Brain and Cognition 60 (2006) 272–281
www.elsevier.com/locate/b&c
Cognitive neuroimaging: Cognitive science out of the armchair
Greig I. de Zubicaray ¤
Centre for Magnetic Resonance, The University of Queensland, Qld, Australia
Accepted 3 June 2005
Available online 9 January 2006
Abstract
Cognitive scientists were not quick to embrace the functional neuroimaging technologies that emerged during the late 20th century. In
this new century, cognitive scientists continue to question, not unreasonably, the relevance of functional neuroimaging investigations that
fail to address questions of interest to cognitive science. However, some ultra-cognitive scientists assert that these experiments can never
be of relevance to the study of cognition. Their reasoning reXects an adherence to a functionalist philosophy that arbitrarily and purposefully distinguishes mental information-processing systems from brain or brain-like operations. This article addresses whether data from
properly conducted functional neuroimaging studies can inform and subsequently constrain the assumptions of theoretical cognitive
models. The article commences with a focus upon the functionalist philosophy espoused by the ultra-cognitive scientists, contrasting it
with the materialist philosophy that motivates both cognitive neuroimaging investigations and connectionist modelling of cognitive systems. Connectionism and cognitive neuroimaging share many features, including an emphasis on uniWed cognitive and neural models of
systems that combine localist and distributed representations. The utility of designing cognitive neuroimaging studies to test (primarily)
connectionist models of cognitive phenomena is illustrated using data from functional magnetic resonance imaging (fMRI) investigations
of language production and episodic memory.
© 2005 Elsevier Inc. All rights reserved.
Keywords: Functional magnetic resonance imaging; Connectionism; Cognitive science; Language production; Episodic memory; Neuroimaging
1. Introduction
Consider the following statements made recently by
Coltheart (2004a) in relation to functional neuroimaging:
“ƒfacts about the brain do not constrain the possible
natures of mental information-processing systems. No
amount of knowledge about the hardware of a computer
will tell you anything about the nature of the software that
the computer runs. In the same way, no facts about the
activity of the brain could be used to conWrm or refute
some information-processing model of cognition” (p. 22).
This position, espoused by ultra-cognitive psychologists (sotermed by Shallice, 1988), is an expression of the functionalist philosophy articulated prominently by Fodor (1981)
and Pylyshyn (1984). Functionalism assumes information
*
Fax: +617 3365 3833.
E-mail address: [email protected].
0278-2626/$ - see front matter © 2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.bandc.2005.11.008
processing occurs at a level of abstraction that does not
depend on the physical composition of the system. Abstract
information processing depends only upon the organization
of the mental system—the relationships among parts that
are deWnable according to the function they perform—and
these are presumed to operate according to discernible psychological rules or principles.
The adoption of the software analogy for the mind
occurred as cognitive science developed an interest in
computational modelling (e.g., Block, 1980; Pylyshyn,
1984). Block (1980) referred to the resulting conXation of
functional and computational descriptions as computation-representation functionalism. In an inXuential publication on visual perception, Marr (1982) proposed three
diVerent levels for machine implementations of information processing; computational theory, representation
and algorithm (i.e., software), and hardware implementation. He also emphasised the independence of the former
two levels from the latter, while acknowledging their
G.I. de Zubicaray / Brain and Cognition 60 (2006) 272–281
interconnections. Later, Block (1995) would write a book
chapter with the unambiguous title “The mind as the software of the brain,” reinforcing the primary research focus
of modern cognitive science.
It is worth noting that the functionalist perspective
reXects only one attempt at a solution to the perennial
mind-body problem in philosophy. Despite the occasional
antipathy expressed among philosophers, such solutions
are not dogma, and they should not be revered as such.
Many scientists, including cognitive scientists, adhere to
materialist philosophies that link cognitive and brain processes. Some, such as Dennett (1991a, 1991b), provide
strong countering arguments to functionalism. Like Dennett (1991a), I think there is something ludicrous about a
position that precludes scientists from attempting empirical explorations of models because others claim to have
an a priori proof that all such attempts are hopeless. By
precluding functional neuroimaging investigations from
providing evidence for or against cognitive models, ultracognitive science is adopting the position of armchair philosopher: ready to observe, ready to berate, never ready to
engage.
Functional neuroimaging is not the only experimental
method to be deemed irrelevant by the ultra-cognitivists.
Over two decades ago the connectionist movement
showed promise to move cognitive modelling closer to
neural modelling. This is because the constituents of its
models are nodes in networks connected in ways that
resemble brain networks (Dennett, 1991b; Medler, 1998).
Seidenberg (1993) noted connectionism’s potential to
extend modelling from being merely descriptive to the
level of providing explanatory theories. However, as
recently as last year, Harley (2004a) had cause to lament
that connectionism appeared largely ignored by cognitive
scientists.
Like functional neuroimaging, connectionism eschews
an arbitrary distinction between software and hardware,
thereby conXating Marr’s (1982) proposed levels of
machine implementation. Within connectionist models,
hardware and software are inexorably intertwined (nodes
are neuron-like), networks develop pragmatically, representations are discovered, learning occurs. No assumptions are
made regarding governing psychological rules or principles
in the functionalist manner, rather connectionist models
demonstrate how the cognitive systems become organised
the way they are (see Bechtel & Abrahamsen, 2002; Medler,
1998).
