Download Independence of Input and Output Phonology in Word Processing

Journal of Memory and Language 41, 3–29 (1999)
Article ID jmla.1999.2637, available online at http://www.idealibrary.com on
Independence of Input and Output Phonology in
Word Processing and Short-Term Memory
Randi C. Martin, Mary F. Lesch, and Michael C. Bartha
Rice University
Recently, theorists have suggested a close relation between the codes activated during language
processing and those involved in STM. In order to further investigate this relationship, we examined
the performance of an anomic patient, MS, to determine whether he would exhibit the same
impairment in retrieving phonology from semantics in short-term memory that he shows in picture
naming. The results indicated a similar pattern of performance in naming and in STM tasks requiring
retention of output phonological codes (those involved in production). However, MS performed at a
high level on STM tasks requiring the retention of input phonological codes (those involved in
perception). The findings support the following conclusions: semantic information contributes to
STM, input and output phonological codes are separable and maintained in separate buffers, and the
same pathway that underlies retrieval of phonology from semantics in naming underlies feedback
from semantics in list recall. © 1999 Academic Press
In contrast to earlier models of short-term
memory (STM) which have focused solely on
the role of phonological and/or articulatory
codes (e.g., Baddeley, 1986), more recent conceptions of short-term memory also incorporate
a role for lexical and semantic representations.
These “language-based” models of STM have
emphasized the close relation between the representations and processes involved in word
perception and production and those involved in
short-term memory (Martin & Lesch, 1996;
Martin, Shelton, & Yaffee, 1994; Martin & Romani, 1994; N. Martin & Saffran, 1992, 1997;
N. Martin, Saffran, & Dell, 1996). The purpose
of the present study was twofold: (a) to investigate the relation between word perception and
production and short-term memory and (b) to
investigate the possible distinction between the
phonological codes involved in perception and
production.
CONCEPTIONS OF WORD PRODUCTION
AND COMPREHENSION
In order to lay out the relationship between
word processing and short-term memory, it is
necessary to consider current conceptualizations of the representations and processes involved in word production and comprehension
(see Fig. 1). In production, the speaker begins
with a concept to be conveyed which is encoded
semantically. The semantic representation is
used to select a lexical representation to convey
the intended concept and then this lexical representation must be encoded phonologically
prior to the onset of articulation. In spoken word
comprehension, the reverse sequence occurs.
Acoustic information is translated into a phonological representation which encodes phoneme
identity and order. This phonological representation then activates a lexical representation
which is connected to semantic information corresponding to that word.
It should be noted that in the language production and comprehension literature, there is
considerable debate concerning the nature of the
lexical representation. Some researchers (see
This research was supported by NIH DCD Grant
DC00218 to Rice University. We thank MS for all the time
and effort that he has given to this project and his family for
their cooperation. Michael C. Bartha is now at the Cognitive
Neuroscience Laboratory, Baylor College of Medicine,
Houston, Texas.
Correspondence concerning this article should be addressed to Randi C. Martin, Psychology Department - MS
25, Rice University, 6100 Main Street, Houston, Texas,
77005-1892. Electronic correspondence may be sent to
[email protected].
3
0749-596X/99 $30.00
Copyright © 1999 by Academic Press
All rights of reproduction in any form reserved.
4
MARTIN, LESCH, AND BARTHA
FIG. 1. Levels of representation in speech perception and
production.
Levelt, 1989) assume that lexical representations contain lexical semantic features (which
are distinguished from conceptual semantic features) as well as syntactic information, such as
word class, subcategorization frame, and grammatical gender (for languages where gender is
relevant). Other researchers (e.g., Levelt, Roelofs, & Meyer, in press) make no distinction
between conceptual and lexical semantics (both
are represented at the semantic level) and assume that the lexical representation contains
only syntactic information. Finally, other researchers (e.g., Dell, Schwartz, Martin, Saffran,
& Gagnon, 1997) assume that the lexical node
is essentially empty and serves only as a placeholder to connect all the information (phonological, semantic, syntactic, graphemic) that
pertains to a particular word (for yet another
position, see Miozzo & Caramazza, 1997). In
the present discussion, we will take the latter
view.
Interactive activation models such as the
TRACE model of word perception (McClelland
& Elman, 1986) and the speech production
model proposed by Dell (1986; Dell &
O’Seaghdha, 1992) incorporate these different
levels of representations and assume feedforward and feedback connections between levels.
For example, in Dell and O’Seaghdha’s (1992)
model of speech production (Fig. 2), semantic
features are connected to lexical representations, which are connected to phonemes. Activation flows through the network in response to
activation at the semantic level. For example,
suppose that the production task involves picture naming and the object to be named is
“lion.” The process of picture recognition
would lead to the activation of semantic features
of the pictured object. Activation of these semantic features would result in a spread of
activation from the semantic level to the lexical
level, and the appropriate lexical node would
become most activated. However, there would
also be weaker activation of other lexical representations that share semantic features (e.g.,
tiger, cat). Activation of these lexical representations would lead to activation of the target
word’s constituent phonemes and to weak activation of the constituent phonemes of the semantically related words. The model also includes feedback from the lexical to the semantic
level and from the phoneme to the lexical level.
Thus, since the lexical node for “lion” will be
most activated from the semantic input, feedback from the lexical node to the semantic level
will tend to reinforce activation of the appropri-
FIG. 2. Dell & O’Seaghdha’s (1992) model of speech
production (from Martin & Lesch, 1996).
INDEPENDENCE OF INPUT
ate semantic features. Similarly, since the phonemes of “lion” will be the most activated,
feedback from the phoneme nodes will reinforce activation of the appropriate lexical node.
However, some inappropriate nodes will be activated from feedback as well, but to a lesser
degree. Thus, the partial activation of the lexical
node for “tiger” would feed activation back to a
semantic feature for “has stripes,” and the activation of the phonemes for “lion” will cause
lexical nodes with similar constituent phonemes
(e.g., “line”) to receive some degree of activation.
An issue that arises when considering models
of language processing is whether the phonological forms involved in perception and production are different. Some models of speech
perception and speech production implicitly assume a separation of input and output phonological forms. For example, the modular model
of speech production proposed by Roelofs,
Meyer, and Levelt (1996; see also Levelt,
Roelofs, & Meyer, in press) assumes that in
word production a single lexical semantic/syntactic representation (termed “lemma”) is chosen prior to access to a lexical phonological
representation (termed “lexeme”). In this model
there is no feedback from the lexeme level to
the lemma level. In order to avoid such feedback, it is necessary to assume a separation
between input and output phonological forms. If
they were one and the same, then activation of
the lexemic representation should lead to automatic spread of activation back to the lemma
level.
A number of other theorists have argued on
the basis of neuropsychological studies and experimental studies with normal subjects that the
phonological forms involved in perception and
production are different (e.g., Howard & Franklin, 1990; Monsell, 1987; Shallice, McLeod, &
Lewis, 1985). For example, Shallice et al.
(1985) demonstrated that there was relatively
little interference between reading words aloud
(processing output phonology), and detecting a
specific name in a speech stream (processing
input phonology) whereas there was massive
interference between shadowing one message
and detecting a name in another (both require
5
processing input phonology). If there are separate input and output phonological forms, then it
would seem to follow that the forms used in
perception are more closely related to acoustic
specifications, whereas the forms used in production are more closely related to articulatory
specifications. For example, Jusczyk’s (1993)
model of speech perception assumes that acoustic features are abstracted from the input and
weighted according to their importance for
making distinctions in that language. On the
production side, Levelt (1989) suggested that
syllable size units are used in production and
these consist of the articulatory phonetic features for each syllable.
RELATION BETWEEN LANGUAGE
PROCESSING AND SHORT-TERM
MEMORY
Our conception of the theoretical underpinnings of memory span performance assumes
that there is a direct relation between the representations and processes involved in word perception and production and those involved in
short-term memory. Thus, according to our approach phonological, lexical, and semantic representations are activated in both language processing as well as in verbal short-term memory.
We further assume that: (1) all levels of representation in short-term memory depend on the
activation of long-term representations within
semantic memory (the knowledge store for information about words) and (2) these representations are activated at encoding and this activation is maintained during retention.
Two different possible instantiations of this
approach are shown in Fig. 3. In Fig. 3a, there
is one short-term memory buffer into which all
the levels of representation for each lexical item
are copied. This approach bears some similarity
to the programmable blackboard model of visual working memory presented by McClelland
(1986). In Fig. 3b, there are separate buffers for
phonemic and lexical–semantic representations.
This approach is similar to the model presented
by Barnard (1985) for verbal short-term memory. In either approach, we assume that the
long-term knowledge store for verbal information (on the left side) is connected to the buffers
6
MARTIN, LESCH, AND BARTHA
FIG. 3. Models of short-term memory incorporating separate lexical–semantic and phonological buffers. (a)
One buffer containing interactive representations. (b) Separate buffers with interactions only in the long-term
knowledge store.
such that activated information from the knowledge store continues to activate information in
the buffers and the activated information in the
buffers feeds back to keep the representations in
long-term knowledge store activated. It may
seem redundant to assume both persisting activation in the long-term knowledge store and the
existence of storage buffers. However, it is unclear how persisting activation in the long-term
store could serve to represent list information—in particular, how such activation could
be used to differentiate repetitions of the same
item in a list (e.g., a list such as “chair leaf chair
book”). There are additional reasons to prefer
the model in Fig. 3b over that in 3a. First, the
presence of several representations in the same
buffer, as in Fig. 3a, would seem to suggest that
these representations could interact, but it is not
clear that such an assumption is necessary. Second, the separation of the buffers provides a
more natural means of accounting for the selective impairment in phonological or lexical–
semantic retention that has been uncovered in
brain-damaged patients (e.g., Martin et al.,
1994, see below for further discussion). Third,
there are data from animal studies indicating
that separate areas of the brain are involved in
maintaining different visual features of an ob-
INDEPENDENCE OF INPUT
ject (such as shape vs. location) over a delay
(Wilson, Scalaidhe, & Goldman-Rakic, 1993).
There are also some neuroimaging data from
humans suggesting separate brain areas involved in working memory for spatial vs. object
information, though so far there has been a lack
of consistency in the anatomical localization of
these areas across studies (e.g., Baker, Frith,
Frackowiak, & Dolan, 1996; Smith, Jonides,
Koeppe, Awh, Schumacher, & Minoshima,
1995). Thus, these findings suggest that different brain areas are involved in the maintenance
of different types of information pertaining to
the same stimulus, and those brain areas involved in maintenance differ from the areas
involved in the primary processing of these
types of information.
Rather than assuming that the different levels
of representation in the short-term memory
buffer can interact, we assume, as in Dell and
O’Seaghdha’s (1992) model and the related approach to short-term memory presented by N.
Martin and Saffran (1992, 1997), that representations in the knowledge structure continue to
interact after an item has been presented such
that there is feedforward and feedback of activation between the different levels. Since activation in the knowledge structure at different
levels serves to activate the representations in
the buffers, the contents of the buffer(s) will
reflect the interactions in the knowledge structure. For instance, if high frequency items have
stronger lexical activations in the knowledge
structure, then high frequency items will tend to
have higher activations in the buffer. For word
list recall, lexical representations in the buffer
are selected for output. For nonword list recall,
activation of buffered representations at the
phonemic level will be used for recall.
