Neuropsychologia,

1973,Vol. 11,pp. 159to 165.PergamonPress.Printedin England.

MEMORY CODING AND AMNESIA

A. D. BADDELEY * and ELIZABETH K. WARRINGTON

The National Hospital, Queen Square, London, W.C.l, England

(Received 10 January 1973)

Abstract-Memory coding in amnesic and control Ss was studied by comparing the free recall
of lists of unrelated control words with recall of lists of words grouped on the basis of phonemic
similarity, taxonomic category membership or a composite visual image. Amnesic .Ss gained
no advantage from visual imagery coding, although all claimed to be able to form the necessary
image. They did, however, take advantage of both phonemic similarity and taxonomic category
grouping and showed unimpaired ability to generate words, on the basis of either sound or
taxonomic category membership. The results are discussed in terms of the proposal that seman-
tic memory comprises two separate systems, one concerned with linguistic relationships, the
other with imagery.

THERE HAS, in recent years, been a growing body of evidence suggesting that the retention
of verbal material is crucially dependent on the way in which it is coded or processed when
presented, with semantic coding leading to much better retention than does coding in
terms of the sound or physical characteristics of the material [l-4]. It has been shown
that amnesic patients with grossly defective long-term memory may have intact short-term
memory [5-71. The present study explores the possibility that the specific long-term memory
defect of amnesic patients may be associated with an inability to utilize semantic coding
adequately. It therefore compares the influence of phonemic and semantic coding on the
verbal memory of amnesic patients.
A previous attempt to investigate memory coding in amnesics by examining intrusion
errors [S] proved unsuccessful since the tasks used led to many phonemic intrusions but
virtually no semantic intrusions in either amnesic or control patients. The present study
therefore attacks the problem more directly using the technique of category clustering [8]
to investigate phonemic and semantic coding. Phonemic coding is studied by comparing
the retention of 16-word lists comprising four groups of four phonemically similar words,
with the retention of 16 phonemically dissimilar words. In order to take advantage of the
clustering, S must encode the words phonemically. Any impairment in this encoding should
therefore reduce the advantage of the clustered list over the control list.
The assessment of semantic coding presents rather more of a problem since there is no
general agreement as to exactly what is meant by semantic memory. It has recently been
suggested that semantic memory comprises two separate but related systems, one based on
linguistic relationships and the other based on imagery [9, lo]. While there are as yet no
generally accepted techniques for studying these two systems, as a first approximation, we
used taxonomic category clustering to study coding in the linguistic system, and coding
in terms of visual imagery to study the non-linguistic system.

*Present address: A. D. Baddeley, Department of Psychology, University of Stirling, Stirling, Scotland.

159
160 A. D. BADDELEYand ELIZABETHK. WARRINGT~N

METHOD
Phonemic clustering
This involved three experimental and three control lists. Both types of list were based on sets of phonemi-
tally similar monosyllables, which were assigned at random to either experimental or control lists. The
three experimental lists comprised four blocks of four phonemically similar monosyllables (e.g. Bag, Tag,
Sag, Rag/Hill, Pill, Will, Till/Rose, Rope, Road, Roam/Grape, Grade, Grace, Grate). Control lists were
constructed using sixteen dissimilar sets and assigning one word from each set to each of the three lists
(e.g. Coil, Four, Found, Gold, Cot, Way, Heat, Can, Brag, Nun, Rage, Wipe, Joy, Wing, Try, Steep).
Each word was written on a 5 x 3 in. flash card, cards were presented manually at a rate of one per set
and the stimulus words were spoken simultaneouslyby E.Each list was presented four times without a break,
followed by a 30 set interpolated task to minimize PM effects and a one min recall period during which S
attempted to recall as many of the 16 words as possible in any order. If S failed to recall items from any
category, he was cued verbally with the first word of the block of four words from that category. In the
control condition hewas cued using theword in the equivalent presentation position (i.e. the first, fifth, ninth,
or thirteenth word if S failed to recall items 14, 5-8, 9-12, or 13-16 respectively). As an intervening task S
was presented with single digits each written on a flash card; he was required to add one to each digit and
give the answer, at which point E presented a further digit. Experimental and control lists were alternated,
with half the Ss beginning with an experimental and half with a control list.

Phonemic generation
The accessibility of words classified on a phonemic basis was assessed using the technique of word genera-
tion [S]. Subjects were given 30 set in which to call out as many words as possible that were similar in sound
to the target word. This task followed the free recall conditions and used as target words Pit, Clay and
Trout.