That connectionist modelling and neuroimaging technologies emerged at roughly the same time in history is an
interesting coincidence. That both methods of empirical
exploration with their similar emphases could be largely
ignored or labelled irrelevant by a group of ultra-cognitive
scientists is not coincidental: They represent alternative
approaches to a functionalist-inspired cognitive science.
More speciWcally, they represent attempts to apply uniWed
models of cognitive and neural processes. As with any type
of model, it is intended that they be tested.
273
1.1. Localist representations and the straw man
Cognitive psychologists continue to debate the value of
including localist representations in their models. Despite
the debate, it seems that non-connectionist modellers tend
to adopt localist representations more frequently than their
connectionist counterparts (primarily because they tend to
eschew distributed processing in their models). I use the
term “non-connectionist” here in the same manner as Coltheart (2004a), referring to largely descriptive computational models that do not a priori describe their
connections as being brain- or neuron-like (apparently to
preserve the functionalist’s theoretical distinction between
mental and physical systems—see the section on models of
language production below).
For example, Coltheart (2004b) recently oVered evidence
in support of models that comprise systems of localist representations of word (phonological and orthographic) and
object (visual/structural) forms, i.e., lexicons. He argued
that models comprising solely distributed representations
fail to account for the range of behavioural observations
from lesion patients. At its outset, connectionism made the
central claim that knowledge is coded in a distributed manner (Rumelhart, McClelland, & the PDP Research Group,
1986). However, localist representations do emerge in connectionist models, and more recent approaches to connectionist modelling have explicitly incorporated localist
representations (see Bowers, 2002; Page, 2000). In fact, preempting Coltheart (2004b), Bowers (2002) showed the deWciencies of purely distributed models of language, and the
advantages of incorporating localist representations within
a connectionist architecture. A considerable number of
connectionist models now employ a combination of localist
and distributed representations (see Page, 2000).
How is this debate about localist and distributed representations relevant to the debate about functional neuroimaging? The activation patterns of nodes in connectionist
models can be recorded to determine whether localist representations have emerged during processing, analogous to
neuroimaging voxel timeseries data (e.g., Medler, Dawson,
& Kingstone, 2005). These nodes can be thought of as consisting of a single neuron or a distinct population of neurons (Page, 2000). While the complexity of these models
cannot approach the complexity of brain networks, the
approach is nevertheless similar. Modern functional neuroimaging methods, such as functional magnetic resonance
imaging (fMRI; with which I am most familiar), provide
neural signal timeseries information during the performance of tasks designed to engage a cognitive process of
interest. In 1997, Marcus Raichle suggested that the construction of cognitive paradigms is the “real Achilles’ heel”
in functional neuroimaging experiments. This is still the
case. However, contemporary experiments frequently use
identical experimental designs to those enjoying the
endorsement of cognitive science, with some obvious constraints introduced by the scanner environment. The neuroimaging timeseries data are treated simply as another
274
G.I. de Zubicaray / Brain and Cognition 60 (2006) 272–281
dependent variable, comparable to conventional behavioural recordings or the recordings of activation levels in
network models (Henson, 2004). In this way, clusters of
voxels comprising activated brain nodes are identiWed and
are inferred to represent, in a localist manner, cognitively
meaningful constructs or entities occurring during information processing.
When faced with voxel-based recordings of neural activity and attributed systems of localist representations, some
of neuroimaging’s detractors have chosen to attack a straw
man in the guise of the long-discredited practice of phrenology. The fact that phrenology can even achieve a passing
mention in recent discussions by prominent cognitive scientists (e.g., Bub, 2000; Coltheart, 2004a; Harley, 2004a)
should be seen as cause for concern. A “straw man argument” belongs to a class of logical fallacy. It refers to a misrepresentation or distortion (such as an over-simpliWcation)
of a position that is substituted and then attacked in its
stead. The defeat of the substitution is then presented as a
proof against the original position. It is a fallacy because
attacking a misrepresentation is hardly identical to attacking the original position. Phrenology did not involve challenging cognitive systems by manipulating processes of
interest and observing concomitant neural activity—the
key features of modern functional neuroimaging (Bechtel,
in press; Henson, 2004). Bub’s (2000) example of a modern
phrenologist who misguidedly uses functional neuroimaging experiments to localise broad cognitive processes is no
more than an anachronism, and still a straw man.
1.2. The crux of the debate: Can neuroimaging experiments
inform cognitive models?
Cognitive neuroscience studies have rarely been conducted with the intent of adjudicating between rival cognitive models. However, this does not mean that they have
never been, nor that they cannot. Coltheart (2002) is right
to distinguish between the aims of cognitive neuroscience
and cognitive neuroimaging. The former attempts to identify
the regional brain activity associated with de facto cognitive processes in the brain. I say “de facto,” because many
of these studies do seem to implicitly accept that the processes being investigated exist in whatever form the
researchers’ preferred cognitive theories have speciWed.
Conversely, cognitive neuroimaging studies aim to test
assumptions about the information processes putatively
involved in producing an eVect of interest. In this way, they
hope to inform and constrain cognitive theories. Some
examples of this method applied to primarily connectionist
models of spoken word production and episodic memory
are discussed below.