According to our approach, if the representations or processes involved in word perception
and production are damaged, then short-term
memory will be affected as well, with predictable consequences depending on the particular
representation or process that is affected. For
example, if semantic representations are damaged, then feedback from the semantic to the
lexical level will be minimal in the knowledge
structure, and consequently the strength of lexi-
7
cal–semantic representations in the buffer will
be reduced. Consequently, the normal advantage of word over nonword list recall (Brener,
1940; Crowder, 1978; Hulme, Maughan, &
Brown, 1991; see below for further discussion)
will be reduced. We have also argued that it is
possible for word perception and production to
be preserved, but for the buffer to be affected
such that patients might show overly rapid decay of representations at a particular level or
greater than normal effects of interference from
other items in the to-be-recalled list.
EVIDENCE FOR MULTIPLE LEVELS OF
REPRESENTATION IN SHORT-TERM
MEMORY
Phonological Representations
Language-based models of STM suggest the
importance of multiple levels of representation
(i.e., phonological, lexical, semantic, syntactic).
The contribution of phonological representations to verbal short-term memory is well documented, and many models of short-term memory have dealt primarily with accounting for
phonological effects (e.g., Baddeley, 1986; Burgess & Hitch, 1996; Houghton, Hartley, &
Glasspool, 1996). Effects of phonological similarity have been taken as evidence for the phonological nature of the short-term store (Conrad
& Hull, 1964; Hintzman, 1965; Luce, Feustel,
& Pisoni, 1983; Schweickert, Guentert, & Hersberger, 1990; Sperling & Speelman, 1970;
Wickelgren, 1966) and word length effects have
been taken as evidence for an articulatory rehearsal process (Baddeley, Thomson, &
Buchanan, 1975; Mackworth, 1963; Schweickert & Boruff, 1986; Schweickert et al., 1990).
Recently, others have argued that word length
effects do not necessarily implicate rehearsal,
but instead reflect the greater number of phonological segments in longer words (Brown &
Hulme, 1995; Neath and Nairne, 1995; Service,
1998).
An issue separate from whether phonological
representations are maintained in STM is
whether there are separate retention capacities
for input and output phonological forms. As
argued above, some models of language pro-
8
MARTIN, LESCH, AND BARTHA
FIG. 4. Model of short-term memory incorporating separate input and output phonological buffers. As in Fig.
3b, interactions occur only in the long-term knowledge store.
cessing assume such a separation. A typical
procedure in standard short-term memory paradigms involves presenting a list of words auditorily and having the subject reproduce the list
orally. Thus, both word perception and word
production are involved. Thus, it is plausible
that there are separate capacities for maintaining
input and output forms. If a separation between
input and output phonology exists, then our
model in Fig. 3b would need to be revised as
shown in Fig. 4 to include separate capacities
for maintaining input and output forms. On the
word processing side there is a direct connection between input and output forms (that is,
that doesn’t go through lexical and semantic
forms). The ability to repeat nonwords is evidence for the existence of this connection. Either through learning or possibly through some
innate endowment, humans can take an input
acoustic form and come up with the articulatory
form needed to reproduce it. The connection in
the opposite direction reflects the ability to activate the input form from an output form—
without involving audible speech production.
That is, one can imagine the sound of a word or
nonword even when the word or nonword might
be pronounced internally. Some have suggested
that the recirculation between input and output
forms constitutes inner rehearsal (see Howard &
Franklin, 1990; Martin, Blossom-Stach, Yaffee,
& Wetzel, 1995; Monsell, 1987).
With regard to separable STM capacities for
retaining input and output forms, there are neuropsychological data that bear on the issue.
Howard and Franklin (1988, 1990) have argued
for a model in which there are separate input
and output phonological stores and a rehearsal
process which circulates information between
input and output forms (see Monsell, 1987, for
a related proposal). Part of their evidence comes
from a patient, MK, who showed good performance on matching span tasks and good speech
production, but severely impaired single word
and nonword repetition. (In a matching span
task, the patient has to say whether two lists are
the same or different. Since overt repetition is
INDEPENDENCE OF INPUT
not required, matching span is presumed to reflect the retention of input phonological representations.) They argued that MK had preserved
input and output stores but a disruption in the
processes which convert between input and output forms. Other evidence comes from Allport’s
(1984) study which contrasted two brain-damaged patients, both of whom showed very reduced memory span on repetition tasks. The
two patients differed, however, in their performance on a matching span task. One patient
performed as poorly on the matching span task
as on the list repetition tasks, whereas the other
showed excellent performance on the matching
span task, at least for three- and four-word lists.
Allport argued that the two patients differed in
that one had a disruption of an input phonological store, whereas the other had a preservation
of the input phonological store, but a disruption
of the output phonological store. Romani (1992)
has presented more recent evidence of a case
who appeared to have a preserved input phonological store and disrupted output phonological
store as the patient performed very well on
matching span and probe tasks for even nonword lists, but performed poorly on list repetition.
Other patients provide evidence of a dissociation opposite to that presented by Romani
(1992). Patient EA reported by Martin et al.
(1994) and patient JB reported by Shallice and
Butterworth (1977) did very poorly on matching span and memory probe tasks (which also
do not require list output), thus implying a disruption of the input phonological store. However, their spontaneous output was normal as
assessed by a variety of measures including
speech rate, syntactic complexity, pausing, etc.
Assuming that spontaneous speech production
involves planning several phonological forms
simultaneously, their preserved production suggests a preserved output phonological store.
In evaluating these neuropsychological data,
it is necessary to take care to distinguish
whether the evidence strongly supports the assumption of separate input and output phonological codes and input and output stores or
whether a model which assumes separate input
and output pathways between a single phono-
9
logical representation and semantics could accommodate the data (see Shallice, 1988, for
discussion). Thus, for example, if a patient has
good spoken word comprehension but poor production, one would not need to postulate separate input and output phonological forms, as the
difficulty might be in the output pathway for
activating phonology from lexical and semantic
representations. However, some of the cases
discussed above present a challenge to the single phonological representation account. For
example, patient EA (Martin & Breedin, 1992;
Martin et al., 1994), who was argued to have an
input store deficit, showed excellent comprehension of single words, and it would thus be
difficult to argue that disruption of the input
pathway from phonology to semantics could
account for her severe deficit on input STM
tasks. Similarly, the patient reported by Romani
(1992), who was argued to have an output store
deficit, was good at producing single words as
names of pictures and in repeating single words,
and thus disruption of the output pathway from
semantics to phonology seems an unlikely account of his output store difficulties.
One source of evidence that seemed to argue
strongly for the separation between input and
output phonology came from cases of deep dysphasia (see Howard & Franklin, 1988, for a
review). These patients have difficulty repeating
nonwords and make semantic errors in the repetition of single words (e.g., saying “mirror” in
response to “reflection”). They also show effects of imageability on word repetition, with
better success at repeating high imageability
words. The patients’ semantic errors suggest
that phonological perception proceeded accurately— otherwise, why would they access a
semantically related word? If input and output
phonological forms were the same, one would
then expect these patients to be able to produce
the phonological form that they had perceived.
Their production of a semantically related word,
rather than the correct word, suggests that the
input form cannot be used for production and
that they had some type of difficulty in retrieving the output form (because of either degraded
semantic representations or output phonological
representations or impaired links between
10
MARTIN, LESCH, AND BARTHA
them). This pattern of symptoms could not easily be accounted for in terms of separate pathways from semantics to phonology on the input
and output sides, but one phonological representation. Even if output links from semantics
to phonology were damaged, the patient should
still be able to produce the phonological form
perceived on the input side if the input and
output forms were the same.
Even though the evidence from deep dysphasia may seem to provide compelling support for
a separation between input and output phonology, N. Martin and Saffran (1992) have proposed an account of these symptoms using their
interactive activation model, which is closely
related to Dell’s (1986; Dell & O’Seaghdha,
1992) model of speech production. In order to
account for repetition, the model is assumed to
work in the opposite direction as well, that is, in
perception as well as in production. Their approach does not assume a separation between
input and output phonological forms in accounting for deep dysphasia. Instead, they assume
that deep dysphasic patients suffer from a rapid
loss of activation at the phonological, lexical,
and semantic levels. Since the phonological representation derived from perception of the input
decays very rapidly, it is not available at the
moment of production and has to be reconstituted from semantics. Decay at the lexical and
semantic levels results in semantic errors (and
other error types) in repetition. 1
Recently, however, Dell, Schwartz, Martin,
Saffran, and Gagnon (1997) encountered difficulties when they attempted to model both the
word production and word repetition patterns of
aphasic speakers with this model. They found
that the parameters needed to account for patients’ accuracy and error patterns in production
could not reproduce their repetition patterns. To
model repetition more accurately, it was necessary to assume that perception was entirely ac1
Howard and Franklin’s patient MK made semantic errors in repetition and showed very poor nonword repetition;
thus, he would fit the clinical designation of deep dysphasia.
MK’s good performance on tasks tapping the retention of
input phonological forms argues against N. Martin and
Saffran’s (1992) account of deep dysphasia as a general
account of all patients showing this symptom complex.
curate for most patients. In order to capture this
difference between perception and production
in their model, there has to be some separation
between input and output phonological processes. They leave as an open question the issue
of exactly where this differentiation occurs.
Lexical and Semantic Representations
With regard to the involvement of lexical and
semantic codes in short-term memory, findings
from both normal and brain-damaged patients
support this assumption. For example, studies
of normal subjects have demonstrated that
memory span is greater for words than nonwords (Brener, 1940; Crowder, 1978) even
when controlling for pronunciation rate for the
two types of materials (Hulme, Maughan, &
Brown, 1991). Thus, there is a lexical contribution to span that appears to be independent of
the functioning of the articulatory loop (but see
Multhaup, Balota, & Cowan, 1996). Also,
memory span is greater for high than low frequency words (Gregg, Freedman, & Smith,
1989; Roodenrys, Hulme, Alban, Ellis, &
Brown, 1994; Tehan & Humphreys, 1988;
Watkins, 1977), greater for words drawn from
the same semantic category than for words
drawn from disparate categories (Poirier & St.
Aubin, 1995), and greater for high than for low
imageability words (Bourassa & Besner, 1994).
Findings from neuropsychological studies support the conclusion that not only are semantic and
phonological information retained in STM, but the
capacities for retaining the two types of information are separable. These studies have demonstrated that dissociations may be obtained between
patients’ ability to retain phonological and semantic information. These dissociations have been
documented by examining the effects of phonological and lexical/semantic variables on span and
by examining serial position effects and error patterns. For example, Martin et al. (1994) reported
that patient EA failed to show normal effects of
phonological similarity and word length on memory span, but showed a normal advantage in memory span for words over nonwords and performed
better on a category probe task than on a rhyme
probe task. Patient AB showed the reverse pattern
of normal phonological effects, equivalent word
INDEPENDENCE OF INPUT
and nonword span, and better performance on a
rhyme probe than category probe task. Another
patient, ML, has recently been documented to
show a pattern similar to that of AB (Martin &
Lesch, 1996). Importantly, patients AB and ML
perform at high levels on tasks that assess the
integrity of semantic information (Martin &
Lesch, 1996). Thus, it is the short-term retention
of semantic information that is impaired and not
semantic knowledge itself. Based on these data,
Martin and colleagues (Martin & Lesch, 1996;
Martin & Romani, 1994; Martin et al., 1994)
propose that the capacities for the short-term retention of semantic and phonological information
are separable.