Tuxonomic category chutering

This involved lists clustered on the basis of taxonomic category membership. The lists were constructed
in the same way as the acoustic lists except that the basic sets were items drawn from the COHEN, BOUSFIELD
and WHITMAR~H[ll] taxonomic category norms by taking four frequent words (usually the lst, 3rd, 5th
and 7th items) from each of 28 categories (e.g. professions, animals, types of building, etc.). Experimental
lists comprised four blocks of four such items. Control lists each comprised one item from each of 16 taxono-
mic categories. Presentation and test procedures were identical with those used for the acoustically clustered
lists. Half the Ss began with the acoustic clustering condition and half with the taxonomic category condition.

Taxonomic category generation

This followed the phonemic generation condition and used the same procedure, except that Ss were asked
to generate on the basis of taxonomic category, using girls’ names, birds and metals as target categories.

Visd imagery
The third condition was based on the visual imagery strategy shown by BOWER,LE~COLDand TIEMAN
(1969) to be an effective mnemonic aid for college student Ss. It involves linking words to form a striking
image which S is instructed to visualize. Thus, the words Irishman, Penny, Monkey, Violin, would be linked
together as “The Irishman gave a Penny to the Monkey playing the Violin” which S was told to visualize.
Ashe described the scene to be visualized, E presented.&e four-words relevant to each scene on flash cards
at a rate of one word per set, vocally emphasizing each of the critical words. There was then a pause of I I
set during which S attempted to form the appropriate image, after which the next set of four words was
presented in a similar manner. Each set was presented only once, so that total presentation time was the
same as in the previous conditions in which each item was presented four times but with no pauses. A pool
of four-word sets that could be linked to form images was generated and sets were assigned at random to
experimental or control lists. Since the experimental condition essentially involved teaching S a mnemonic
system, we were anxious to avoid influencing performance on the other conditions or on the control lists. FOI
this reason, the imagery experiment was always run after the other two, and instead of alternative experimen-
tal and control lists an ABBA design was used in which S began with two control lists, went on to do four
imagery lists and finished with two more control lists. Furthermore, the control list presentation procedure
was arranged so as to minimize the possibility of using imagery by using the same rapid presentation as was
used in the other conditions. BUGELSKI,KIDD and SEGMEN[13] have shown that rapid presentation dis-
courages the use of imagery. This seemed to be effective in suppressing the mnemonic technique since there
was no apparent difference between performance on the first two and the last two control lists. It might be
argued that the control condition used was inappropriate since it confounds imagery instructions with
rate of presentation. Unfortunately, given the constraints described, with too few Ss to allow an indepen-
dent group design, some confounding is inevitable. However, our main aim is not to show that imagery is
helpful (there is ample evidence of this already, e.g. PAIVIO [lo]; BOWER [9], but rather to compare the
 MEMORY
CODINGANDAMNESIA 161

ability of two types of S to take advantage of imagery instructions, given the appropriate conditions. Recall
and cueing procedures were identical with the prior conditions.

Subjects
Six amnesic Ss were selected on the basis of clinically diagnosed amnesia with the absence of any intel-
lectual impairment other than a memory defect (see BADDELEYand WARRINGTON[5], for further details).
Control Ss were patients suffering from peripheral nerve lesions. They were matched with amnesic Ss for
age, sex and intelligence as measured by performance on the block design and vocabulary sub-tests of the
WAIS.

RESULTS

Recall data before and after prompting are shown in Table 1.

Phonemic clustering
All six amnesic Ss performed better on the clustered lists than on the control lists, as
measured by total words correct before and after prompting. (On the Wilcoxon test T = 0,
N = 6, p <O-O5 in each case). The equivalent effects also reached significance for control
patients although one S did not show an effect (T = 1, N = 6, z = 1.99, p ~0.05). Com-
parison between the amnesic and control groups using the Mann Whitney test showed better

Table 1. Mean number of words recalled initially and after prompting by amnesic and control subjects
Amnesic subjects Control subjects
Clustered Control Clustered Control
lists lists Diff. lists lists Diff.
Initial 10.5 $2) 3.0 19.1 12.8 6.9
Phonemic lists recall (2.39) (3.50) (4.05)
(Max = 48) Total 15.0 7.5 21.7 13.1 8.6
recall (5.03) $2) (2.14) (3.89)