1.3. Neuroimaging tests of models of spoken word production
Harley (2004a) raised the prospect of neuroimaging
being used to adjudicate between rival serial and interactive-activation (IA) models of spoken word production,
and concluded that such an experiment, if ever conducted,
would probably be of little consequence. In the same journal issue, Coltheart (2004a) described Levelt, Roelofs, and
Meyer’s (1999) serial model of spoken word production as
an example of a descriptive functional information-processing model that is “not intended to say anything whatsoever
about the neural level” (p. 23). Coltheart therefore classiWed
the model as being non-connectionist (see above).1 However, Coltheart’s view appears to be a selective one designed
to maintain the functionalist distinction between software
and hardware. Dell and Sullivan (2004) consider the Levelt
et al. (1999) model to be a connectionist model “in the sense
that computation is carried out by spreading activation
through a network of units representing lexical knowledge”
(p. 68). This is consistent with Bechtel and Abrahamsen’s
(2002) view of the properties of connectionist models (see
also Medler, 1998). Indefrey and Levelt (2000) conducted a
meta-analysis that related the key stages of word production as implemented in Levelt et al.’s (1999) serial model to
neuroimaging data (Indefrey & Levelt, 2000; see also
Levelt, 2001).2 Clearly, these researchers are of the view that
the model does have something to say about neural processes.
What of Harley’s prophetic viewpoint? Indefrey and
Levelt’s (2000) meta-analysis identiWed a left-lateralised
cerebral network responsible for the core process of word
production, in which the mid section of the left middle temporal gyrus appears to be involved in lexical-conceptual
processing, and the posterior superior and middle temporal
gyri (Wernicke’s area) are speciWcally involved in phonological code retrieval. We had previously used this information in fMRI investigations to test predictions from the
Levelt et al. (1999) serial model, contrasting them with predictions derived from rival IA models (de Zubicaray,
McMahon, Eastburn, & Wilson, 2002; de Zubicaray, Wilson, McMahon, & Muthiah, 2001). The reasoning for conducting these experiments, and their outcomes, is discussed
in the following paragraphs.
Most contemporary theories of spoken word production
propose access occurs to at least two distinct levels of lexical representation, consisting of the conceptual and syntactic information of a to-be-produced word and the encoding
of its phonological information. The term lemma is often
used to describe the former representations, whereas the
latter are typically referred to as (phonological) word forms
(Levelt, 1999; Nickels, 2002). Evidence for these two levels
of lexical access comes from a variety of sources: Speech
errors of normal speakers (“tip of the tongue” states) and
1
In contrast to Coltheart (2004a), Henson (2004) views the assumptions
of these “non-connectionist” models to be amenable to testing with neuroimaging experiments.
2
While the meta-analysis was conducted to identify a cerebral architecture consistent with the Levelt et al. (1999) serial model, the results are
generalisable to other models of spoken word production. This is because
the diVerences in the models’ assumptions relate only to the nature of the
processes occurring between the levels of lexical representation, not to the
levels themselves.
G.I. de Zubicaray / Brain and Cognition 60 (2006) 272–281
patients with brain lesions (aphasics/anomics), and data
from chronometric (reaction time) studies of the time
course of normal word production (Levelt, 1999; Nickels,
2002). Although there is general agreement regarding the
existence of these two levels, there is comparatively little
agreement about their nature, especially with respect to the
processes that occur between them (Dell & Sullivan, 2004;
Nickels, 2002). For example, strict serial models assume
feedforward processing with a single target lemma selected
from among competitors and then phonologically encoded
(e.g., Levelt et al., 1999), whereas cascade (e.g., Morsella &
Miozzo, 2002; Peterson & Savoy, 1998) and interactiveactivation models (e.g., Dell & Sullivan, 2004; Harley, 1993;
Starreveld & La Heij, 1996) assume that all lemma nodes,
target and competitor, are encoded phonologically (a process called phonological coactivation). Interactive models
further assume that phonological encoding in turn inXuences lemma selection, i.e., that there is feedback between
the two levels.
One of the methods applied to investigate spoken word
production is picture-word interference. This is a generalisation of the Stroop (1935) interference task, requiring participants to name a target picture while ignoring an
accompanying spoken or written distractor word
(MacLeod, 1991). The presence of a distractor word slows
the naming response (e.g., Lupker, 1982). Two well-documented priming eVects on the picture-word interference
task are that naming latencies are slower when the distractor word is semantically related to the depicted object (e.g.,
picture DOG, word CAT) and faster when it is orthographically or phonologically related (e.g., picture DOG, word
DOT) compared to when it is an unrelated word (e.g., picture DOG, word RANK). These eVects are referred to as
semantic interference and phonological facilitation, respectively. The major serial and cascade-IA models diVer in
terms of where they assume the locus of semantic interference to occur. The former (e.g., Levelt et al., 1999) assume it
to be at the conceptual processing level (the lemma node),
while the latter typically assume it to be at word form
encoding (e.g., Dell & Sullivan, 2004; Harley, 1993; Starreveld & La Heij, 1996). According to Levelt et al.’s (1999)
serial model, the main eVect of phonological relatedness
occurs at encoding of the whole word form. Cascade-IA
models likewise make this assumption (e.g., Starreveld & La
Heij, 1996).