Additional neuropsychological evidence for
the involvement of lexical and semantic codes
comes from the work of Patterson and colleagues and of N. Martin and Saffran. Patterson,
Graham, and Hodges (1994) and Knott, Patterson, and Hodges (in press) have demonstrated
that patients with dementia that affects semantic
representations show a larger word span for
known words than for unknown words (that is,
words for which they no longer know the meaning). Saffran and N. Martin (1990; Martin &
Saffran, 1992, 1997) have analyzed the word
processing and short-term memory deficits of
patients with phonological vs. semantic processing deficits and found distinctive short-term
memory patterns with regard to recency and
primacy effects in serial recall and the effects of
imageability and frequency at early vs. late serial positions. That is, patients with phonological processing deficits tend to show greater primacy than recency effects and greater effects of
imageability at late serial positions, whereas
patients with semantic processing deficits tend
to show greater recency than primacy effects
and large effects of imageability at early list
positions. Thus, there is evidence from both
normals and brain-damaged patients supporting
the involvement of multiple linguistic codes in
verbal short-term memory.
GOALS OF THE CURRENT
INVESTIGATION
The present study investigated the relation
between word perception and production and
11
short-term memory and the possible distinction
between input and output phonology by examining the performance of MS, an anomic patient—that is, a brain-damaged individual with
a severe and selective deficit in naming. According to the models shown in Figs. 1 and 2, a
naming deficit could result from a breakdown at
any point in the process going from semantics to
articulation (including, for example, disrupted
semantic representations or a disruption of connections between lexical and phonemic representations). For MS, we will present evidence
that semantic and lexical levels are preserved
and that his difficulty is in activating a phonological representation.
According to the models shown in Figs. 3 and
4, a disruption of the connection between lexical and phonological representations in speech
production should affect short-term memory,
since the same representations and connections
between representations are involved in feeding
activation into the short-term memory buffer.
Moreover, to the extent that the naming deficit
is specific to certain items or types of items,
then short-term memory for those items should
also be impaired.
The proposed distinction between input and
output phonological capacities (see Fig. 4) is
also relevant to the analysis of MS’s short-term
memory performance. The contrast between his
preserved comprehension and his very impaired
naming suggests that input phonology, semantics, and the connections between them are preserved even though the connections between
lexical representations and output phonology
are disrupted. Thus, to the extent that input and
output phonological representations are distinguished in short-term memory, one would predict that MS would perform poorly on tasks
tapping the retention of output phonological
codes, but might perform well on tasks tapping
the retention of input phonological codes.
PATIENT DESCRIPTION
MS is a 30-year-old, right-handed male with
2 years of college education. He contracted herpes encephalitis in 1993 which resulted in an
impairment of language abilities together with a
sparing of most other cognitive abilities. An
12
MARTIN, LESCH, AND BARTHA
MRI scan and an EEG recording suggested left
temporal damage. The testing reported here
took place between 6 months and 3 years postonset. MS displays a pattern of surface dyslexia
and dysgraphia on single word reading and writing tasks. That is, he reads and spells regularly
spelled words (e.g., belt) much better than irregular words (e.g., sword) and tends to make
regularization errors in reading and spelling
(e.g., reading “yacht” as /yækt/ and spelling
“pheasant” as “fezent”). On the word lists from
the Psycholinguistic Assessments of Language
Processing in Aphasia (PALPA) (Kay, Lesser,
& Coltheart, 1992), MS obtained a mean of
93% correct (across two testing sessions) in
reading regular words, but only 52% correct in
reading irregular words. Because of his difficulties with reading and spelling, the current investigation probed his abilities with tasks that used
auditory presentation and oral responses. Portions of these data were presented previously in
Martin and Lesch (1996).
WORD PRODUCTION AND
COMPREHENSION
The tests reported below examined MS’s single word production and comprehension. These
tests were administered in order to explore the
nature of MS’s naming difficulty. Anomia may
have multiple causes. For example, a patient
may have difficulty naming pictures because the
semantic representations corresponding to the
depicted objects have been damaged (Warrington, 1975; Howard & Orchard-Lisle, 1984)
or because the phonological representations for
certain words have been disrupted (see Ellis &
Young, 1988) or because the patient has difficulty activating a phonological representation
from an intact lexical–semantic representation
(Kay & Ellis, 1987). The following tests were
designed to provide a rigorous assessment of
MS’s semantic knowledge in order to rule out a
semantic impairment as the cause of his anomia.
Word Production
Naming. MS obtained a score of 10 on the
60-item Boston Naming Test (Kaplan, Goodglass, & Weintraub, 1983), indicating a severe
impairment in naming ability (control mean 5
TABLE 1
Number Correct on the Philadelphia Naming Test as a
Function of Frequency and Length
Number of syllables
Frequency
1
2
3
4
Low
Medium
High
14/48
19/26
19/26
10/37
3/9
5/5
4/16
2/4
n/a
0/3
n/a
n/a
55.86, SD 5 2.86, Kaplan et al., 1983). MS was
also tested on the Philadelphia Naming Test
(Roach, Schwartz, Martin, Grewal, & Brecher,
1996), which consists of 175 items with names
varying on number of syllables (one to four) and
frequency (frequency ranged from 1 to 20 for
low, from 21 to 66 for medium, and from 70 to
1771 for high frequency items; means 5 7.69,
38.69, and 278.84 for the low, medium, and
high items, respectively, Francis & Kucera,
1982). MS performed poorly, obtaining only
76/174 (44%) correct (the picture “ambulance”
was removed as the name appeared within the
picture). Performance varied as a function of
frequency, with high frequency items (77% correct; 24/31 correct) being named correctly much
more often than low frequency items (27% correct; 28/104 correct), x 2 5 25.71, P , .005;
however, there was no evidence for an effect of
number of syllables (see Table 1). Ninety-five
percent of MS’s errors consisted of circumlocutions (i.e., descriptions of the meaning and/or
use of the item). For example, for cane: “This is
something you use to walk with if you have
trouble walking, if you have, you broke your leg
or something, and you need something to help
you walk.” This description clearly suggests
that MS comprehends the concept “cane.” The
preponderance of semantic descriptions in MS’s
responses suggests that detailed semantic information is available for objects that he cannot
name.
It has been found that some anomic patients
who are initially unable to name a picture are
successful when provided with a phonemic cue
(Funnell & Hodges, 1991; Howard & OrchardLisle, 1984; Myers Pease & Goodglass, 1978).
INDEPENDENCE OF INPUT
In a follow-up to the Philadelphia Naming Test,
MS was asked to name just those pictures that
he had not been able to name on that test (98
pictures). If unsuccessful, he was then given a
phonemic cue (the first phoneme of the target
word). On this retest, MS was unable to name
60 of the 98 pictures. When provided with a
phonemic cue, he was then able to name 31/60
of those pictures. A beneficial effect of a phonemic cue on naming is consistent with the
hypothesis that output phonological representations persist, but MS has difficulty activating
those representations. In terms of the model in
Figs. 3 and 4, the connections between lexical
and phonological representations have been
weakened but have not been totally disrupted.
Access to syntactic features of nonnameable
pictures. In a recent study, Vigliocco, Vinson,
Martin, and Garrett (in press) examined MS’s
ability to make count/mass judgments when in
an anomic state, that is, when unable to retrieve
the phonological form when trying to name a
picture or produce a name to a definition. Count
vs. mass is a syntactic feature of a lexical representation that determines various syntactic
frames into which a word can be fit (e.g., “There
is a _______“ and “There won’t be many ____”
for count nouns vs. “There is _________” and
“There won’t be much ________” for mass).
Although the count vs. mass distinction relates
to conceptual features of the objects to be
named, there are some nouns for which the
distinction is quite arbitrary (consider “artichoke” vs. “broccoli”). The objects chosen for
this study were all selected such that the count/
mass distinction was nonobvious from the conceptual features of the picture or definition.
When MS was unable to provide a name, he
was asked to choose which syntactic contexts
were appropriate for the word he was trying to
retrieve. He was highly accurate in making
these judgments (.80% correct on 91 trials),
even on trials for which he was unable to provide any segmental or syllabic information
about the word. His accuracy on these judgments was within the range of 10 control subjects who were provided the names and asked to
make the same judgments about appropriate
syntactic contexts (mean accuracy 5 85%).
13
MS’s good access to this syntactic feature
would suggest that even when he cannot retrieve phonological information, he can access a
lexical representation (irrespective of which
theory of lexical representations is employed)
and that his naming difficulty does not derive
from weak connections between semantic and
lexical representations.
Speech Perception
Auditory discrimination task. In order to assess MS’s speech perception abilities, an auditory discrimination task (provided by N. Martin
& Saffran, personal communication) was administered. There were 80 pairs of nonwords
and 80 pairs of words— half of which were
“same” and half of which were “different.” On
the “different” trials the word (e.g., funnel–
tunnel) and nonword pairs (beck–feck) differed
in one phoneme and the position of the difference was varied. As a relatively pure test of
auditory discrimination (that is, one that minimizes a short-term memory component), MS
was asked to judge whether a pair of words was
the same or different when there was no delay
between presentation of the two items. In this
condition, MS obtained 99% correct. When a
delay of 5 s was introduced between presentation of the two members of the pair, MS again
obtained 99% correct. Finally, when there was a
filled 5-s delay (when the patient and the examiner counted out loud), MS obtained 98% correct. Thus, MS demonstrates a good ability to
perform auditory discrimination in both no-delay and delay (unfilled or filled) conditions.
MS was also tested on an auditory lexical
decision task which included 180 words (90
abstract and 90 concrete words that were
matched in frequency) (from Kroll & Merves,
1986) and 180 nonwords which differed from
the real words by either one or two phonemes
(provided by N. Martin and Saffran, personal
communication). MS scored 96% correct overall with little difference in performance between
words (172/180) and nonwords (173/180). Five
of eight of the word errors were to abstract
words. In a second test (see N. Martin & Saffran, 1992) of 160 items (80 words; 80 nonwords) the words varied on imageability and
14
MARTIN, LESCH, AND BARTHA
frequency and the nonwords differed from the
words by one or two phonemes. MS again
scored at a high level, obtaining 97% correct
overall, with little difference in performance
between the words (76/80 correct) and nonwords (79/80 correct). All four word errors
were to low frequency–low imageability words.
Peabody picture vocabulary test. MS was
tested on the Peabody Picture Vocabulary Test
(Dunn & Dunn, 1981), which involves hearing
a spoken word and choosing from four picture
alternatives the one that matches the word. MS
performed in the low-normal range (standard
score 5 86 relative to age-matched controls
where m 5 100 and s 5 15). This level of
performance clearly contrasts with his extremely poor performance on a standardized
naming test (i.e., the Boston Naming Test;
Kaplan et al., 1983). However, the PPVT was
not specifically designed to include close semantic distractors among the nonmatching pictures. Thus, it might be possible to obtain a
good score on this task despite having only
partial semantic knowledge of many of the
words. Thus, several additional experimental
tasks were administered that tested semantic
knowledge more closely. These were based on
tests developed by Hillis and Caramazza
(1995), who demonstrated that these tests revealed semantic deficits in a so-called optic
aphasic patient where other more standard tests
had not.
