, UR, UL and URL represent successful completion of
the task requirements and so were given no score. A
summary of the error score results is given in Table 3.
Under these conditions the difference in performance
between the control group, unilateral anterior group and
the posterior group was significant at the 0.002 level
[ANOVA, d.f. = 2, 91; F= 7.34]. A
Tukey-Kramerpost-
hoc
comparison showed the difference between the
anterior group and the control group is significant at
the 0.05 level [Q =4.1, r = 3, d.f. = 91], with the difference
between the anterior group and the posterior group
significant at the 0.01 level [Q=4.4, r=3, d.f.=91] (see
Table 3). Multiple linear regression showed that age and
IQ were significant predictors of performance as
measured by the error score. In order to check that
the significant anterior-posterior effect in this analysis
was not explicable in terms of these two factors, an
analysis of covariance was performed on the patients'
scores using age and FSIQ as covariates. Under these
conditions, the anterior-posterior difference was sig-
nificant at the 0.005 level.
Investigation of the effects of lesion involvement
P. W. Burgess and T. Shallice/Response suppression following frontal lobe lesions 267
Step Classification
1. Does the word reasonably complete C
the sentence (i.e. it's a word you might
give yourself if asked to provide a
word that would fit at the end).
2. Is the word an opposite of what you SO
might expect as an answer?
3. Is the word obviously semantically SA
connected to the subject of the sentence?
4. Is the word obviously semantically SB
connected to the expected response?
5. SC
6.
7.
8.
9.
Does the word vaguely fit at the end
of the sentence, but in a way that
makes the sense of the sentence
ludicrous - or is the word a slang
semi-obsenity?
If the word completely unconnected to
the sentence:
Is the word an item you might find in UR
a hospital office?
Is the word semantically connected to UL
the subject's last response?
Is the word both an item you might URL
find in an office, and is also related to
the
last answer?
Are none of the above true? U
Fig. 1. Rater's classification guide.
Error
score
[anterior or posterior; one or two lobes affected;
laterality; lesion type (tumour, vascular, other)] by
multiple regression revealed that only the anterior/
posterior classification was a significant predictor of
error score [t-ratio 2.72, P<0.01].
Analysis of variance of each subjects' mean latencies
on section B (log transformed) according to the three
main groups of error type (e.g. completions, semanti-
cally connected or unconnected) described above
revealed no significant differences, despite some tend-
ency towards longer response latencies in patients with
frontal lobe involvement. In other words, none of the
groups showed significantly different response latencies
on those trials where they produced a word which was
related to the sentence, thus breaking the task require-
ments. The group x error type latencies interaction,
similarly, did not approach significance.
Responses revealing the use of strategies. Further
analyses were carried out on the third category of
Table 3. Error scores across groups (S.D. in brackets)
Group Error score (max 45)
Controls 4.2 (5.7)
Posteriors 4.8 (4.8)
Anteriors 11.5 (11.3)
Bifrontals 19.4 (11.1)
responses (i.e. production of words unrelated to the
sentence, as requested). Observation of the performance
of controls suggested that two strategies were used
frequently when the subject was performing the task.
The first method for producing unrelated words was to
look round the room, choosing objects within sight and
name these in response to the stimulus sentences. The
second method was for the subject to generate a
semantic category and choose members of that category
as responses. (In both cases the responses could be
checked before being produced that they satisfied the
test conditions.) Raters were therefore asked to classify
responses unrelated to the sentence into one of four
additional categories. First they had to judge whether
the completion responses were objects normally found
in a standard hospital office/room (UR). Second they
had to judge whether the words were related to the last
answer they gave (UL). Obviously on occasions
responses might satisfy both conditions, and so there
was a category (URL) for such responses. If the word
was unrelated to the sentence, was not an object
normally found in an office and was not semantically
related to the last response, it was classified as U (see
Fig. 1).