Much of the debate concerning the nature of the processing that occurs between the two levels of lexical representation has been informed by time course data from
behavioural studies that manipulated the onsets of both
pictures and distractors. However, both serial and cascadeIA theorists view the available behavioural data as supporting their respective positions, somewhat diminishing its
capacity to adjudicate between model types (see e.g., Rapp
& Goldrick, 2004; Roelofs, 2004). We viewed this impasse
as promoting a need for alternative approaches to investigating picture-word interference eVects. The approach we
proposed involved using fMRI (de Zubicaray et al., 2001,
275
2002). We reasoned that an fMRI investigation of semantic
interference should produce diVerential activity in either (1)
the left middle temporal cortex, if the assumption of Levelt
and colleagues’ (1999) serial model was correct or, (2) in
Wernicke’s area, if the assumptions of Starreveld and La
Heij’s (1996) cascade-IA model were correct, given the Wndings of Indefrey and Levelt’s (2000) meta-analysis mentioned earlier.
The two models were chosen for comparison as they
provided mutually exclusive hypotheses concerning the
locus of the eVect. Instead, we found diVerential activity in
both cerebral regions (de Zubicaray et al., 2001). We interpreted this result as evidence of phonological coactivation,
refuting the strict serial model assumption that only a single target lemma is selected and then phonologically
encoded (e.g., Levelt et al., 1999). While this result did not
provide direct support for IA models either, we argued that
it could nevertheless be accommodated within a cascade-IA
model framework. In fact, the most parsimonious explanation for the result simply involves assuming cascaded processing (e.g., Morsella & Miozzo, 2002; Peterson & Savoy,
1998). Interactive models presuppose cascaded processing,
as in order to inXuence lexical selection phonological activation must occur beforehand.
Phonological priming in picture-word interference was
the subject of a second study (de Zubicaray et al., 2002).
Given the assumption by both model types that phonological facilitation occurs at word form encoding, we predicted
diVerential activity associated with this eVect in Wernicke’s
area. In addition, following from the results of previous
neuroimaging studies of priming eVects, we expected these
activity changes to manifest as signal reductions (Henson,
2003). For instance, Henson, Shallice, and Dolan (2000)
argue that the reduced activity observed in association with
repetition priming represents more eYcient or faster processing due to lowered thresholds for activating existing
representations via competitive spreading-activation mechanisms. Phonological facilitation is generally attributed to
targets requiring less activation to exceed a threshold for
selection due to the activation they receive from related
competitors (Levelt, 2001; Levelt et al., 1999; Starreveld &
La Heij, 1996). Both of these hypotheses were conWrmed,
providing support for the word form account (de Zubicaray et al., 2002). In addition, these results were interpreted
as supporting the view that facilitation and interference are
dissociable processes, as originally proposed by Lupker
(1982; see also MacLeod and MacDonald, 2000).
An additional and somewhat novel proposal that we
examined in our fMRI studies concerned how attentional
control might be implemented within the context of picture-word interference. Most models of Stroop interference
propose an important role for selective attention in its resolution, either through active suppression of distractor information or selective enhancement of target information
(MacLeod, 1991). Models of spoken word production have
tended not to elaborate upon the allocation of attention for
resolving picture-word interference: Carr (1999) suggested
276
G.I. de Zubicaray / Brain and Cognition 60 (2006) 272–281
that this oversight needs to be remedied in order for them
to progress to integrated models of human performance.
We reasoned that if attentional control processes were
involved in picture-word interference then, in the context of
an fMRI experiment, diVerential activity should be observable in cerebral regions analogous to those observed in neuroimaging studies of the Stroop eVect, especially the
prefrontal cortex (PFC) and anterior cingulate cortex
(ACC; MacLeod & MacDonald, 2000). The rationale for
implicating these regions was based upon a conXict monitoring role for the ACC proposed by Carter and colleagues
(e.g., Botwinick, Cohen, & Carter, 2004), and a well-established role for the PFC in the inhibitory control of interference (e.g., Fuster, Van Hoesen, Morecraft, & Semendeferi,
2000). Signal increases were observed in the ACC and PFC
in both fMRI studies (de Zubicaray et al., 2001, 2002). The
results motivated us to consider the nature of the attentional control system involved in spoken word production
in more detail.
Like Botwinick, Braver, Barch, Carter, and Cohen
(2001), we envisaged an attentional control mechanism
involving conXict monitoring; conXict being operationally
deWned as the simultaneous activation of incompatible,
mutually inhibiting representations. Following the detection of conXict, control processes are triggered for the purpose of resolving it, facilitating the production of the
appropriate response. Botwinick et al. (2001) provide
explicit computational implementations of conXict monitoring and cognitive control units connected via a feedback
loop, added to several well-known examples of IA models
of performance on varying tasks (e.g., Stroop task). Our
adoption of the conXict monitoring hypothesis necessitated
a reconsideration of the mechanisms for competitor (i.e.,
non-target node) deactivation implemented in the computational models of spoken word production we had examined. The serial and cascade-IA models of Levelt et al.
(1999) and Starreveld and La Heij (1996) both assume that
the activation levels of competing lexical representations
return quickly (decay) to a resting level once the node with
the highest activation has been selected. However, for the
proposed addition of a conXict monitoring unit to work, we
needed a word production model that incorporated inhibitory links.