Which is more closely related task. In this
task, the subject sees a picture of one object and
is asked to pick the object that is most closely
related to it from a picture of a pair of objects
(e.g., which is more closely related to a desk, a
table, or a bureau?). In this task, both alternatives are related to the target object, but one is
more closely related. This task was administered to eight control subjects and only items for
which there was a high degree of consistency
(greater than 75%) in responding were included
in the test. MS obtained 19/22 (86%) correct.
This level of performance corresponds to the
mean for the control subjects (range 5 13 to 22
items correct).
Single spoken word-to-single-picture matching. In this task, the subject sees a single pic-
tured object and is asked, “Is this a ____?” For
each word (e.g., “cat”), four trials are presented.
The four trials included a correct picture (cat), a
picture of a semantically related object (dog), a
picture of an object with a phonologically similar name (hat), or an unrelated (semantically or
phonologically) picture (nailclippers). The test
was administered over four sessions, with each
word occurring once per session. This test was
not administered to control subjects as it was
assumed these subjects would obtain 100% correct. MS obtained 205/216 (95%) correct overall, with all 11 errors occurring when words
were paired with semantically related pictures
(e.g., “thread” paired with a picture of yarn).
MS’s 20% error rate (11/54) on the semantically
related pictures might suggest that he has some
semantic disruption for these items. We decided
to further probe MS’s comprehension of these
11 objects by presenting the pictured object and
asking four questions designed to determine
whether MS understood the differences between the pictured object and the word that had
been presented (e.g., for the picture yarn which
had been presented with “thread,” one of the
questions was: Do you knit with this?). One
question was asked at a time and the 11 pictures
were cycled through four times in order to ask
four questions per picture. MS answered all of
the questions correctly, suggesting that his errors on the word-to-picture matching task were
not due to comprehension failures, but more
likely to a low criterion for a match decision—
when questions were focused so as to distinguish between the members of each pair, MS
performed perfectly.
Attribute questions for pictured objects. In
this task, the subject is presented with a picture
of an object from one of the following categories: animals, vehicles, clothing, and edible
plants. There were 33 animals, 26 vehicles, 25
articles of clothing, and 20 edible plants. Five
questions were asked about attributes of the
pictured objects from each category. For animals: Does it have fur? Does it eat other animals? Is it found in water? Is it domestic? Is it
dangerous? For vehicles: Does it require fuel? Is
it primarily used for recreational purposes? Is it
used on land? Is it primarily used to transport
15
INDEPENDENCE OF INPUT
TABLE 2
Performance of MS and Control Subjects on Questions Probing Semantic Knowledge
for Pictured Objects—Number of Items Correct
Category
MS
Controls
Mean (SD)
Range
Animals
Vehicles
Clothing
Plants
134/139
109/111
103/105
83/90
131 (5.09)
120–136
105 (3.08)
98–109
100 (3.86)
93–105
88 (1.45)
85–90
people? Is it quiet? For clothing: Is it used in
cold weather? Is it for men? Is it formal wear?
Is it worn on the extremities? Is it an accessory?
For edible plants: Is it green? Is it primarily
eaten raw or cooked? Is it a fruit or a vegetable?
Do you eat the skin? Can it be used in preparing
a dessert? Presentation of the pictures was
blocked by category and, on each presentation
of a picture, only one question was asked. After
completing one question for a given category,
the category was changed and one question was
asked about all the pictured objects in that category. This procedure continued until all five
questions for each category had been asked.
Testing occurred over several sessions. This
task was also administered to controls such that,
for each category, data were obtained from 16
subjects. Unlike MS, no individual control subject answered questions from all four categories. Therefore, comparison to MS will be made
on a category-by-category basis.
Before examining MS’s performance, the
performance of controls was examined for consistency of responding. Items (questions for particular objects) for which there was less than
75% agreement were eliminated. Additionally,
one question for the clothing category (is it
worn on the extremities?) was removed because
MS seemed unfamiliar with the concept of “extremity.” After the removal of these items, MS
obtained 134/139 (96%) correct on the questions about animals, 109/111 (98%) correct for
vehicles, 103/105 (98%) correct for clothing,
and 83/90 (92%) correct for edible plants. For
all the categories except for edible plants, MS’s
performance was above the mean for the control
subjects (see Table 2). However, MS’s performance for edible plants fell just outside of the
range for the controls. 2
In order to compare comprehension and naming performance for the same set of items, MS
was asked to name the pictures that had been
used in this comprehension test. MS named
45/104 (43%) of these items. There was no
indication that MS’s naming deficit is due to a
semantic impairment: MS failed to name 61%
(51/84) of the items for which he scored 100%
on the comprehension task, whereas he failed to
name only 35% (7/20) of the items for which he
made an error on one of the comprehension
tests.
SUMMARY OF PERFORMANCE ON
COMPREHENSION TASKS
MS performed at, or above, the mean for the
control subjects on several of the comprehension tasks. The exceptions were the questions
probing knowledge about pictured edible plants
and the semantically related trials for the picture–word matching test (in fact, controls were
not tested on the picture–word matching test as
it was assumed that their performance would be
at 100%). On the edible plants questions, MS’s
performance fell just outside the range of the
2
There are several instances reported in the literature of
herpes encephalitis cases who have a selective disruption of
semantic knowledge for living things (see Saffran &
Schwartz, 1994), and it might be hypothesized that MS’s
deficit for plants reflects a mild version of this pattern.
However, these patients tend to show impairments for both
animals and plants and there was no evidence that MS had
an impairment for animals.
16
MARTIN, LESCH, AND BARTHA
controls. With regard to the semantically related
picture trials, further probing indicated that MS
has a good ability to distinguish the semantic
properties of pictures and words for the trials on
which he had made errors. MS’s high level of
performance on tasks designed to require detailed semantic knowledge (Hillis & Caramazza, 1995) argues strongly that his naming
deficit is not the result of an impaired semantic
system. Even though there may be some indication of a semantic deficit limited to the category edible plants, this potential deficit clearly
cannot account for the pattern of results observed with MS. Additionally, when asked to
name the pictures involved in some of the comprehension tasks, there appeared to be little relationship between his naming and comprehension. For those items for which he performed
perfectly on the questions probing semantic
knowledge, he named only 39% (33/84), which
again argues that his naming deficit is not related to a semantic impairment.
These results, along with the finding that a
phonemic cue benefits naming performance, are
consistent with the hypothesis that MS has intact semantic representations and that the connections between lexical–semantic representations and output phonological representations
remain, but have been weakened. For normal
subjects, we assume that the connections between lexical and phonological representations
are stronger for high frequency words than for
low frequency words (for discussions of word
frequency effects in production see Dell, 1990,
and Jescheniak & Levelt, 1994). To explain the
large frequency effect in naming for MS, we
would hypothesize that the connections between lexical and phonological representations
have been damaged for words of all frequencies, but because of the weaker connection
strengths for low frequency items, the disruption to naming is more evident for these items.
representations or damaged phonological word
forms. If this same ability to activate output
phonology from semantics is involved in word
list recall, then MS should be impaired on shortterm memory tasks as well.
SHORT-TERM MEMORY
Words vs. nonwords. One piece of evidence
that semantic information supports recall comes
from the observation that memory span is
greater for word lists than for nonword lists
(Brener, 1940; Crowder, 1978; Hulme et al.,
1991; Multhaup et al., 1996). MS was thus
MS’s performance on the naming tasks and
comprehension tasks suggests that his anomia
reflects an inability to activate phonological
representations from lexical–semantic representations and is not a result of degraded semantic
Digit Span and Digit Matching Span
Our initial evaluation of MS included the
forward digit span task from the WAIS-R
(Wechsler, 1981). On this test MS repeated the
lists perfectly up to the eight-item lists, at which
point he missed 1 of 2 lists. On a subsequent test
of digit span using a pointing response (with 20
lists at each list length), MS scored at a normal
level, obtaining 100, 90, 85, and 80% correct for
four-, five-, six-, and seven-item lists, respectively. He also performed well on a digit matching span task. In this task, the subject hears 2
lists containing the same digits and is asked to
judge whether the order of the items is the same
or different across the 2 lists. On the nonmatching trials, the second list reverses the order of
two adjacent items. Position of the reversal was
balanced across serial positions. There were 20
lists with 10 matching and 10 nonmatching trials. MS’s performance was excellent, as he obtained 95% correct on six-item lists (the mean
level of performance for six control subjects
was 93% correct; range, 85–100% correct).
The digit span results suggested that MS has
normal short-term memory and thus appeared to
contradict the hypothesis of a close relation
between word production processes and list recall. However, digits are high frequency words
(the word frequencies for the digits 0 to 9 range
from 81 to 3372 per million, with a mean of
729.33 and standard deviation of 1074.93; Francis & Kucera, 1982) and, as indicated in the
discussion of the naming results, MS’s naming
ability was much worse with low frequency
words.
Word List Repetition
17
INDEPENDENCE OF INPUT
TABLE 3
Percentage of Lists Correct for Word and Nonword Lists
Matched in Constituent Phonemes—Repeated Sampling
from Same Pool
List length
MS
Words
Nonwords
2
3
4
5
100
100
100
100
60
80
20
20
List length
3
4
5
Control subjects—Mean level of performance (range)
Words
98
90
52
(90–100)
(80–100)
(10–80)
Nonwords
90
63
25
(90)
(30–90)
(0–60)
tested on the recall of word and nonword lists.
The words were of low frequency (,10 per
million, Kucera & Francis, 1967) and the nonwords were formed by switching phonemes
among the words. A standard span procedure
was used in which list recall began with 10
two-item lists and list length was increased by
increments of one until lists were five items in
length. Items were drawn from the same pool
for each presentation. Spoken recall was used.
The results are shown in Table 3. MS did not
show the standard pattern of better recall for
word than for nonword lists—at no list length
did MS show an advantage of word over nonword recall. Recall was actually slightly worse
for the word lists than the nonword lists, suggesting that MS’s recall does not benefit from
the semantic information provided by words.
Six control subjects were tested at list lengths
three through five. As can be seen in Table 3,
MS’s recall of the nonword lists was within the
normal range, while his word recall was below
the range of controls for the four-item lists.
Collapsing across list length, all six control subjects showed an advantage of word recall over
nonword recall (mean advantage, 21%; SD,
11.43). For only one subject, at one list length,
did nonword recall exceed word recall: At list
length four, one subject recalled one more nonword list than word list.
We also had MS recall word and nonword
lists that did not repeatedly sample from the
same pool (in order to make the task more
comparable to the recall tasks reported below).
Word lists consisted of low frequency words
(,10 per million, Kucera & Francis, 1967),
while the nonwords again were formed by rearranging the phonemes contained within the
words. As can be seen in Table 4, we again
obtained evidence that MS’s recall does not
benefit from the semantic information provided
by words (indeed, for four-item lists substantially more nonword lists were recalled than
word lists). A comparison with seven control
subjects indicates that MS performed at a high
level in recalling nonwords. However, in recalling word lists, MS again scored below the range
of control subjects on the four-item lists. Collapsing across list length, all seven control subjects showed an advantage of word recall over
nonword recall (mean advantage, 30%; SD,
7.14). For only one subject, at one list length,
did nonword recall exceed word recall: At list
length five, one subject recalled one nonword
list and no word lists.