The unilateral anterior group, the posterior group and
the control group differed in the proportion of their
responses that fell in one or other of these categories,
268 P. W. Burgess and T. Shallice/Response suppression following frontal lobe lesions
Table 4. Response latencies by response type across groups (mean sec; S.D. in brackets)
Group Completions Semantically related Unrelated Response failures*
Controls 2.1 (2.0) 3.3 (4.6) 1.6 (2.4) 0
Posteriors 5.0 (8.2) 3.4 (3.5) 2.5 (3.6) 0
Anteriors 7.4 (9.0) 6.6 (9.0) 3.4 (5.2) 9
Bifrontals 6.9 (9.3) 7.7 (9.8) 8.7 (10.6) 6
*This column gives the numbers of occasions that a member of the group was unable to produce
any response within 60 sec.
namely in being either semantically related to their last
response, or being objects normally found in an office,
or both [Kruskal-Wallis H=6.53, d.f.=2, P<0.05].
Multiple comparison procedures showed that the uni-
lateral anterior group produced a significantly smaller
proportion of these types of response (at the <0.05
level) than either the posterior group or the control
group, which did not differ significantly from each other
(see Table 5). The proportion of the responses that came
from the strategy categories was, on average, less than
half for the bifrontal group than for the unilateral
anterior group.
Clearly however, as the anterior patients in section B
showed a greater tendency to complete the sentence,
whilst at the same time showing decreased evidence of
strategy use, these characteristics may not be independ-
ent; there may be some characteristic of the task which
leads to a trade-off between them. This possibility was
explored using factor analysis. The two types of task
failure (either straightforward completion, or the
production of one of the four types of semantically
connected response--SO, SA, SB, SC) were considered
alongside the two forms of task success (either produ-
cing a 'non-strategy' unconnected response (category U)
or one of the types of 'strategy' unconnected responses
(categories UL, UR, URL). The numbers of instances of
responses falling into these categories were considered
across the unilateral patients and controls, using
principal components analysis.
The results indicated an inverse relationship between
occurrence of strategy-type answers and the two levels
of task failure. Thus number of strategy responses
loaded positively on the first principal component
(eigenvalue 2.05, explaining 51% of the variance), and
both number of completions and number of semanti-
cally connected responses loaded negatively. Number of
non-strategy but satisfactory (category U) responses was
comparatively unrelated to this component. However
they were positively loaded on the second principal
component (eigenvalue 1.3, explaining a further 32% of
the variance), while strategy answers (UR, UL, URL),
by contrast, had a negative loading. The two types of
error responses were not loaded on this factor (see Table
6).
The factor loadings suggest a complex relationship
between strategy use and completions of the sentence
(i.e. failure of response suppression) in the Hayling task.
Removing the variance explicable by group from the
raw data by regression techniques before the principal
component analysis is carried out did not alter the factor
structure, suggesting that this complex causal relation-
ship is a characteristic of the specific demands made by
the Hayling task rather than being a characteristic of
patients who are unable to perform it well. There was,
however, a significant relationship between the
first
principal component factor scores (derived for each
individual) and group [F=5.74, d.f.=2, 91, P<0.005]
with this effect largely attributable to the anterior/other
group comparison [t-ratio - 3.15, P<0.005]. In other
words, the performance of the anteriors was charac-
terised by a combination of decreased strategy use
together with simultaneous increase in responses
Table 5. Percentage responses in each category across groups
C SO SA SB SC UR UL URL U*
Cont 4.2 2.7 2.5 1.5 5.8 31.5 5.0 1.7 45.2
Posts 4.7 3.7 2.7 3.7 9.6 27.2 8.9 1.7 37.5
Ants 16.9 8.2 3.4 3.1 10.9 24.8 2.7 1.1 27.7
BiFr 32.2 11.0 6.7 3.1 12.5 8.2 3.9 0.4 22.7
Section totals
Completions Semantically related Unrelated
Cont 4.2 12.5 83.3
Posts 4.7 19.8 75.3
Ants 16.9 25.7 56.2
BiFr 32.2 32.5 35.3
*See Fig. 1 for explanation of categories.
Cont = controls; Posts = unilateral posteriors; Ants = unilateral anteriors; BiFr = bifrontals.
P. W. Burgess and T. Shallice/Response suppression following frontal lobe lesions
Table 6. Summary of principal component loadings across response types
269
PC 1 PC 2
Failure type Completions - 0.54 NS*
(Error score 1 or 3) Semantically connected -0.57 NS
Success type Strategy responses 0.57 -0.82
(Error score = 0) Non-strategy response NS 0.50
*Indicates factor loading < 0.3.
NS = not significant.
semantically connected to the sentence, thus breaking
the task rules (see Tables 5 and 6).