The application of inhibitory links for competitor deactivation in spreading-activation models of word production
has been a contentious issue (Berg & Schade, 1992a; Dell &
O’Seaghdha, 1994). In fact, it is yet to be resolved, despite
the more recent literature being relatively quiescent on this
point (although see Miozzo & Caramazza, 2003 who favour
inhibition). We conducted a search for word production
models that could account for picture-word interference
data and implemented activation of incompatible, mutually
inhibiting representations. Harley’s (1993) IA model satisWed these criteria. This model includes inhibitory connections both within and between levels, and assumes that the
word forms of all competitors are activated and this activation is then suppressed by mutual inhibition. Harley’s
model is not alone in proposing inhibitory connections:
Berg and Schade (1992b) similarly proposed a model that
implemented lateral inhibition among competing nodes. It
should be noted that the inclusion of inhibitory links does
not preclude the involvement of decay-based mechanisms:
Harley’s (1993) model combines both.
According to this interpretation of picture-word interference, when conXict produced by competing, non-target
nodes is detected by the ACC, processes for implementing
inhibitory control in the PFC are engaged, managing the
activation in the network in order to achieve correct
response selection. Thus, this account counter-intuitively
expects similar involvement of both cerebral regions during
semantic interference and phonological facilitation, due to
the activation of competing non-target nodes (see de Zubicaray et al., 2002). Although considerations of conXict
monitoring are typically restricted to response selection
processes, it should be noted that conXicts can in theory
occur at any stage of processing, such as the perceptual
input level, resulting in ACC involvement (Botwinick et al.,
2001). While further research is needed to delineate the circumstances under which conXict occurs, Ferreira and Pashler (2002) have recently provided evidence of central
processing bottlenecks during lemma selection and word
form retrieval, indicating that both can probably be characterised as response selection processes.
Thus, a cognitive neuroimaging approach was used to
adjudicate between rival theories of spoken word production, with the conclusion being that only models incorporating cascaded processing were supported by the data. In
addition, the Wnding of additional cortical activity during
picture-word interference indicates that the competition is
not resolved completely at the lemma or word form levels
as existing models assume. Accordingly, an additional
attentional control mechanism was proposed to be added
to an existing cascade-IA model. In this way, the fMRI data
were used to reveal something new about the cognitive
architecture of spoken word production.
What impact have these fMRI studies had? Obviously,
Harley (2004a) and Coltheart (2004a) overlooked them
entirely. The studies were also overlooked by Indefrey and
Levelt (2004) in their updated meta-analysis. However, an
attentional control mechanism was subsequently added to
Levelt et al.’s (1999) serial model by Roelofs (2003), who
associated its operations with those of the prefrontal cortex
and ACC during Stroop task performance. It is yet to be
applied to model picture-word interference. Luckily, the
connectionists were more forthcoming. Dell and Sullivan
(2004) consider the Wnding of activity in both semantic and
phonological processing regions during semantic interference to be inconsistent with a serial model account attributing the eVect to only lemma selection, while supporting the
notion of multiple stages of lexical access. When the results
of the studies were discussed with Trevor Harley (personal
communication, 2003), he interpreted them as supporting
the inclusion of inhibitory mechanisms in word production
models (see Harley, 2004b).
G.I. de Zubicaray / Brain and Cognition 60 (2006) 272–281
1.4. Neuroimaging tests of single- vs. dual-process episodic
memory models
Another example of the application of cognitive neuroimaging comes from research into episodic memory. Two broad
classes of model have been proposed: Single-process models
derived from signal-detection theory (SDT) assume that recognition memory is based on a single strength-of-evidence
variable or signal (e.g., Glanzer, Adams, Iverson, & Kim,
1993; ShiVrin & Steyvers, 1997). In contrast, dual-process
theories assume that as well as a strength-of-memory variable (called, in this context, familiarity), recognition memory
is also based on an additional recall-like process called, recollection (e.g., Norman & O’Reilly, 2003; Yonelinas, 2002). As
per the cognitive science literature on spoken word production, two opposing camps claim the existing behavioural evidence supports their respective positions. The goal of
cognitive neuroimaging is to provide information that can
adjudicate between the diVerent classes of model.
Much of the recent cognitive neuroscience research has
focussed upon identifying single dissociations at the cognitive and neural levels to support the notion of separable
memory processes. In order to do so, these investigations
have generally employed two paradigms favoured by dualprocess theorists, namely the Remember-Know (RK) and
process-dissociation tasks (see Rugg & Yonelinas, 2003).
Neural correlates of the dual-process theoretical constructs
of recollection and familiarity have been proposed based
upon results from these experiments. Responses in the left
inferior parietal cortex and, less frequently, in the hippocampal formation in the medial temporal lobe (MTL) have
been attributed to recollection, while activity in adjacent
anterior MTL structures such as the amygdala, rhinal, and
parahippocampal cortices is proposed to represent a familiarity-based signal (Henson et al., 2004; Rugg & Yonelinas,
2003). A dissociation according to the direction of these
responses has also been proposed: Recollection should be
reXected in positive activity for studied vs. unstudied items,
while familiarity should show the opposite pattern (Henson
et al., 2004; Rugg & Yonelinas, 2003).