Words varied on frequency and imageability.
MS was also tested on list repetition tasks provided by N. Martin and Saffran (see Saffran &
Martin, 1990; Martin, Saffran, & Dell, 1996) in
order to examine effects of imageability and
TABLE 4
Percentage of Lists Correct for Word and Nonword Lists
Matched in Constituent Phonemes—New Items on each
List
List length
MS
Words
Nonwords
3
4
5
100
100
30
70
20
10
Control subjects—Mean level of performance (Range)
Words
99
84
17
(90–100)
(70–100)
(0–40)
Nonwords
80
27
1
(60–90)
(10–40)
0–10)
18
MARTIN, LESCH, AND BARTHA
TABLE 5
Percentage of lists correct as a function of frequency
(HF, LF) and imageability (HI, LI)
MS
HF
HI
LI
Mean
LF
HI
LI
Mean
Overall mean
Controls
3-word
4-word
4-word
5-word
95
100
98
80
73
77
97
96
97
71
65
68
70
65
67
83
60
7
34
55
95
89
92
94
53
41
47
58
frequency on list recall. As indicated previously, N. Martin and Saffran (1992, 1997; Martin et al., 1996; Saffran & Martin, 1990; see also
R. Martin & Lesch, 1996) have obtained evidence suggesting an association between type
of STM impairment (phonological, semantic, or
mixed) and patterns of imageability and frequency effects on list recall. Half of the lists
were composed of high imageability words and
half of low imageability words. High imageability words had ratings greater than 4.97 (on a
scale of 1–7), while low imageability words had
ratings less than 4.97 (Paivio, Yuille, & Madigan, 1968). Within each level of imageability,
half of the lists were composed of high frequency words and half of low frequency words.
High frequency words had frequencies of
greater than 40 per million, while low frequency
words had frequencies of less than 30 per million (Kucera & Francis, 1967). MS was presented with 80 three-word lists and 60 fourword lists. Spoken recall was used. Overall, MS
recalled 83% of three-word lists and 55% of
four-word lists. The percentage of lists correct
by frequency and imageability is shown in Table 5. MS showed substantially better recall for
the high than low frequency lists for both list
lengths (for list length three, 39/40 lists vs.
27/40: x 2 5 12.47, P , .005; for list length
four, 23/30 vs. 10/30: x 2 5 11.38, P , .005).
An effect of imageability was evident within the
four-word low frequency lists where he showed
much better recall of the high than low imageability lists (9/15 vs. 1/15: x 2 5 9.60, P , .005).
In terms of items correct at each serial position,
the pattern of recall was similar for both list
lengths (see Fig. 5)—a reduced recency effect
and an effect of frequency particularly at later
list positions. This pattern is consistent with
data that N. Martin and Saffran (1997) have
obtained from patients with presumed phonological deficits.
Table 6 presents a breakdown of the errors
(collapsed over list length) that MS made for the
different list types. These errors do not include
order errors (i.e., reporting an item in the incorrect serial position). As can be seen from Table
6, a high percentage of MS’s errors shared a
phonological relationship with the target word,
indicating that MS retained phonological information about the words in the lists. Further, it is
FIG. 5. Frequency and imageability effects as a function
of serial position (Martin & Lesch, 1996).
19
INDEPENDENCE OF INPUT
TABLE 6
Percentage of Error Types in List Recall as a Function of Frequency (HF, LF) and Imageability (HI, LI)
Error types
Patient MS
HF
HI
LI
LF
HI
LI
Controls
HF
HI
LI
LF
HI
LI
Sa
SD
P-W
P-NW
0
0
20
0
0
33
40
33
0
10
32
5
11
22
0
0
0
0
,1
0
0
0
(SD 1 P)
U
O
Other
N
0
0
0
17
40
17
0
0
5
6
11
37
37
5
0
0
11
17
0
5
19
41
29
18
0
0
3
0
4
5
55
70
11
7
21
17
,1
1
,1
,1
0
2
70
77
8
3
a
Note. S, semantically related word; SD, semantic description; P-W, phonologically related word; P-NW, phonologically
related nonword; SD 1 P, semantic description combined with a phonologically related response; S 1 P, semantically and
phonologically related response; U, unrelated; O, omission; Other, some other type of response. N, total number of errors
for that list type.
interesting to note that phonologically related
nonword responses appeared somewhat more
likely than phonologically related word responses. We will return to this point later.
MS’s responses also quite often included semantic descriptions of the words within the
lists. This type of response was particularly
likely in the recall of low frequency, high imageability lists. For example, for the list lobster,
castle, and bagpipe, he responded: “losser—the
thing you eat, the place the kings go in,” and “it
comes from the place where men wear the same
things as women” (mimics use). As shown in
Table 6, 32% of MS’s errors on the low frequency, high imageability lists were semantic
descriptions of the items. An additional 37% of
his errors consisted of a semantic description
combined with a phonologically related response. On the low frequency, low imageability
lists, 5% of MS’s error responses consisted of
semantic descriptions, while an additional 10%
were semantically related words. The occurrence of semantic descriptions in MS’s recall
suggests that MS at times retained semantic
information about a word that he could not
produce correctly. The greater likelihood of
these errors on high imageability items most
likely occurred because of the richer semantic
information in these items.
Control subjects were tested on the recall of
both four- and five-item lists varying on frequency and imageability in order to determine
whether MS’s level and pattern of performance
would differ from those for controls. The fouritem lists were those that MS had been asked to
recall, while the five-item lists were constructed
by rearranging words from the four-item lists.
Five subjects participated in the recall of the
four-item lists, while 36 subjects participated in
the recall of the five-item lists.
The recall performance of the control subjects demonstrated effects of both frequency
and imageability: In the recall of five-item lists
subjects recalled significantly more high frequency (68%) than low frequency lists (47%),
t(33) 5 8.96, P , .001, and significantly more
high imageability (62%) than low imageability
lists (53%), t(33) 5 3.83, P , .001. Finally, the
Frequency X Imageability interaction was marginally significant, F(33) 5 1.94, p 5 .06. The
20
MARTIN, LESCH, AND BARTHA
results thus replicate previous studies demonstrating frequency and imageability effects in
list recall (e.g., Hulme et al., 1991; Bourassa &
Besner, 1994) and further provide some evidence that these factors interact in supporting
retention (this pattern of results is similar to that
obtained by Lesch & Martin, 1998).
As is evident in Table 5, MS performed substantially worse than the controls. Control subjects recalled 94% of the four-item lists (range,
5 87–100%) compared to 55% of lists correct
for MS. The mean performance of the controls
on the five-item lists (mean, 58%; range, 7–95%
correct) was similar to MS’s level of recall for
the four-item lists (55% correct). Although both
MS and the control subjects showed effects of
frequency and imageability, there was some indication of greater effects of these variables for
MS than the controls. These comparisons are
somewhat difficult to make, given the different
level of performance for MS and the controls.
However, some comparisons can be made by
evaluating the effects of frequency and imageability when MS and the controls are matched
for performance on certain cells (necessarily at
different list lengths). For example, MS’s level
of performance on the high frequency threeword lists (98%) was similar to that of controls
on the four-word high frequency lists (97%).
However, MS showed a much greater frequency
effect on the three-word lists (31%) than did the
controls on the four-word lists (3%). Also, the
level of performance for MS on the high frequency four-word lists (77%) was fairly similar
to that of the controls on the high frequency
five-word lists (68% correct), but again MS
showed a much larger frequency effect on the
four-word lists (43%) than did the controls on
the five-word lists (21%). It should be noted,
though, that on the four-word lists, the larger
frequency effect for MS was due primarily to
his very poor performance on the low frequency, low imageability lists. For the low frequency four-item lists, MS showed a much
larger effect of imageability (53%) than did the
controls for the low frequency five-word lists
(12%) even though MS and the controls were
approximately matched in their level of recall
for the low frequency, high imageability items.
In order to explore the types of errors made
by control subjects, mean error rates were calculated as a function of frequency, imageability,
and error type. Subjects who accurately recalled
all the lists in any frequency–imageability condition were excluded. The mean error rates are
presented in Table 6. The error distributions
appear quite different for MS and for the control
subjects. Whereas MS provided a large percentage of semantic descriptions, control subjects
never provided semantic descriptions of the
items to be recalled and only very rarely produced semantically–related responses. Additionally, control subjects very rarely made phonologically related nonword responses, while
this type of response was quite common in
MS’s recall. The predominant error types for
the controls were omissions and the substitution
of phonologically related words. Thus, not only
does MS’s level and pattern of performance
differ from that of control subjects, the types of
errors also differ (even when overall level of
recall is approximately matched).
List recall—Names of pictures named/not
named and words with and without comprehension errors. MS performed at a high level on a
variety of tasks that require the access of detailed semantic information, suggesting that his
poor performance on list recall is not the result
of insufficient semantic support due to disrupted
semantic representations. In order to provide
further evidence on this point, MS was tested on
recall of lists of words corresponding to the
names of pictures that he could or could not
name and words for which he made or did not
make comprehension errors. Given MS’s generally high level of performance on the semantic
tasks, it is difficult to claim that his semantic
knowledge for items with errors was actually
disrupted relative to controls, since his level of
performance was similar to that of controls on
most tests. That is, even though MS made an
error on, for example, one attribute of an object
in the attribute questions task, his performance
could not be said to differ from controls since
they also made such errors at a similar rate—
though perhaps not on the same items. However, if one takes a conservative approach and
hypothesizes that MS does in fact have some
21
INDEPENDENCE OF INPUT
degree of semantic deficit for the items on
which he made errors, then recall should be
worse for lists of items for which MS demonstrated some comprehension difficulty (see
Warrington, 1975; McCarthy & Warrington,
1987; Patterson et al., 1994). 3 On the other
hand, if difficulty in phonological retrieval underlies his naming deficit, and if list recall involves the same phonological retrieval processes involved in naming, then list recall
should be worse for items he could not name
than for items he could name.
Items that MS could/could not name were
taken from the Philadelphia Naming Test. Items
that MS could/could not comprehend were
taken primarily from the comprehension tasks
reported above. Items for which there was the
strongest evidence of comprehension difficulty
(items that were responded to incorrectly in
more than one task) were selected first. MS was
presented with 10 four-item lists of the following types: (a) items that MS could name, (b)
items that MS could not name, (c) items for
which MS made errors on the comprehension
tests, and (d) items for which MS made no
errors on these tests. The named/not named lists
were approximately matched in terms of frequency and number of syllables. The comprehended/not comprehended lists were also
matched in terms of number of syllables, but as
there were relatively few items for which MS
demonstrated comprehension difficulty, it was
difficult to match lists in terms of frequency. All
lists were presented auditorily and spoken recall
was used. This task was administered twice,
3
Warrington (1975; McCarthy & Warrington, 1987) first
employed the known/unknown technique to study patients
with characteristics similar to the patients of Patterson et al.
(1994) and failed to observe a significant advantage for
known compared to unknown items. Patterson et al. point to
the different methods used to select known/unknown items
as one possible source of the discrepant findings. They
further suggest that the near ceiling levels of performance
by Warrington’s (1975) patient on the unknown items suggests that, at the time of initial testing, the patient still
maintained semantic knowledge for these items. At the time
of a second testing, when a comparison of unknown words
to nonwords was made, performance had dropped considerably for these items. However, no comparison was made
for the known items.
with the second testing occurring approximately
6 months after the first testing.