Response latencies in both sections of the task
The response latencies of all groups were significantly
longer on section B, compared with section A [paired t-
tests, all groups P<0.005]. A one-way ANOVA of the
unilateral patients and controls revealed differences in
response latencies significant at the P<0.01 level
[F=6.39, d.f.=2, 91], with Tukey-Kramer
post-hoc
comparisons showing the anterior-control comparison
to be significant at the P<0.01 level; none of the other
comparisons were significant.
The effects in the unilateral patients, of lesion size
(affecting one or two lobes), laterality (left or right
hemisphere), location (anterior or posterior) and type
(tumour, vascular or other) were investigated using
multiple linear regression (backward elimination). Of
these factors, only the anterior/posterior classification
was a significant predictor of response latencies in
section B [F=4.73, d.f. = 1, 72, P<0.05]. As with the
performance on section A, the bifrontal group were
significantly slower than the unilateral anterior group on
this section [following log transformation: t=2.92,
d.f. =36, P<0.01] (see Table 2).
One patient (with a right frontal lesion) was unable to
produce a word within 60 sec in eight of the 15 trials. A
number of other patients with frontal lesions failed to
produce words within the time limit on single trials (such
a failure to respond, which was never seen in section A,
was scored as a 'completion' for the error score
analyses). No patient with a posterior lesion or a
control ever showed this pattern.
B-A latencies.
As a measure of the difference that the
experimental manipulation in section B made, the
section B latencies minus the section A latencies were
considered for each subject (see Table 2). This pre-
sumably represents the additional 'thinking time' that is
required in having to produce a novel word rather than
a straightforward sentence completion. Obviously this
serves to remove the possible confounding factor of
initiation problems when considering the latencies in the
response suppression condition.
The B-A times for the unilateral anterior, posterior
and control groups were significantly different [F= 4.35,
d.f.=2, 91, P<0.01], with Tukey-Kramer
post-hoc
comparisons showing the anterior-control differences
to be significant at the P< 0.01 level. None of the other
comparions was significant. Moreover, the correlation
between the B-A latencies in the unilateral patients (i.e.
suppression condition minus initiation condition laten-
cies) and the proportion of responses revealing the use
of strategies was -0.45, which is significantly different
from zero [P<0.01, 2-tailed]. In other words, those
patients that showed longer thinking times tended to
produce fewer strategy responses.
Overall relation between performance on section A and
section B
Correlations between performance on section A
(response latency) and section B (error score) were
considered separately for each group. The controls and
posteriors showed correlations between the two sections
of the test which were significantly different from zero
[controls: r=0.593, P<0.01; posteriors: r=0.389,
P<0.05]. This was not the case for the unilateral
anteriors, where performance on section A appeared
unrelated to performance on section B [r=0.191,
P=0.2]. This pattern was confirmed by considering the
scores of individual patients. Using a strict criterion for
a dissociation between tasks as being where the subject
performs with 1 S.D. of the mean of the control group
on one measure, but worse than 3 S.D. below their mean
on another [see 19], four patients with unilateral anterior
lesions, one bifrontal and one with a posterior lesion
showed this pattern for good performance on section A
with poor section B performance, and one patient with a
unilateral anterior lesion showed the opposite pattern
(poor A with good B performance). Such low correla-
tions in performance might occur if the reliability of the
measure were low. However when the Spearman-Brown
split-half reliabilities for section A and section B (0.63
and 0.78, respectively) were taken into account, there
were individual cases (in the anterior group only) who
exhibited individual difference scores (following the
procedure outlined in [1] pp. 136-137, which calculates
the likelihood of occurrence of a given magnitude of
discrepancy between a subject's performance on two
different tests) that were significant at the 0.01 level in
favour of each section of the task. In addition, one
bifrontal showed a significant individual difference score
in favour of section A. Thus it seems that performance
270 P.W. Burgess and T. Shallice/Response suppression following frontal lobe lesions
on section A may be independent of performance on
section B.
On the correlation analysis the bifrontal group
showed a different pattern from the unilateral anteriors,
with performance on the two sections of the test being
highly related [r = 0.634, P< 0.01]. This difference in the
correlations between the unilateral anteriors and bi-
frontals is significant at the 0.05 level. The bifrontal-
unilateral anterior difference may suggest differences in
laterality of function between the two sections of the
test. However, within the unilateral anteriors there was
no effect at all of laterality of lesion on either parts of the
Hayling test (laterality of lesion explaining 0% of the
variance in a regression analysis for each section). It
would appear instead that the bifrontals tended to show
more uniform impairments across the test (10/17
bifrontals were more than 2 S.D.s below the controls
on section A, and 11/17 were at least 2 S.D.s below the
controls on section B).