Initial 12.5 8.2 4.3 29.9 14.5 15.4

Taxonomic lists Recall (4.07) (4.60) (4.21) (3.10)
(Max = 48) Total 16.0 (& 7.2 33.7 14.6 19.1
recall (3.96) (5.06) (3.04)

Initial 9.5 8.8 0.7 33.5 21.2 12.3

Imagery lists recall (2.66) (3.81) (7.02) (7.02)
(Max = 64) Total 10.7 9.3 1.4 38.3 21.5 16.8
recall (2.87) (3.40) (8.42) (7.41)
(Standard deviations in parentheses)

total performance by control Ss on both clustered lists (U = 4.5, N, = Nz = 6, p <O-5)

and control lists (U = 3, N1 = N, = 6, p<O-02). There was, however, no difference in the
extent to which amnesic and control Ss were helped by clustering as measured by the
difference between clustered and control list performance (U = 9, N, = N, = 6, p >O*lO).
It appears then that amnesic patients perform at a somewhat lower overall level on this
test but do not differ markedly in the extent to which they are assisted by acoustic clustering.

Taxonomic category clustering

Amnesic Ss performed significantly better on clustered than on control lists as measured
by words correct both before, (T = 1, N = 6, z = 1.99, p = 0.05), and after prompting
(T = 0, N = 6, p <0.05). Control patients also performed better on recall both before and
after prompting (T = 0, N = 6, p <0.05 in both cases). Although control Ss performed
somewhat better than amnesic Ss on control lists, this difference was not statistically signi-
 162 A. D. BADDELEYand ELIZABETHK. WARRINGTON

ficant (U = 8.5, N, = N, = 6, p>O*l), but a very reliable difference in favour of control

Ss occurred on clustered lists (c’ = 0, N1 x N, = 6, p ~0.005). There is also a significant
difference in clustering effect as measured by the difference between control and experi-
mental list recall (U = 2.5, N1 = N, L= 6, p <0.02).* It seems then that both amnesic and
control Ss are assisted by taxonomic clustering. Control Ss appear to benefit more from
clustering than amnesic Ss though in view of the smallness of the sample and the difference
in overall level, this last result should perhaps be treated with caution.

Imagery clustering
The results of this condition are also shown in Table 1. It should be noted that since the
design involved four control and four experimental lists rather than three, the possible
maximum score is 64 rather than 48. Although all the amnesic Ss claimed to be able to
form the relevant images, they did not recall significantly more words from imagery lists
than from control lists either before prompting (T = 5, N = 6, p >0.05) or after (T = 5,
N = 6, p >0.05). In contrast to this, control Ss all performed better on imagery than on
control lists, in terms of recall both before prompting and after (T 1 0, N =: 6, p co.05
in both cases). A comparison between amnesic and control Ss showed significantly better
performance by control Ss on both control lists (U = 2.5, N, N, = 6, p <0.02) and
imagery lists (U = 0, N, = N2 = 6, p (0.005) together with a greater imagery facilitation
effect, as measured by the difference between recall on imagery and control lists (II == 0,
Ni = PJ? = 6, p <0.005).

INTRUSIONS

Distribution of intrusions in the various conditions is shown in Table 2. None of the

differences between clustered and control lists are statistically significant; there is however, a
reliable tendency for amnesic Ss to make more intrusions than control Ss. This applies to the

Table 2. Total intrusions by amnesic and control subjects

Amnesic subjects Control subjects
Clustered Control Clustered Control
lists lists lists lists
Phonemic lists 62 74 39 36
Taxonomic lists 50 62 17 14
Imagery lists 49 76 13 19

phonemic condition (U = 5, Ni == N, --= 6, p <0.05), the taxonomic category condition

(V = 4, N, == Nz = 6, p (0.025) and the imagery condition (U = 2.5, N, m=m N, = 6,
p (0.025). The theoretical implication of the higher intrusion rate of the amnesic group
is not clear. While it may reflect increased interference (WARRINGTON and WEISCRANTZ,
1970), it could equally well be simply an attempt to cope with a situation in which recall
responses are required but not available, an experimental equivalent of the tendency to
confabulation which has sometimes been attributed to Korsakov patients [14].

*A percentage score might seem preferable, but is not appropriate with large differences in baseline levels
of performance. This can be illustrated using the case of improvement due to acoustic clustering. If an
items correct score is used, amnesic Ss appear to improve more (100 per cent) than do control Ss (68 per
cent). Scored in terms of the reduction in errors however, the pattern appears to reverse with amnesic Ss
helped less (18.5 per cent) than control Ss (24.7 per cent).
 MEMORY CODING AND AMNESIA 163

Word generation
Table 3 shows the mean number of words generated in 30 set on each of the tests for
amnesic and control Ss. There is clearly no difference between the two groups of patients
in their ability to produce items from the relevant categories.