As the pattern of performance was similar
across the two administrations, we will report
means collapsed over administration. MS was
better able to recall lists composed of items that
he could name (95% lists correct; 99% items
correct) than lists of items that he could not
name (55% lists correct; 81% items correct), for
lists correct (19/20 vs. 11/20), x 2 5 8.53, P ,
.005, for items correct (79/80 vs. 65/80), x 2 5
13.61, P , .005. In contrast, there was no
difference in the recall of lists composed of
items for which he showed perfect comprehension (13/20 lists correct; 72/80 items correct)
versus those for which he made any comprehension errors (13/20 lists correct; 71/80 items
correct). MS’s pattern of performance on this
recall task strongly argues that a semantic deficit is not the basis of his poor performance in
list recall—MS’s level of recall appears to be
unrelated to his accuracy in responding to the
comprehension questions for particular items.
This contrasts with the observation that MS’s
recall was poorer for lists composed of items
corresponding to pictures that he was unable to
name in a naming task than for items that he had
been able to name, again indicating a strong
correspondence between his ability to retrieve
phonological representations in word production and list recall.
DISCUSSION
To summarize, MS showed normal list recall
for digits and nonwords, but impaired recall for
words. Word recall was particularly impaired
for low frequency words and within low frequency words, for low imageability words.
MS’s normal recall of nonword lists indicates
that he does not have difficulty in maintaining
and reproducing phonological information per
se. However, MS does not show the boost to
recall from the lexical–semantic information
available in words that is shown by normal
subjects. Consistent with the suggestion of a
reduced lexical influence on recall is the prevalence of phonologically related nonword responses in MS’s recall. The phonological relatedness of the responses suggests an ability to
22
MARTIN, LESCH, AND BARTHA
retain phonological information. For normal
subjects, feedback from the lexical to the phonological level should prevent the production of
nonword responses, and few were observed in
our control data. The few that were observed
occurred for low frequency lists, suggesting that
some of the subjects may not have been familiar
with some of the items. (Note that MS made
nonword responses to both low and high frequency items—see Table 6.) MS’s recall of
word lists suggests that recall is characterized
by the same difficulty in retrieving phonological
representations from semantics that is characteristic of his naming. A large effect of frequency was evident in both his naming and list
recall presumably due to the weaker connections between semantics and phonology for low
frequency words. MS’s very good performance
on digit span lists compared to word lists can be
attributed to the high word frequency of digits.
Further evidence of the concordance between
MS’s word production and list recall was evident in the errors that MS made in list recall.
That is, MS at times produced circumlocutions
in list recall that were similar to his circumlocutions in naming. In these cases, MS recalled
the meanings of the items contained within the
list rather than the items themselves indicating
that the words had been accurately perceived
(indicating intact input phonological representations) and that appropriate semantic representations had been activated. These semantic descriptions were most likely to occur in the
condition where retrieval of phonological forms
would be assumed to be most difficult (low
frequency words), but where semantic information should be most readily available (high imageability words), indicating that, while semantic information had been activated, it could not
be used to retrieve the appropriate phonological
form. A comparison of MS’s error types with
those of controls indicated that control subjects
never provided semantic descriptions of the
items to be recalled. Other evidence of a correspondence between word production and list
recall came from MS’s recall of lists corresponding to the names of pictures which he
could or could not name. Better performance
was obtained for words he produced as names
than for those he could not produce as names.
This result contrasted with equivalent performance on lists composed of words for which he
did or did not make errors on the comprehension tests.
MS’s naming and list recall responses diverged in that he produced many more phonological approximations (both word and nonword forms) in list recall than he had in naming.
This difference can be attributed to the fact that
phonological information is presented to MS in
the list recall tasks, and, as indicated by his
recall of nonwords, he has a good ability to hold
onto phonological information. Thus, he can
use this surviving phonological information to
constrain his responses. In naming, phonological information is not provided, but has to be
retrieved.
The results of the word vs. nonword recall
tasks and the task manipulating frequency and
imageability of words might appear contradictory. Specifically, the failure to find an advantage for words over nonwords would seem to
contradict the finding of an imageability effect
for MS. 4 Our discussion of the word vs. nonword results focused on the low frequency of
the words and MS’s poor ability to retrieve low
frequency words. However, since words would
be bound to be more imageable than nonwords
(and the words we used in these tasks were
fairly imageable), an advantage for words might
be predicted. One possible explanation for the
failure to find an advantage for words (and, in
fact, the finding of a slight numerical disadvantage for words) is that with words, a semantic
representation is activated which leads to the
activation of incorrect but semantically related
words on the output side. If these related words
are of higher frequency than the target words,
the phonological forms of the related words
may become highly activated and interfere with
the retrieval of the phonology of the target
words. With nonwords, there would be no competition from words activated from semantics.
According to this logic one would see both
imageability and frequency effects for words,
but the performance for nonwords would not
4
We thank an anonymous reviewer for this suggestion.
23
INDEPENDENCE OF INPUT
necessarily be worse than that for any type of
word.
Up to this point, we have provided evidence
that MS demonstrates difficulties in recall tasks
that require verbal output (i.e., output phonological representations). Within a model that
distinguishes between input and output phonology, the data suggest that MS has difficulty on
the output side in retrieving phonological forms.
If one further assumes that there are separate
input and output phonological buffers, MS
might be predicted to perform well on tasks that
require only the retention of input phonological
forms.
STM for Input Phonology
For the most part, these tasks did not require
output of words from the memory lists, but
rather a yes/no judgment about whether a probe
word or list matched the preceding list. Previous
evidence indicated that MS has a good ability to
retain semantic information from word lists
even when he cannot retrieve appropriate phonological forms corresponding to these semantic representations. Thus, it was necessary to
ensure that these tasks tapped the retention of
phonological information rather than semantic
information. The first task, a rhyme probe task,
required phonological information in order to
make the rhyme/nonrhyme judgment. The next
task required the retention of order information,
which has been argued to rely on the retention
of phonological information (e.g., Wickelgren,
1965).
Rhyme probe task. In the rhyme probe task
the subject judges whether a probed word
rhymes with any item in a preceding list. Lists
varied from 1 to 7 items in length and number of
lists at each list length varied from 20 to 28. For
each list length, half of the trials were rhymepresent and half rhyme-absent. On rhymepresent trials, each serial position was probed
equally often. MS’s performance was similar to
that of control subjects who were roughly
matched with MS in terms of age and education
level. MS’s estimated span (75% correct performance) was 6.4 items, while the mean for the
control subjects was 7.08 (SD, 1.44) items.
As MS’s performance on the repetition tasks
TABLE 7
Percentage of Correct as a Function of List Length—
Rhyme Probe Task (Low Frequency–Low Imageability
Words)
List length
MS
Controls
Mean
Range
4
5
6
94
95
100
89
(81–94)
86
(80–95)
70
(58–83)
was more impaired for low frequency than for
high frequency words, we constructed a new
rhyme probe task that used only low frequency
(,30 words per million; Kucera & Francis,
1967) and low imageability words (,497 on a
scale of 100 to 700; Oxford Psycholinguistic
Database, Quinlan, 1992). While a frequency
value of 30 may not seem a particularly low
cutoff for low frequency words, this is the same
cutoff that was used in the lists for recall that
varied on frequency and imageability and for
which MS showed a large effect of frequency.
List length varied from four to six items and
number of lists at each list length varied from 16
to 24. As can be seen in Table 7, MS performed
at a high level relative to six control subjects. In
order to assess whether this level of performance would be stable across test sessions, MS
was tested a second time on this task. He obtained 100, 85, and 83% correct for list lengths
of four, five, and six, respectively—a level of
performance which again compared favorably
to that of control subjects (see Table 7).
Order recognition task. In the order recognition task the subject hears an eight-item list of
words followed by a pair of adjacent words
from the list (with all serial positions being
probed equally often). The pair of words either
are in the same order as they occurred within the
list or they are in the reverse order. The subject
is asked to decide whether the word order
within the pair is the same as it had been within
the preceding list. Items were one, two, or three
or more syllables in length and presentation was
blocked by item length. There were 48 lists at
24
MARTIN, LESCH, AND BARTHA
TABLE 8
can be seen in Table 9, MS performed at a high
level on this difficult nonword probe task.
Percentage Correct as a Function of Item
Length—Order Recognition Task
DISCUSSION OF PERFORMANCE ON
STM TASKS TAPPING INPUT
PHONOLOGY
Number of syllables
MS
Controls
1
2
3
79
77
81
76
73
76
each list length and an equal number of same
and different trials. As can be seen in Table 8,
MS performed at least as well as control subjects on this task.
We also constructed an order recognition task
using only low frequency (,30 words per million, Kucera & Francis, 1967) and low imageability words (,497 on a scale from 100 to 700;
Oxford Psycholinguistic Database, Quinlan,
1992). In this version of the task, subjects heard
28 lists of eight two-syllable words. MS performed at a high level, obtaining 89% (25/28)
correct. Again, MS’s performance compared favorably to that of six control subjects (mean,
70%; range, 54 –79% correct). Indeed, he performed at a higher level than the control subjects. In a second administration, MS’s level of
performance was equivalent to his performance
on the initial testing (89% correct), indicating
that he reliably performs well on this order
recognition task.
Nonword probe task. In the nonword probe
task the subject hears a list of nonwords followed by a nonword probe. The subject must
decide whether the probe item occurred in the
preceding list. List length varied from three to
five items and each serial position was probed
equally often. Half of the probes occurred in the
preceding list, while half did not. Of those
probes that did not occur in the preceding list,
half were phonologically similar (they shared
an onset with one item and a body with another
item— e.g., the probe “lirb” given the list items
“jirb” and “lez”) and half were phonologically
dissimilar. Number of lists at each list length
varied from 20 to 24. The results for MS and for
six control subjects are presented in Table 9. As
MS demonstrates an impairment in STM
tasks that require word production—MS recalled only 83% of the three word lists and 55%
of the four word lists that were varied on frequency and imageability. Interestingly, MS at
times reported the meanings of the words in the
list rather than the words themselves, indicating
that the words had been accurately perceived
and the appropriate semantic representations activated. We suggested that MS’s deficit reflects
a difficulty in retrieving output phonological
representations from semantics. Consistent with
this interpretation, MS’s pattern of performance
is quite different on STM tasks that do not
require verbal output. On all the tasks reported
here, MS’s performance was comparable to that
of control subjects. Our results are thus consistent with an interpretation in terms of a separation between input and output phonology and a
separation between input and output phonological buffers.
However, there is some indication that we
may actually be underestimating the extent of
MS’s impairment in the retention of output phonological codes since the control subjects did
not tend to perform as well as MS on the input
tasks. The differing patterns of performance exhibited by MS on input and output tasks is
underscored by a comparison of MS with an
additional control subject whom we tested on
TABLE 9
Percentage Correct as a Function of List
Length—Nonword Probe Task
List length
MS
Controls
Mean
Range
3
4
5
6
96
96
75
88
95
(79–100)
90
(79–100)
93
(80–100)
86
(75–96)
INDEPENDENCE OF INPUT
both the order recognition task (with low frequency–low imageability words) as well as the
recall of word lists varying on frequency and
imageability: On list recall, MS only recalled
55% of four-word lists compared to 100% correct recall for the control subject. However, this
control subject only obtained 75% correct on
the order recognition task, while MS obtained
89% correct (no individual subject performed as
well as MS on this task). Even though none of
the controls performed as well as MS did on the
order recognition task, there appeared to be
little difficulty in performing a task requiring
output phonological codes—a task for which
MS only obtained 55% lists correct.