Discussion
This study has two main findings. The first is that,
compared with a group of age- and IQ-matched patients
with lesions to other parts of the brain (the posterior
group), patients with lesions that included the frontal
lobes (the anterior group) showed deficits on a measure
of response initiation. The second finding is that the
anterior group were also poorer than the posterior
group on a task which involved suppression of the
habitual response. In subsidiary analyses it was shown
that a bilateral anterior group had significantly greater
deficits than the unilateral anterior group.
In comparision with the patients with posterior
lesions, the patients with anterior lesions performed in
a qualitatively poorer fashion. They made significantly
more straightforward completions of the sentence, and
those answers they gave which were not completions
nevertheless had a higher number judged to be
semantically related to the sentence.
Turning to the latency of responding, the anterior
group were significantly slower than the posterior group
in section A. When the latencies in the two sections were
compared, all groups took longer in section B than
section A. Since all characteristics of the task are the
same in the two sections except that the subject must
produce a word unconnected to the meaning of the
sentence in section B, latency B minus latency A
presumably gives a measure of the thinking time in
section B. The two unilateral lesion groups did not differ
significantly on this measure.
In the context of the anterior group's qualitatively
worse performance this means that the anterior group's
thinking time was used less effectively in fulfilling the
task requirements. However, although the unilateral
anterior group's performance was poorer than that of
the posterior group on two different aspects of the
task--the overall error score on part B and the speed of
responding on part A--the relationship between per-
formance on the two sections was not the same for all
groups. Thus whilst performance on section A was
related to section B performance for the controls and
posteriors, this was not the case for the unilateral
anteriors. The bifrontal group, by contrast, was
generally poor on both sections of the test.
One possible explanation of the anterior patients'
deficits on the Hayling test is that they might be
suffering from a greater degree of generalised cognitive
impairment than the posterior group. However, this can
be rejected as the two groups were matched for
performance on a standard neuropsychological measure
of general intellectual functioning (the WAIS).
A second possible explanation of the results might be
that the anterior group performed the tasks badly due to
distractibility or to motivational problems. However,
this seems implausible since the anterior group did not
perform other neuropsychological tests poorly (e.g.
WAIS). An explanation of the results therefore needs
to be based upon the way the tests employ particular
processing resources or abilities that are damaged in
patients with lesions primarily in the frontal lobes.
Perret [16] suggested that the high correlation in his
study between performance on the Verbal Fluency test
(VFT) and the Stroop task suggests that both VFT and
the Stroop task tap the same process--suppression of
the habitual response. The alternative hypothesis was
that VFT requires good initiation or verbal planning
abilities and the Stroop requires a different process, the
suppressing of a prepotent response, but that initiation
and suppression abilities tend to be damaged together
because they are part of the same system or that they
tend to be impaired together due to the anatomical
proximity of the cortical areas subserving these pro-
cesses.
However, the findings of the present study suggest
that initiation or verbal planning abilities and response
suppression may be impaired singly in unilateral
anteriorly lesioned patients. These findings seem un-
likely to be due to any overall lack of reliability in either
a response time measure
per se
or in the error score in
part B, as the error score correlated with (B-A) response
times at the 0.01 level. In addition, when reliability was
taken into account statistically, individual anterior
patients still showed significant individual difference
scores [1] on the two measures. In any case, low task
reliability is unlikely to be a good explanation of the
double dissociations noted within single cases [see 9, p.
158]. Instead, the most likley explanation of the present
results is that the two processes are separable and that
where both processes are impaired, they are so because
the cerebral areas which subserve such processes are
damaged together.
Why then should there be a positive correlation in the
bifrontal group? One possibility is that medial frontal
lesions, which would of course be almost inevitably
P. W. Burgess and T. Shallice/Response suppression following frontal lobe lesions 271
produced in this group, give rise to general initiation
problems which affected performance on both sections
of the task. An alternative explanation could be given in
terms of lesion size effects. Neither of these accounts are
likely to be satisfactory in explaining the positive
correlation in the posterior group, however. Here it is
more plausible that the correlation is a result of deficits
in the processing requirements of the task which are
incidental to the main focus of this study. For instance,
mild sub-clinical aphasic problems would be expected to
affect sections of the task roughly equally. No doubt
many other forms of disorder would act similarly.