Table 3. Mean number of words generated in 30 set by

amnesic and control subjects

Amnesic subjects Control subjects

Phonemic 5.2
generation (1.60)
Taxonomic 7.6
generation (1.65) (0%

(Standard deviations in parentheses)

DISCUSSION
The most striking feature of these results is the almost complete inability of the amnesic
Ss to gain any advantage from visual imagery coding. This suggests a previously unsuspected
defect in utilizing a type of memory coding which has proved to be extremely powerful
for normal Ss. We do not as yet, however, have evidence as to why the amnesic Ss gained
so little from imagery. All Ss claimed to be able to form the images, although this was not
the case in a pilot study in which particularly bizarre imagery combinations were used.
If the amnesic patients are to be believed therefore, the defect presumably occurs at some
stage beyond that of image formation. The work of BOWER and his associates, [15] suggests
that visual imagery assists memory only when the images to be associated are visualized
in an interactive way. Hence, in a paired-associate task if S visualizes the stimulus items
on one wall of the room, and the response items on another, no imagery facilitation occurs.
This suggests that the value of visual imagery may lie in its flexibility, in that it enables S to
produce an interactive relationship between a very wide range of items. If this is the case,
it is possible that our amnesic Ss do not show a defect in imagery per se, but rather fail
to gain the normal benefit from constructing a new relationship or episode. Hence, further
elucidation of the defective use of imagery in amnesic Ss is tied to the more basic questions
of what aspect of semantic or imagery coding is so beneficial to retention, and why.
A second feature of our results is the apparent lack of impairment in the utilization of
phonemic coding. It is tempting to contrast this with the taxonomic coding condition, in
which amnesics gained significantly less from clustering than control Ss. Unfortunately,
however, the number of phonemically similar words that might potentially have been used
in the various categories is generally much smaller than the number of items in the taxo-
nomic categories used. This being so, the probability of guessing an item correctly is likely
to be greater for the phonemically clustered list, so that quantitative comparisons should
be treated with caution. At present, it would perhaps be wise to conclude only that am-
nesic Ss are able to take advantage of both phonemic and taxonomic coding, in contrast
to their inability to benefit from the visual imagery mnemonic.
Our results confirm the observation of WARRINGTON and WEISCRANTZ [16] that amnesic
Ss can take advantage of category cueing. The absence of any substantial cueing effect in
the control lists for either amnesic or control Ss (reflected in the absence of a difference
between intitial and total recall in Table 1) is consistent with the results of SLAMECKA [17]
164 A. D. BADDELEY and ELIZABETH K. WARRING~~N

who found that when S has recalled as many items as he can from a list, prompting him
with items he has failed to recall does not help him to retrieve further unrecalled items. Our
results (see Table 1) suggest that prompting with unrecalled items is useful in lists that are
both organised and blocked. The cueing effect is significant for all three types of organi-
zation (p <0.05 in each case), and there is no reliable difference in the magnitude of the
cueing effect between amnesic and control subjects. Since Slamecka found no prompting
advantage with unblocked categorized lists, this implies that blocking of items from the
same category is necessary if subsequent prompting is to be effective.
In conclusion, our results indicate that amnesic Ss with defective LTM show an im-
paired ability to utilize one type of semantic coding. They lend further support to the
suggestion that semantic memory comprises two separate systems, one concerned with
linguistic relationships and the other an imagery system.

Acknowledgements-We wish to thank the physicians of the National Hospital for permission to study
patients under their care and to report our findings. We are grateful to Dr. R. T. C. Pratt for providing
facilities to carry out this study, to Dr. T. Shallice for useful discussion, and to the British Medical Research
Council for financial support.

