Brain and Cognition 54 (2004) 251–253
www.elsevier.com/locate/b&c

Testing predictions of the interactive activation model
in recovery from aphasia after treatment
Regina Jokel,* Elizabeth Rochon, and Carol Leonard
Graduate Department of Speech-Language Pathology, University of Toronto, 500 University Avenue, Room 1034, Toronto, Ont., Canada M5G 1V7
Accepted 12 February 2004

Abstract
This paper presents preliminary results of pre- and post-treatment error analysis from an aphasic patient with anomia. The
Interactive Activation (IA) model of word production (Dell, Schwartz, Martin, Saffran, & Gagnon, 1997) is utilized to make
predictions about the anticipated changes on a picture naming task and to explain emerging patterns. Error patterns are viewed in
light of two putative mechanisms within the IA network, connection strength, and rate of decay. The results suggest that while these
mechanisms can successfully account for the magnitude and order of specific errors (e.g., predominance of semantic paraphasias
and nonwords followed by formal paraphasias under the conditions of weak connections), the treatment may have influenced the
occurrence/absence of some error types.
Ó 2004 Elsevier Inc. All rights reserved.

1. Introduction
Anomia, a difficulty in retrieving words, is one of the
most common symptoms in patients with brain damage
due to a stroke. Numerous theoretical explanations have
been proposed to account for different patterns of anomic errors. One of the most viable accounts is the Interactive Activation model (Dell, Schwartz, Martin,
Saffran & Gagnon, 1997), which postulates that the
lexical retrieval proceeds in two steps, semantic, and
phonological. Activation in the retrieval network cascades in a top-down fashion, while connections are
bidirectional and excitatory, allowing for both feedforward and feedback between levels of processing. The
degree of activation in the network is a sum of both
feedback and feedforward activation that changes over
time, and it is relative to the semantic and phonological
closeness to the target. The role of feedback within the
IA network is to stabilize the pattern of activation over
the target nodes and assure that the selected entry is a
real word (Ôlexical biasÕ).
The model has been used to assist in making predictions about normal word access and to provide insights
*
Corresponding author.
E-mail address: rjokel@baycrest.org (R. Jokel).

0278-2626/$ - see front matter Ó 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.bandc.2004.02.033

into the hypothesized mechanisms linked to errors in
word selection, which are seen in aphasic speakers (Dell
et al., 1997). Two such mechanisms have been suggested;
the strength of connections between levels of processing
and the decay of the signal traveling between the levels.
Schwartz, Saffran, Bloch, and Dell (1994) utilized the
IA model in their study to demonstrate the consequences of the weak connection mechanism in their
fluent aphasic patient, FL, who showed high rates of
nonwords in spontaneous speech. Under the principles
of the IA account, weakening of connection strength
reduces the amount of activation that spreads between
levels of processing, leading in turn to a decreased
amount of feedback. When the system is deprived of
feedback, the lexical bias will be replaced by nonwords.
An overproduction of nonwords in aphasia was therefore linked to weak connections.
Accelerated decay, on the other hand, was hypothesized to cause semantic errors in single word repetition,
an inability to repeat non-words, and the occurrence of
formal (phonemic) paraphasias in naming (Martin, Dell,
Saffran, & Schwartz, 1994). Using the IA model,
(Martin et al., 1994) were able to characterize the deficits
of NC, a patient with deep dysphasia. Deep dysphasia is
a disorder defined by the inability to repeat non-words,
formal paraphasias in naming and spontaneous speech,