We have suggested that the dissociations between performance on input and output phonological STM tasks can be interpreted as support
for the notion that there are separate input and
output phonological representations and separate buffers for retaining these representations.
As discussed in the introduction, it is necessary
to consider whether a model such as those
shown in Fig. 3 with a single phonological
representation could accommodate the findings.
In such a model, MS’s disruption would be in
the connections from semantics on the output
side. However, this approach runs into difficulty
in accounting for MS’s STM data. That is, if
there is one phonological buffer associated with
storing phonological forms derived from either
input or output, it would be difficult to account
for MS’s normal performance on input STM
tasks and poor performance on output STM
tasks. That is, if one assumed that a disruption
in the connections from semantics to phonology
was the source of MS’s poor performance on
output STM tasks, poor performance on input
STM tasks would be expected as well because
of the absence of feedback from the semantic to
the phonological level. This reasoning assumes
that input tasks are affected by lexical–semantic
factors. One source of evidence in this regard
comes from a recent event-related potential
(ERP) study (Ruchkin, Berndt, Johnson, Grafman, Ritter, & Canoune, 1998) that examined
performance for words versus nonwords on a
recognition probe task, which would be a task
tapping input retention. These researchers found
25
differences in the topographical pattern of ERPs
for words versus nonwords that were evident
both during encoding and during a 6-s retention
interval.
SUMMARY AND CONCLUSIONS
We have reported data from an anomic patient, MS, whose naming performance can be
characterized as demonstrating greater difficulty in producing lower frequency words and
retention of detailed semantic information about
objects that cannot be named. Rigorous testing
provided evidence of preserved semantic and
lexical information about names that could not
be produced and we concluded that MS’s naming deficit derived from weakened connections
between lexical and phonological representations.
According to models such as those shown in
Figs. 3 and 4 that postulate a close connection
between word processing and short-term memory, the weakened connections between lexical
and phonological representations should have
predictable consequences for STM performance. That is, MS should not show the benefit
to recall that derives from the flow of activation
from semantic and lexical representations to
phonological representations on the output side,
and the factors that influenced his ability to
retrieve phonology in single word tasks should
influence his STM performance. We found this
to be the case: (a) MS’s performance did not
show the advantage normally seen in word recall relative to nonword recall; (b) word frequency exerted a strong influence on his STM
performance, with higher frequency items being
more likely to be recalled than lower frequency
items; (c) specific items that were unlikely to be
named were less likely to be recalled in a shortterm memory task than were items that had been
correctly named; and (d) MS produced circumlocutions in list recall as he did in naming tasks.
The data from MS also provided evidence
relevant to distinguishing between a model such
as those in Fig. 3, with a single phonological
representation underlying both input and output,
and that in Fig. 4, with separate input and output
phonological representations and input and output phonological buffers. In support of the sep-
26
MARTIN, LESCH, AND BARTHA
arate buffers interpretation we observed that,
while MS’s performance on short-term memory
tasks suffered when verbal output was required,
his performance was normal on short-term
memory tasks that did not require verbal output.
A model such as that in Fig. 3 would have
difficulty accounting for these findings, unless it
were assumed that somehow performance on
input STM tasks were uninfluenced by feedback
from the lexical and semantic levels.
The current investigation thus provides further support for the notion that verbal STM
involves the maintenance of the various levels
of representation involved in word perception
and production. These results are consistent
with the claim that the short-term retention of
verbal material depends on the activation of
long-term memory representations (phonological, lexical, and semantic) for words and the
maintenance of this activation in short-term
memory. While our approach suggests that
LTM makes its contribution during encoding
and retention, an alternative view is that LTM
representations exert an influence at the point of
recall. 5 For example, Hulme et al. (1991) hypothesize that at recall lexical representations
are automatically retrieved and are used to reinstate information in decayed memory traces—
since nonwords lack these representations, pattern completion should be more successful for
words than for nonwords. A similar account of
5
It should be noted that these relatively recent attempts to
attribute various aspects of span performance to a LTM
contribution differ from earlier accounts of span performance as deriving from recall from both short- and longterm stores (e.g., Waugh & Norman, 1965; Craik, 1971). In
these recent accounts, what is meant by LTM is the use of
“semantic” memory to aid recall—that is, long-term knowledge of lexical and semantic representations of words. In the
earlier accounts what was meant by LTM was “episodic”
memory. According to these earlier views, information
about a specific list (i.e., what words were presented in what
order at a particular point in time) was represented in both
STM and LTM. At the time of recall, items could be
recalled from either STM or LTM. Neuropsychological
evidence argues against this earlier view in that amnesic
patients with very impaired ability to learn new information
(i.e., impaired episodic memory) but preserved knowledge
of words (i.e., preserved semantic memory) show normal
memory span performance (Baddeley & Warrington, 1970;
Levin, Benton, & Grossman, 1982).
the lexicality effect has been offered by Schweickert and colleagues (Schweickert, 1993;
Schweickert, Ansel, & McDaniel, 1991), who
term the reconstruction process “redintegration.” Roodenrys et al. (1994) have suggested
that redintegration can also account for the frequency effect—phonological representations in
long-term memory are more accessible for high
frequency words than for low frequency words.
Thus, according to a redintegration account,
effects of lexicality and frequency are due to the
relative retrievability of lexical phonological
representations at the time of recall. However,
an effect of imageability (e.g., Bourassa &
Besner, 1994) is not easily attributed to redintegration as there is no obvious reason to suspect that it would be easier to reconstruct high
than low imageability words from a degraded
phonological trace. That is, lexical phonological
information would seem to be as readily available for low as for high imageability words.
Thus, while a redintegration approach encounters difficulty in accounting for semantic effects
on STM, the language-based approach accounts
for both phonological and semantic effects on
STM by positing a close relation between the
codes stored in LTM (phonological, lexical, and
semantic) and those active in STM—the phonological, lexical, and semantic representations
shown in the left side of Fig. 3 are the long-term
memory representations for the phonological
units in the language and for the semantic specifications of known lexical items. The current
findings form part of a growing body of data
that indicate that a disruption of these representations or access to these language representations will have predictable effects on STM.
Moreover, the findings point to a need for separate representations for input and output phonology in both language processing and shortterm memory.
REFERENCES
Allport, D. A. (1984). Auditory verbal short-term memory
and conduction aphasia. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control
and language processes (pp. 351–364). Hillsdale, NJ:
Erlbaum.
Baddeley, A. D. (1986). Working memory. New York: Oxford Univ. Press.
INDEPENDENCE OF INPUT
Baddeley, A. D., Thomson, N., & Buchanan, M. (1975).
Word length and the structure of short-term memory.
Journal of Verbal Learning and Verbal Behavior, 14,
575–589.
Baddeley, A. D., & Warrington, E. K. (1970). Amnesia and
the distinction between long- and short-term memory.
Journal of Verbal Learning and Verbal Behavior, 9,
176 –189.
Baker, S. C., Frith, C. D., Frackowiak, R. S. J., & Dolan,
R. J. (1996). Active representation of shape and spatial
location in man. Cerebral Cortex, 6, 612– 619.
Barnard, P. (1985). Interacting cognitive subsystems: A
psycholinguistic approach to short-term memory. In A.
Ellis (Ed.) Progress in the psychology of language,
Vol. 2, pp. 197–258. London: Erlbaum.
Bourassa, D., & Besner, D. (1994). Beyond the articulatory
loop: A semantic contribution to serial order recall of
subspan lists. Psychonomic Bulletin and Review, 1,
122–125.
Brener, R. (1940). An experimental investigation of memory span. Journal of Experimental Psychology, 26,
467– 482.
Brown, G., & Hulme, C. (1995). Modeling item length
effects in memory span: No rehearsal needed? Journal
of Memory and Language, 34, 594 – 621.
Burgess, N., & Hitch, G. J. (1996) A connectionist model of
STM for serial order. In S. E. Gathercole (Ed.), Models
of short-term memory. Hove, England: Psychology
Press.
Conrad, R., & Hull, A. J. (1964). Information, acoustic
confusion and memory span. British Journal of Psychology, 55, 429 – 432.
Craik, F. I. M. (1971). Primary memory. British Medical
Bulletin, 27, 232–236.
Crowder, R. G. (1978). Memory for phonologically uniform
lists. Journal of Verbal Learning and Verbal Behavior,
17, 73– 89.
Dell, G. S. (1986). A spreading activation theory of retrieval
in sentence production. Psychological Review, 93,
283–321.
Dell, G. S. (1990). Effects of frequency and vocabulary type
on phonological speech errors. Language and Cognitive Processes, 5, 313–349.
Dell, G. S., & O’Seaghdha, P. G. (1992). Stages of lexical
access in language production. Cognition, 42, 287–
314.
Dell, G. S., Schwartz, M. F., Martin, N., Saffran, E. M., &
Gagnon, D. A. (1997). Lexical access in aphasic and
nonaphasic speakers. Psychological Review, 104, 801–
838.
Dunn, L., & Dunn, L. (1981). Peabody Picture Vocabulary
Test - Revised. Circle Pines, MN: Guidance Service.
Ellis, A. W., & Young, A. W. (1988). Human cognitive
neuropsychology. Hove, England: Erlbaum.
Francis, W. N., & Kucera, H. (1982). Frequency analysis of
English usage: Lexicon and grammar. Boston: Houghton Mifflin.
Funnell, E., & Hodges, J. R. (1991). Progressive loss of
27
access to spoken word forms in a case of Alzheimer’s
disease. Proceedings of the Royal Society of London,
Series B-Biological Sciences, 243, 173–179.
Gregg, V. H., Freedman, C. M., & Smith, D. K. (1989).
Word frequency, articulatory suppression and memory
span. British Journal of Psychology, 80, 363–374.
Hillis, A. E., & Caramazza, A. (1995). Cognitive and neural
mechanisms underlying visual and semantic processing: Implications from “optic aphasia.” Journal of
Cognitive Neuroscience, 7, 457– 478.
Hintzman, D. L. (1965). Classification and aural coding in
short-term memory. Psychonomic Science, 3, 161–162.
Houghton, G., Hartley, T., & Glasspool, D. W. (1996). The
representation of words and nonwords in short-term
memory: Serial order and syllable structure. In S. E.
Gathercole (Ed.), Models of short-term memory. Hove,
England: Psychology Press.
Howard, D., & Franklin, S. E. (1988). Missing the meaning?
Cambridge, MA: MIT Press.
Howard, D., & Franklin, S. E. (1990). Memory without
rehearsal. In G. Vallar and T. Shallice (Eds.), Neuropsychological impairments of short-term memory (pp.
287–318). Cambridge, UK: Cambridge Univ. Press.
Howard, D., & Orchard-Lisle, V. M. (1984). On the origin
of semantic errors in naming: Evidence from the case
of a global dysphasic. Cognitive Neuropsychology, 1,
163–190.