For example, Eldridge, Knowlton, Furmanski, Bookheimer, and Engel (2000) reported increased hippocampal
activity for remember (R) vs. know (K) judgements in recognition memory. This paper is often cited as providing evidence for a neural basis for a separate process of
recollection (e.g., Rugg & Yonelinas, 2003). However, the
RK paradigm was criticised some time ago by Donaldson
(1996), who suggested that R and K responses instead represent diVerent levels of decision conWdence, or the application of more or less stringent decision criteria in relation to
the strength of a single memory trace. Dunn (2004) subsequently conWrmed this empirically. Wixted and Stretch
(2004), themselves dual-process theorists, likewise acknowledged that the results of R > K comparisons in fMRI studies such as that of Eldridge et al. (2000) could not
necessarily be interpreted as providing evidence for the
operation of two distinct memory processes.
277
Process-dissociation paradigms have involved contrasting activity during source memory and item memory tasks
in fMRI studies. Weis et al. (2004) recently reported a dissociation in the form of increased hippocampal activity
during source recognition and reduced anterior MTL activity during item recognition. Dobbins, Rice, Wagner, and
Schacter (2003) used a similar paradigm and reported
increased hippocampal activity solely during successful
source retrieval that they interpreted as supporting a separable neural basis for recollection. However, single-process
theorists consider process-dissociation paradigms to simply
reXect the mental operations supporting two diVerent memory tasks with diVerent instructions. For example, Glanzer,
Hilford, and Kim’s (2004) SDT model represents source
and item recognition discrimination accuracy as occurring
along two diVerent dimensions. Thus, the results of the process-dissociation fMRI studies can also be interpreted as
being consistent with single process models.
Humphreys, Dennis, Chalmers, and Finnigan (2000)
suggested that neuroimaging studies should address
whether the diVerential activity observed in the processdissociation or RK procedures also occurs in yes/no recognition. This would answer the question of whether the
activity observed was due to the special paradigms used,
or whether it was a more general phenomena of episodic
memory. We recently used fMRI to test hypotheses from
single- and dual-process accounts concerning two wellestablished eVects in the yes/no recognition memory literature: The strength and word frequency eVects. The
strength eVect refers to the Wnding of increased hit rates
(HR; correct responses to studied words) at test when
items are repeated in a study list, whereas the word frequency eVect (WFE) refers to the superior HRs for low
frequency (LF) than high frequency (HF) words. Additionally, unstudied LF words are less likely than unstudied HF words to be judged incorrectly as belonging to the
study list (a “false alarm”; FA)—an example of a mirror
eVect (Glanzer et al., 1993).
In the Wrst study (de Zubicaray, McMahon, Eastburn,
Finnigan, & Humphreys, 2005a), we examined memory
models that attributed these eVects to processes occurring
during encoding. For example, Glanzer et al.’s (1993)
Attention Likelihood Theory (ALT) assumes that the WFE
is due to relatively greater attentional allocation to the features of LF words at encoding (see also Maddox & Estes,
1997). Many memory theories hypothesise that item repetition strengthens episodic memory representations by either
adding features to an existing trace or storing a novel one
(e.g., Landauer, 1975; ShiVrin & Steyvers, 1997). However,
Cary and Reder (2003) had noted during testing of their
dual-process model that repeated presentations of items
produced identical amounts of memory strengthening,
resulting in a poor Wt to the data from existing behavioural
studies. In order to achieve a better Wt, they suggested that
less attention or processing eVort might be entailed for
encoding each repeated presentation of an item compared
to the Wrst.
278
G.I. de Zubicaray / Brain and Cognition 60 (2006) 272–281
Buckner, Kelley, and Petersen (1999) had earlier
reviewed the results of a number of neuroimaging studies of
word encoding. These studies conWrmed a role for the left
prefrontal cortex (LPFC) in encoding verbal material. In
addition, the activity in the LPFC appears to be modulated
according to the eVort involved or allocation of attentional
resources during encoding. For example, Kensinger,
Clarke, and Corkin (2003) observed reductions in LPFC
activity during word encoding under divided attention conditions. We therefore considered the LPFC to be a plausible
candidate region for testing the predictions of both the
ALT account of the WFE and Cary and Reder’s (2003)
account of encoding processes contributing to the strength
eVect. The data conWrmed both assumptions: relatively
greater LPFC activity was observed during encoding of LF
vs. HF words, and words repeated during encoding showed
reduced LPFC activity compared to single word presentations (de Zubicaray et al., 2005a). The latter Wnding might
also be interpreted as supporting the notion that repetition
adds information to an existing representation rather than
establishing a new one (cf. Landauer, 1975).
In a second study, we contrasted the predictions of single- and dual-process models that attributed the WFE and
strength eVects to processes occurring solely at test (de
Zubicaray, McMahon, Eastburn, Finnigan, & Humphreys,
2005b). Unlike single-process models that assume item repetition to augment a unitary memory trace signal, dual-process theories consider repetition to strengthen both
familiarity and recollection (Cary & Reder, 2003; Joordens
& Hockley, 2000; Stretch & Wixted, 1998). With respect to
the WFE, the original single-process REM model attributed the LF word HR advantage to a familiarity-like signal
due to the relatively more diagnostic or distinctive features
of LF words (ShiVrin & Steyvers, 1997). The dual-process
extension of REM makes a similar assumption: the recall
or recollection mechanism does not favour LF words
(Malmberg, Holden, & ShiVrin, 2004). However, this view
contrasts with several other dual-process models that consider LF words to instead be more recollectable than HF
words (Cary & Reder, 2003; Joordens & Hockley, 2000;
Stretch & Wixted, 1998).