However, there is a complicating factor for the
preceding discussion. More detailed examination of
performance on section B shows that an important
factor in the way the different groups performed in
section B is how well they employed strategies appro-
priate to the task. A failure to utilise an appropriate
strategy when dealing with a novel task has often been
treated as one aspect of the patient's difficulties [e.g. 13,
20]. However, it has been rare for the utilisation of
appropriate strategies to be actually measured in frontal
lobe group studies, the only case known to us being the
study of Owen
et al.
[15].
In the present situation control subjects frequently
employed some heuristic which enabled them to
generate a word likely to be unrelated to the sentence
stem. Among the many possible heuristics available
two strategies were commonly used by control
subjects. One was to pick objects from the surround-
ings, and the other was to generate a semantic
category, and then choose exemplars of this category
as responses, maybe after checking that it produced
an appropriate result. A strategy measure was there-
fore employed of the frequency with which subjects
produced the names of items typically found in
hospital offices or words related to the previous
response. The anterior group demonstrated a greater
tendency towards completion of the sentence, whilst
simultaneously showing a decrease in responses
showing evidence of strategy use.
One possibility is that the frontal group might be
unable to inhibit inappropriate responses and thus
differed on the strategy measure because they did not
have the time to operate the strategy. In fact there was a
significant negative correlation between B-A latencies
and the production of strategy-indicating responses. The
patients who did not employ strategies effectively took
relatively longer on task B than task A. These results
therefore do not support the lack of time hypothesis.
Instead they raise the possibility that the cause of the
anterior group's poor task performance was their
inability to acquire or realise an appropriate strategy
and not being merely due to a deficit in response
suppression abilities. In performing the Hayling task
there would then be a necessary reciprocal causal link
between strategy generation and response suppression
abilities. Indeed it is plausible that the first factor
derived in the principal components analysis reflects this
reciprocal link.
The extent to which the factor of strategy use is
critical in other situations where a frontal deficit in
prepotent response suppression has been held to be
important e.g. Stroop [16] or Wisconsin Card-Sorting
[11] remains to be investigated. In the present situation,
however, the critical unilateral anterior group show no
significant relation between their performance on sec-
tion A and that on section B. There appears to be little
necessary relation other than the possible anatomical
contiguity of their material substrates, between pro-
cesses that concern strategy generation and realisation
and those concerned with initiation or verbal planning
that are involved in straightforward sentence comple-
tion.
Acknowledgements--We
would like to thank Professor E. K.
Warrington for providing the facilities for this research to be
carried out, and two anonymous referees for their helpful
comments on an earlier draft. This research was supported by
a grant from the U.K. Medical Research Council.
References
1. Anastasi, A.
Psychological Testing.
Macmillan, New
York, 1990.
2. Benton, A. L. Differential behavioural effects in
frontal lobe disease.
Neuropsychologia 6,
53-60,
1968.
3. Bloom, P. A. and Fischler, I. Completion norms for
329 sentence contexts.
Mem. Cognit.
38, 631-642,
1980.
4. Costello, A. de L. and Warrington, E. K. Dynamic
aphasia: The selective impairment of verbal
planning.
Cortex
25, 103-114, 1989.
5. Crawford, J. R., Allan, K. M., Besson, J. A. O.,
Cochrane, R. H. B. and Stewart, L. E. A
comparison of the WAIS and WAIS--R in matched
UK samples.
Br. J. clin. Psychol.
29, 105-109, 1990.
6. Drew, E. A. Go-no go learning after frontal lobe
lesions in humans.
Cortex
11, 8-16, 1975.
7. Kleist, K. Gehirnpathologie. Barth, Leipzig, 1934.
8. Luria, A. R.
Traumatic Aphasia.
Mouton, The
Hague, 1970.
9. McNemar, Q.
Psychological Statistics.
John Wiley,
London, 1964.
10. Miller, E. Verbal fluency as a function of a measure
of verbal intelligence and in relation to different
types of cerebral pathology.
Br. J. clin. Psychol.
23,
53-57, 1984.