252

R. Jokel et al. / Brain and Cognition 54 (2004) 251–253

and semantic errors in repetition. When the signal
within the network decays rapidly, there is not enough
activation to complete the feedback loop, therefore selection of a lexical target will be more influenced by the
most recently activated level. As both spontaneous
speech and picture naming tasks evoke the phonological
nodes last, a phonological influence will exert the
greatest power on final production. Subsequently, the
selected item will most likely be phonologically rather
than semantically related to the target. Similarly, when a
lexical item is to be repeated, the semantic nodes will be
activated last. As the initial activation forwarded from
the phonological to the semantic nodes may provide an
insufficient boost to select a single lemma, an activated
semantic relative may be selected instead. Under conditions of rapid decay, subject to decay severity, the
system will not receive confirmatory feedback as to the
accuracy of the semantic competitor. This mechanism is
thought to cause an inability to repeat nonwords.
In addition, Dell and his colleagues (Dell et al., 1997;
Gagnon, Schwartz, Martin, Dell, & Saffran, 1997)
showed that simple quantitative changes applied to a
normal processing model could account for a variety of
naming errors. By modifying a single parameter (i.e.,
either connection strength or decay rate), they were able
to simulate a wide range of errors made by aphasic
patients.
This paper presents preliminary error analysis from
an aphasic patient who was administered a picture
naming task before and after treatment for anomia. To
date, the IA model was utilized to examine error patterns at different time points in spontaneous recovery.
We wished to examine the potential influence of treatment effects on error pattern. Data presented here is
extracted from one of nine patients participating in our
project.

2. Method
Data presented comes from MB, a 71-year-old male
12 years after stroke, who was randomly assigned to
receive a phonologically based treatment for anomia.
(Patients in this project received one of three treatment
options; phonological, semantic features, and mixed
approach.) A set of 175 line drawings of common objects (Philadelphia Naming Test, Roach, Schwartz,
Martin, Grenwal, & Brecher, 1996) was presented for
naming. MB was asked to provide single word responses
(whenever possible). All responses were transcribed,
coded, and scored by two independent scorers using the
International Phonetic Association rules. The overall
inter-judge reliability score for MB was 92.1%. Errors
receiving separate coding categories consisted of: (a)
formal paraphasias (e.g., mat for cat), (b) semantic paraphasias, (e.g., dog for cat), (c) mixed errors (e.g., rat

for cat), (d) nonwords (e.g., strek for cat), and (e) unrelated words (e.g., bench for cat).

3. Results
Fig. 1 illustrates the pattern of MBÕs errors on a preand post-treatment picture naming task. Although the
results indicate post-treatment improvements (27 errors
pre- versus 20 post-treatment), factors other than the
overall severity were analyzed in this project.
According to the IA model, it is the order and magnitude of error types that lead to the identification of
each of the two postulated mechanisms underlying a
naming impairment. Despite the fact that MBÕs pattern
shows no unrelated errors, the order and magnitude of
error types appear to be most consistent with reduced
connection strength both pre- and post-treatment (see
Fig. 1). An intriguing finding post-treatment was the
emergence of mixed errors. The IA model applied to
recovery without treatment does not predict the occurrence of error types not seen on initial evaluation.
Therefore, the emergence of a new error type after
treatment may be suggestive of therapeutic influences.
The provision of phonologically based treatment in a
patient with the predominance of semantic errors at the
outset of the experiment, may have lead to the appearance of errors that carry both semantic and phonological resemblance to the target (hence mixed errors in
MBÕs post-treatment profile).

4. Conclusions
Findings from this investigation suggest that while
the IA model can be successfully utilized in making
predictions regarding recovery, the treatment that patients receive may alter the anticipated pattern. Data
from a larger sample are currently being analyzed to
provide more compelling support for this conclusion.
Our findings will also provide directions for explorations
in the area of word production that are currently under

Fig. 1. Naming profile of patient MB pre- and post-treatment.

R. Jokel et al. / Brain and Cognition 54 (2004) 251–253

way, i.e., comparisons of error patterns seen in recovery
from naming impairments due to a stroke with those of
naming decline seen in primary progressive aphasia, a
neurodegenerative disorder.

References
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.

253

Gagnon, D. A., Schwartz, M. F., Martin, N., Dell, G. S., & Saffran, E.
M. (1997). The origins of formal paraphasias in aphasicsÕ picture
naming. Brain and Language, 59, 450–472.
Martin, N., Dell, G. S., Saffran, E., & Schwartz, M. (1994). Origins of
paraphasia in deep dysphasia: Testing the consequences of a decay
impairment to an interactive spreading activation model of lexical
retrieval. Brain and Language, 47, 609–660.
Roach, A., Schwartz, M. F., Martin, N., Grenwal, R. S., & Brecher, A.
(1996). The Philadelphia naming test. Clinical Aphasiology, 24,
121–133.
Schwartz, M. F., Saffran, E. M., Bloch, D. E., & Dell, G. S. (1994).
Disordered speech production in aphasic and normal speakers.
Brain and Language, 47, 52–88.