An alternative class of memory model, the context
noise model, makes diVerent assumptions about the
mechanisms responsible for the WFE and strength eVect.
Most memory models focus upon item information that
pertains to the features (e.g., orthographic, graphemic)
describing each word, whereas context information might
be best considered a lexico-semantic construct related to
the way in which a word is used (Steyvers & Malmberg,
2003). Context-noise models (e.g., Dennis & Humphreys,
2001; Sikström, 2001) instead assume that HF words are
subject to greater interference or noise due to the larger
number of pre-experimental contexts in which they have
been encountered. As LF words tend to have been
encountered in fewer contexts, they are more strongly
associated with the study context and the context that the
participant reinstates at test. Thus, their traces receive
relatively greater activation compared to those of HF
words. However, strengthening the association between a
word and the experimental context by repeating it at
study is not considered to interfere with the memory
retrieved to a diVerent word in the list (Dennis &
Humphreys, 2001).
To test these hypotheses using fMRI, we Wrst identiWed
regions of interest (ROIs) demonstrating old/new item
information eVects in recognition memory. These ROIs
included regions identiWed consistently in fMRI studies of
recognition memory, such as the inferior parietal cortex,
hippocampus, and anterior MTL. Analyses performed on
these ROIs revealed only one cerebral region that demonstrated a signiWcant eVect of item repetition or memory
strength; the anterior MTL, and this manifested as a signal
reduction. This is consistent with the assumption that
familiarity and implicit priming represent similar memory
processes (e.g., Mandler, 1980), as implicit priming eVects
are usually associated with reductions in cerebral activity in
fMRI experiments (Henson, 2003).3 In fact, this region was
virtually identical to that identiWed by Henson, Cansino,
Herron, Robb, and Rugg (2003) as representing a single,
strength-like memory factor such as familiarity in a metaanalysis of fMRI studies. None of the regions putatively
associated with recollection in RK or process-dissociation
fMRI studies showed a similar eVect (i.e., left inferior parietal cortex, hippocampus; Rugg & Yonelinas, 2003). We
therefore considered the results to be consistent with the
predictions of single-process models.
The ROI analyses similarly failed to Wnd an eVect of
word frequency in any of the regions that had demonstrated item information eVects. We interpreted this result
as refuting the predictions of single- and dual-process models that attribute the LF word HR advantage to item
retrieval processes occurring at test. This conclusion applies
equally to competing dual-process models that attribute the
eVect to either familiarity or recollection (Cary & Reder,
2003; Joordens & Hockley, 2000; Malmberg et al., 2004;
Stretch & Wixted, 1998). As context-noise models instead
attribute the WFE to lexico-semantic context dependent
processing, we predicted a priori that this should manifest
as diVerential activity in the left lateral temporal cortex
(LTC). Lesion and cognitive neuroscience studies have indicated a prominent role for the left LTC in lexical-semantic
processing across a variety of tasks (Levy, Bayley, & Squire,
2004), including lexical decision (word vs. non-word; Rissman, Eliassen, & Blumstein, 2003), semantic priming
(related vs. unrelated words; Copland et al., 2003), spoken
word production (Indefrey & Levelt, 2000; see above section on language production), and episodic memory
3
Some researchers consider Wndings of intact priming in amnesic patients to be evidence that priming and recognition memory might instead
be mediated by separable neural systems. However, Kinder and Shanks
(2003) were able to demonstrate the same pattern of Wndings using a single-system connectionist model. The model was also able to account for
double dissociations.
G.I. de Zubicaray / Brain and Cognition 60 (2006) 272–281
279
retrieval (Menon, Boyett-Anderson, Schatzberg, & Reiss,
2002). This prediction was conWrmed by the data (de Zubicaray et al., 2005b).
These studies illustrate the use of a cognitive neuroimaging approach to test the assumptions of various cognitive
models of memory concerning two prominent memory phenomena. The method is distinct to the cognitive neuroscience approach applied to date that has used diVerent
paradigms to support primarily one class of model through
the identiWcation of single dissociations. Of course, the conclusions can only pertain to the memory phenomena under
study. It is necessary to test single- and dual-process explanations of other memory eVects in yes/no recognition memory in order to refute either class of model. Luckily, the
memory literature is replete with such eVects.
testing theories that have been formalised in computational models. These models represent limited implementations of theoretical assumptions, and have the advantage
that all processes are made explicit (Seidenberg, 1993). I
also think it is a distinct advantage to cognitive neuroimaging if the nodes in these models are connected in ways
that resemble brain networks, as the signals derived from
neuroimaging studies can then be interpreted meaningfully
in context. However, it is probably suYcient for the
purposes of cognitive neuroimaging if computation is
carried out by simply spreading activation through a
network of units—as is the case with most current models.
1.5. Premises, premises
Various cognitive scientists have questioned, not unreasonably, the relevance of neuroimaging investigations that
fail to address questions of interest to cognitive science.
There is no doubt that such investigations have been conducted, and will continue to be conducted, under the rubric
of cognitive neuroscience. This is because neuroscientists
like to have their questions answered too. However, the
cognitive neuroimaging studies of language production and
episodic memory discussed above provide some initial support for the use of the method to inform and constrain uniWed models of cognitive and neural systems. It is not
appropriate for ultra-cognitive scientists to insist that they
have an a priori proof that neuroimaging investigations are
irrelevant to the study of cognition based upon one attempt
at a solution to the mind-body problem. The question is
whether cognitive neuroimaging can enhance our knowledge about the architecture of information-processing systems. The answer will not be determined in an armchair: It
will be decided empirically.