11. Milner, B. Effects of different brain lesions on card
sorting.
Arch. Neurol.
(Chicago) 9, 90-100, 1963.
12. Milner, B. Brain mechanisms suggested by studies
of the temporal lobes. In
Brain Mechanisms
Underlying Speech and Language, C. W. M.
Whitty and O. L. Zangwill (Editors), pp. 122-145.
Grune and Stratton, New York, 1967.
272 P.W. Burgess and T. Shallice/Response suppression following frontal lobe lesions
13. Milner, B. Petrides, M. and Smith, M. L. Frontal
lobes and the temporal organization of memory.
Hum. Neurobiol.
4, 137-142, 1985.
14. Nelson, H. E. and O'Connell, A. Dementia: The
estimation of pre-morbid intelligence levels using the
new adult reading test.
Cortex
14, 234-244, 1975.
15. Owen, A. M., Downes, J. J., Sahakian, B. J., Polkey,
C. E. and Robbins, T. W. Planning and spatial
working memory following frontal lobe lesions in
man.
Neuropsychologia
28, 1021-1034, 1990.
16. Perret, E. The left frontal lobe in man and the
suppression of habitual responses in verbal catego-
rical behaviour.
Neuropsychologia
12, 323-330,
1974.
17. Ramier, A. M. and Hecaen, H. Role respectif des
atteintes frontales et de laterallisation lesionelle
dans les deficits de la "fluence verbale".
Rev.
Neurol. (Paris)
123, 17-22, 1970.
18. Shallice, T. Specific impairments of planning.
Phil.
Trans. R. Soc. Lond.
B295, 199-209, 1982.
19. Shallice, T.
From Neuropsychology to Mental
Structure.
Cambridge University Press, New York,
1988.
20. Shallice, T. and Evans, M. E. The involvement of
the frontal lobes in cognitive estimation.
Cortex
14,
294-303, 1978.
21. Verfaellie, M. and Heilman, K. M. Response
preparation and response inhibition after lesions
of the medial frontal lobe.
Arch. Neurol.
44, 1265-
1271, 1987.
22. Wechsler, D.
Wechsler Adult Intelligence Scale.
The
Psychological Corporation, New York, 1955.
Appendix
Hayling sentence completion test--raters' instructions and examples of test items
[Raters were also given the 'decision tree" shown in Fig. 1.]
The sheets you have been given have two sets of 15 sentences on them. We are only interested in scoring the SECOND SET of
sentences.
Each subject had each sentence read out to them in turn, and the response the subject made is written at the end of that sentence.
We would like you to rate each of these words according to the "decision tree" below.
Examples:
C rating:
(a straightforward completion of the sentence)
The captain wanted to stay with the sinking BOAT or SHIP
He bought them in the sweet SHOP or STORE
SO rating:
(an opposite of what might be expected)
They went as far as they COULDN'T
The whole town came to hear the major REMAIN SILENT
SA rating:
(a word obviously semantically connected to the subject of the sentence)
Most sharks attack very close to FISHES or SEA or BITE
She called the husband at this PHONE or WIFE or DIAL
SB rating:
(a word showing an obvious semantic connection to the word(s) that you would expect to appear at the end of the
sentence)
None of the books made any UNDERSTANDING
Most cats see very well at DAWN or DUSK or WINTER
SC rating:
(a word which makes vague sense at the end of the sentence but which makes the sentence ludicrous. Also swear-words,
obscenities or another inappropriate word.)
The whole town came to hear the major FART
The dog chased our cat up the TOWER OF PISA
UR rating:
(a word that is completely unconnected to the sentence
and
which might reasonably be expected to be found in a normal
office room--including office fixtures and fittings and clothes that someone in an office might wear)
The dough was put in the hot COATHANGER/PAPERCLIP
Jean was glad the affair was DESK/CHAIR/WINDOW/DOOR
UL rating:
(a word that is unconnected to the sentence
and
which is semantically connected to their response to the last sentence.
NB----cannot apply to sentence 1)
Most cats see well at TALK (where previous answer was
CONVERSATION)
The whole town came to hear the major SANDCASTLE (where previous
response was SEA)
URL rating:
this is where both UL and UR are true.
U rating:
a word that is unconnected to the sentence and which is not connected to the subject's response to the previous sentence
and you would not generally expect to find in an office.