1. BADDELEY, A. D. The influence of acoustic and semantic similarity on long-term memory for word
sequences. Q. J. exp. Psychol. 18, 302-309, 1966.
2. BADDELEY, A. D. Retrieval rules and semantic coding in short-term memory. Psycho/. Bull, (In Press).
3. HYDE, T. S. and JENKINS, J. J. Differential effects of incidental tasks on the organisation of recall of a
list of highly associated words. J. exp. Psycho/. 82,472481, 1969.
4. SACHS, J. S. Recognition memory for syntactic and semantic aspects of connected discourse. Perceptiozz
and Psychophysics, 2, 47342, 1967.
5. BADDELEY, A. D. and WARRINGTON, E. K. Amnesia and the distinction between long and short-term
memory. J. verb. Learn. verb. Behav. 9, 176-189, 1970.
6. DRACHMAN, D. A. and ARBIT, J. Memory and the hippocampal complex II. Is memory a multiple pro-
cess? Arch. Neural. 15, 52-61, 1966.
7. WICKELGREN, W. A. Sparing of short-term memory in an amnesic patient. Neuropsychologia 6,235-244,
1968.
8. BOUSFIELD, W. A. The occurrence of clustering in the recall of randomly arranged associates. J. gen.
Psychol. 9,229-240, 1953
9. BOWER, G. H. Mental imagery and associative learning. In Cogfzition irz Learrriug arzd Memory. (Editor
L. GREGG. Wiley, New York. In Press).
10. PAIVIO, A. Imagery and language. In Imagery: Current Cognitive Approaches, S. J. SEGAL (Editor).
Academic Press, New York.
11. COHEN, B. H., BOUSFIELD, W. A. and WHITMARSH, G. A. Cultural norms for verbal items in 43 cate-
gories. Tech Report No, 22, 1957, University of Connecticut, Contract Nonr 631 (00), Office of Naval
Research.
12. BOWER, G. H., LESGOLD, A. M. and TIEMAN, D. Grouping operations in free recall. J. verb. Leanz. verb.
Behav. 8,481-493, 1969.
13. BUGELSKI, B. R., KIDD, E. and SEGMEN, J. Image as a mediator in one-trial paired-associate learning.
J. exp. Psycho!. 76, 69-73, 1968.
14. ZANGWILL, 0. L. The amnesic syndrome. In Anznesia. C. W. M. WHITTY and 0. L. ZANGWILL (Editors).
Butterworths, London, 1966.
15. BOWER, G. H. Organisational factors in memory. Cog. Psych. 1, 18-46, 1970.
16. WARRINGTON. E. K. and WEISKRANTZ. L. Organisational aspects of memory in amnesic patients.
Neuropsycholigia 9, 67-73, 1971. -
17. SLAMECKA, N. J. An examination of trace storage in free recall. J. exp. Psychol. 76, 504-513, 1968.

RBsumB-On a Btudie l’encodage mnCsique chez des sujets amnCsiques et chez des sujets con-
tr6les en comparant le rappel libre de listes de mots iontrales sans rapport entre eux, avec le
rappel de listes de mots group&s sur la base de la similarit& phonkmique, de l’appartenance taxo-
nomique ou 8 une image visuelle complexe. Les sujets amnCsiques ne tiraient aucun avantage
de l’encodage par l’imagerie visuelle bien que tous soutenaient qu’ils etaient capables de
former l’image nkcessaire. 11s tiraient cependant avantage des groupements par similarit&
 MEMORY CODING AND AMNESIA 165

phonemique et par categoric taxonomique et leur capacite de produire des mats en partant du
son ou de l’appartenance a la categoric taxonomique n’etait pas perturb&e. On discute ces
r&hats en proposant que la memoire semantique comprend deux systbme &pares, l’un en rap-
port avec les relations linguistiques, I’autre avec I’imagerie.

Zusammenfassung-Bei amnestischen Kranken und gesunden Versuchspersonen wurde die

Gedachtnisspeicherung untersucht. Die Priifung erfolgte in der Weise, da8 man eine Reihe von
beziehungslosen Wiirtern wiederholen m&e und das Ergebnis mit der Wiederholung von
Wortgruppen ahnlichen Klanges oder taxonomischer Zugehijrigkeit oder mit der Wiedergabe
eines zusammengesetzten optischen Bildes verglich. Amnestische Kranke hatten keinen Vorteil
durch das optische Bild, obgleich alle behaupteten, sie seien imstande die optischen Details
wiederzugeben. Andererseits war es vorteilhaft fur sie, wenn sie klanglich ihnliche und taxo-
nomisch verwandte Wortgruppen behalten m&en. Sie waren such unbehindert Worte zu
bilden, entweder auf der Basis ihnlichen Klanges oder im Rahmen der taxonomischen
Kategorie. Die Ergebnisse wurde dahingehend gedeutet, daB das Begriffsgedlchtnis sich auf
zwei getrennte Systeme aufbaut, das eine urnfaSt linguist&he Bereiche, das andere fugt auf
Erinnerungsbildem.