Hulme, C., Maughan, S., & Brown, G. D. A. (1991). Memory for familiar and unfamiliar words: Evidence for a
long-term memory contribution to short-term memory
span. Journal of Memory and Language, 13, 685–701.
Jescheniak, J., & Levelt, W. (1994). Word frequency effects
in speech production: Retrieval of syntactic information and of phonological form. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20,
824 – 843.
Jusczyk, P. (1993). How word recognition may evolve from
infant speech perception capacities. In G. Altmann &
R. Shillcock (Eds.), Cognitive models of speech processing: The second Sperlonga meeting (pp. 27-55).
Hillsdale, NJ: Erlbaum.
Kaplan, E., Goodglass, H., & Weintraub, S. (1983). Boston
Naming Test. Philadelphia: Lee & Febiger.
Kay, J., & Ellis, A. (1987). A cognitive neuropsychological
case study of anomia: Implications for psychological
models of word retrieval. Brain, 110, 613– 629.
Kay, J., Lesser, R., & Coltheart, M. (1992). Psycholinguistic Assessments of Language Processing in Aphasia:
Reading and Spelling. Hove, England: Erlbaum.
Knott, R., Patterson, K., & Hodges, J. R. Lexical and
semantic binding effects in short-term memory: Evidence from semantic dementia. Cognitive Neuropsychology, 14, 1165–1216.
Kroll, J. F., & Merves, J. S. (1986). Lexical access for
concrete and abstract words. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 12,
92–107.
Kucera, H., & Francis, W. N. (1967). Computational anal-
28
MARTIN, LESCH, AND BARTHA
ysis of present-day American English. Providence, RI:
Brown Univ. Press.
Lesch, M. F., & Martin, R. C. (1998). Lexical, semantic,
and phonological effects on span: Support for a psycholinguistic approach to short-term memory. [manuscript in preparation]
Levelt, W. J. M. (1989). Speaking: From intention to articulation. Cambridge, MA: MIT Press.
Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (In press). A
theory of lexical access in speech production. Behavioral and Brain Sciences.
Levin, H. J., Benton, A. L., & Grossman, R. G. (1982).
Neurobehavioral consequences of closed head injury.
New York: Oxford Univ. Press.
Luce, P. A., Feustel, T. C., & Pisoni, D. B. (1983). Capacity
demands in short-term memory for synthetic and natural speech. Human Factors, 25, 17–32.
Mackworth, J. F. (1963). The duration of the visual image.
Canadian Journal of Psychology, 17, 62– 81.
Martin, N., & Saffran, E. (1992). A computational account
of deep dysphasia: Evidence from a single case study.
Brain and Language, 43, 240 –274.
Martin, N., & Saffran, E. (1997). Language and auditoryverbal short-term memory impairments: Evidence for
common underlying processes. Cognitive Neuropsychology, 14, 641– 682.
Martin, N., Saffran, E., & Dell, G. (1996). Recovery in deep
dysphasia: Evidence for a relation between auditoryverbal STM capacity and lexical errors in repetition.
Brain and Language, 52, 83–113.
Martin, R. C., Blossom-Stach, C., Yaffee, L., & Wetzel, F.
(1995). Consequences of a central motor programming
deficit for rehearsal and reading comprehension. Quarterly Journal of Experimental Psychology, 48A, 536 –
572.
Martin, R. C., & Breedin, S. (1992). Dissociations between
speech perception and phonological short-term memory. Cognitive Neuropsychology, 9, 509 –534.
Martin, R. C., & Lesch, M. (1996). Associations and dissociations between language impairment and list recall: Implications for models of STM. In S. E. Gathercole (Ed.), Models of short-term memory (pp. 149 –
178). East Sussex, UK: Psychology Press.
Martin, R. C., & Romani, C. (1994). Verbal working memory and sentence comprehension: A multiple-components view. Neuropsychology, 8, 506 –523.
Martin, R. C., Shelton, J., & Yaffee, L. S. (1994). Language
processing and working memory: Neuropsychological
evidence for separate phonological and semantic capacities. Journal of Memory and Language, 33, 83–
111.
McCarthy, R. A., & Warrington, E. K. (1987). The double
dissociation of short-term memory for lists and sentences. Brain, 110, 1545–1563.
McClelland, J. L. (1986). The programmable blackboard
model of reading. In J. L. McClelland & D. E. Rumelhart (Eds.), Parallel distributed processing: Explorations in the microstructure of cognition, Vol. 2: Psy-
chological and biological models (pp. 122–169).
Cambridge, MA: MIT Press.
McClelland, J. L., & Elman, J. L. (1986). The TRACE
model of speech perception. Cognitive Psychology, 18,
1– 86.
Miozzo, M., & Caramazza, A. (1997). Retrieval of lexicalsyntactic features in tip-of-the tongue states. Journal of
Experimental Psychology: Learning, Memory, and
Cognition, 23, 1410 –1423.
Monsell, S. (1987). On the relation between lexical input
and output pathways for speech. In D. A. Allport, D.
MacKay, W. Prinz, & E. Scheerer (Eds.), Language
perception and production: Relationships between listening, speaking, reading and writing (pp. 273–311).
London: Academic Press.
Multhaup, K., Balota, D., & Cowan, N. (1996). Implications
of aging, lexicality, and item length for the mechanisms underlying memory span. Psychonomic Bulletin
and Review, 3, 112–120.
Myers Pease, D., & Goodglass, H. (1978). The effect of
cuing on picture naming in aphasia. Cortex, 14, 178 –
189.
Neath, I., & Nairne, J. S. (1995). Word length effects in
immediate Memory: Overwriting trace decay theory.
Psychonomic Bulletin and Review, 2, 429 – 441.
Paivio, F., Yuille, J. C., & Madigan, S. A. (1968). Concreteness, imagery and meaningfulness values for 925
words. Journal of Experimental Psychology Monograph Supplement, 76(3, Part 2).
Patterson, K., Graham, N., & Hodges, J. R. (1994). The
impact of semantic memory loss on phonological representations. Journal of Cognitive Neuroscience, 6,
57– 69.
Poirier, M., & St. Aubin, J. (1995). Memory for related and
unrelated words: Further evidence on the influence of
semantic factors in immediate serial recall. Quarterly
Journal of Experimental Psychology, 48A, 384 – 404.
Quinlan, P. T. (1992). The Oxford psycholinguistic database. Oxford: Oxford Univ. Press.
Roach, A., Schwartz, M., Martin, N., Grewal, R., & Brecher, A. (1996). The Philadelphia Naming Test: Scoring and rationale. Clinical aphasiology (Vol. 24, pp.
121–134). Austin: Pro-Ed.
Roelofs, A., Meyer, A., & Levelt, W. (1996). Interaction
between semantic and orthographic factors in conceptually driven naming: Comment on Starreveld and La
Heij (1995). Journal of Experimental Psychology:
Learning, Memory and Cognition, 22, 246 –251.
Romani, C. (1992). Are there distinct output buffers? Evidence from a patient with an impaired output buffer.
Journal of Language and Cognitive Processes, 7, 131–
162.
Roodenrys, S., Hulme, C., Alban, J., Ellis, A., and Brown,
G. D. (1994). Effects of word frequency and age of
acquisition on short-term memory span. Memory and
Cognition, 22, 695–701.
Ruchkin, D. S., Berndt, R. S., Johnson, R., Grafman, J.,
Ritter, W., & Canoune, H. (1998). Lexical-semantic
INDEPENDENCE OF INPUT
contributions to retention of verbal information in
working memory: Event-related potential evidence.
[unpublished manuscript]
Saffran, E. M., & Martin, N. (1990). Neuropsychological
evidence for lexical involvement in short-term memory. In G. Vallar & T. Shallice (Eds.), Neuropsychological impairments of short-term memory (pp. 428 –
447). London: Cambridge Univ. Press.
Saffran, E. M., & Schwartz, M. F. (1994). Of cabbages and
things: Semantic memory from a neuropsychological
perspective—A tutorial review. In C. Umilta & M.
Moscovitch (Eds.), Attention and performance XV (pp.
507–536). Cambridge, MA: MIT Press.
Schweickert, R. (1993). A multinomial processing tree
model for degradation and redintegration in immediate
recall. Memory and Cognition, 21, 168 –176.
Schweickert, R., Ansel, B., & McDaniel, M. A. (1991).
Lexicality and generation effects on memory span.
Paper presented at the 32nd Annual Meeting of the
Psychonomic Society. San Francisco (Nov.).
Schweickert, R., & Boruff, B. (1986). Short-term memory
capacity: Magic number or magic spell? Journal of
Experimental Psychology: Learning, Memory and
Cognition, 12, 419 – 425.
Schweickert, R., Guentert, L., & Hersberger, L. (1990).
Phonological similarity, pronunciation rate, and memory span. Psychological Science, 1, 74 –77.
Service, E. (1998). The effect of word length on immediate
serial recall depends on phonological complexity, not
articulatory duration. The Quarterly Journal of Experimental Psychology, 51A, 283–304.
Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge Univ. Press.
Shallice, T., & Butterworth, B. (1977). Short-term memory
impairment and spontaneous speech. Neuropsychologia, 15, 729 –735.
Shallice, T., McLeod, P., & Lewis, K. (1985). Isolating
cognitive modules with the dual-task paradigm: Are
speech perception and production separate processes?
Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 37(A) 507–532.
29
Smith, E. E., Jonides, J., Koeppe, R. A., Awh, E., Schumacher, E., & Minoshima, S. (1995). Spatial vs. object
working memory: PET investigations. Journal of Cognitive Neuroscience, 7, 337–358.
Sperling, G., & Speelman, R. G. (1970). Acoustic similarity
and auditory short-term memory. In D. A. Norman
(Ed.), Models of human memory (pp. 151–202). New
York: Academic Press.
Tehan, G., & Humphreys, M. S. (1988). Articulatory loop
explanations of memory span and pronunciation rate
correspondences: A cautionary note. Bulletin of the
Psychonomic Society, 26, 293–296.
Vigliocco, G., Vinson, D., Martin, R. C., & Garrett, M. (In
press). Is “count” and “mass” information available
when the noun is not? An investigation of tip of the
tongue states and anomia. Journal of Memory and
Language.
Warrington, E. K. (1975). The selective impairment of
semantic memory. Quarterly Journal of Experimental
Psychology, 27, 635– 657.
Watkins, M. J. (1977). The intricacy of memory span.
Memory and Cognition, 5, 529 –534.
Waugh, N. C., & Norman, D. A. (1965). Primary memory.
Psychological Review, 72, 89 –104.
Wechsler, D. (1981). Wechsler Adult Intelligence ScaleRevised (WAIS-R). New York: Harcourt Brace Jovanovich.
Wickelgren. W. A. (1965). Short-term memory for phonemically similar lists. American Journal of Psychology,
78, 567–574.
Wickelgren, W. A. (1966). Distinctive features and errors in
short-term memory for English consonants. Journal of
the Acoustical Society of America, 39, 388 –398.
Wilson, F. A. W., Scalaidhe, S. P. O., & Goldman-Rakic,
P. S. (1993). Dissociation of object and spatial processing domains in primate prefrontal cortex. Science,
260, 1955–1958.
(Received August 18, 1997)
(Revision received November 16, 1998)