The cognitive neuroimaging approach that I have outlined above proceeds from a core assumption that information processing is achieved via the operations of neural
networks. This assumption is shared by the connectionists
who “base their models upon the known neurophysiology
of the brain and attempt to incorporate those functional
properties thought to be required for cognition” (Medler,
1998, p. 63). Localist representations are incorporated in a
number of connectionist models, and emerge during processing. The activation patterns of nodes or groups of
nodes in these networks can be recorded. Likewise, functional neuroimaging methods such as fMRI are presently
able to record the activity patterns of groups of neurons
(the fMRI signal having an approximately linear relationship to the local Weld potentials recorded by implanted
microelectrodes; see Logothetis & PfeuVer, 2004).
Of course, unlike cognitive modellers, cognitive neuroimagers do not have the luxury of an exhaustive a priori
knowledge of the network architecture they are studying.
Herein lies a signiWcant limitation of cognitive neuroimaging. How do we know what the localist representations
detected by functional neuroimaging actually represent?
Aside from applying identical cognitive paradigms to those
adopted in the cognitive science literature to manipulate
only the process(es) of interest, there needs to be a means of
ensuring that the localist representations detected in neuroimaging experiments are attributed appropriately. The
solution we adopted for our spoken word production studies above, for example, was to rely upon Indefrey and
Levelt’s (2000) theory-driven meta-analysis of neuroimaging data. In this way, the regions we tested were only those
displayed reliably across multiple studies of spoken word
production and capable of supporting directional hypotheses concerning speciWc processes of interest.
Given the aforementioned arguments and examples,
does this mean that the term cognitive neuroimaging should
pertain only to studies testing hypotheses from connectionist models of information processing? I do think it
necessary that cognitive neuroimaging restricts itself to
2. Conclusions: Toward cognitive neuroimaging tests of
information-processing models
Acknowledgments
I thank John Dunn and Katie McMahon for their helpful comments during the drafting of the manuscript. This
study was supported by Australian Research Council
(ARC) Discovery Project Grants DP0342945 and
DP0342656. Greig de Zubicaray is supported by an ARC
Research Fellowship.
References
Bechtel, W. (in press). The epistemology of evidence in cognitive neuroscience. In R. Skipper Jr., C. Allen, R. A. Ankeny, C. F. Craver, L. Darden,
G. Mikkelson, & R. Richardson (Eds.), Philosophy and the life sciences:
A reader. Cambridge, MA: MIT Press. <Available from: http://mechanism.ucsd.edu/~bill/epist.evidence.bechtel.july2004.pdf>.
Bechtel, W., & Abrahamsen, A. (2002). Connectionism and the mind: Parallel processing, dynamics, and evolution in networks (2nd ed.). Oxford:
Basil Blackwell.
Berg, T., & Schade, U. (1992a). The role of inhibition in a spreading-activation model of language production. I. The psycholinguistic perspective. Journal of Psycholinguistic Research, 21, 405–434.
280
G.I. de Zubicaray / Brain and Cognition 60 (2006) 272–281
Berg, T., & Schade, U. (1992b). The role of inhibition in a spreading-activation model of language production. II. The simulational perspective.
Journal of Psycholinguistic Research, 21, 435–462.
Block, N. (1980). Introduction: What is functionalism? In N. Block (Ed.),
Readings in philosophy of psychology (Vol. 1, pp. 171–184). Cambridge,
MA: Harvard University Press.
Block, N. (1995). The mind as the software of the brain. In D. Osherson, L.
Gleitman, S. Kosslyn, E. Smith, & S. Sternberg (Vol. Eds.), An invitation to cognitive science (Vol. 3, An invitation to cognitive science (Vol. 3,
2nd ed.). Cambridge, MA: MIT Press.
Botwinick, M. M., Braver, T. S., Barch, D. M., Carter, C. S., & Cohen, J. D.
(2001). ConXict monitoring and cognitive control. Psychological
Review, 108, 624–652.
Botwinick, M. M., Cohen, J. D., & Carter, C. S. (2004). ConXict monitoring
and anterior cingulate cortex: An update. Trends in Cognitive Sciences,
8, 539–546.
Bowers, J. S. (2002). Challenging the widespread assumption that connectionism and distributed representations go hand-in-hand. Cognitive
Psychology, 45, 413–445.
Bub, D. N. (2000). Methodological issues confronting PET and fMRI
studies of cognitive function. Cognitive Neuropsychology, 17, 467–484.
Buckner, R. L., Kelley, W. M., & Petersen, S. E. (1999). Frontal cortex contributes to human memory formation. Nature Neuroscience, 2, 311–
314.
Carr, T. H. (1999). How does WEAVER pay attention? Behavioral and
Brain Sciences, 21, 39–40.
Cary, M., & Reder, L. M. (2003). A dual-process account of the list-length
and strength-based mirror eVects in recognition. Journal of Memory
and Language, 49, 231–248.
Coltheart, M. (2002). Cognitive neuropsychology. In J. Wixted & H. Pashler (Eds.), Stevens’ handbook of experimental psychology. Methodology
in experimental psychology (3rd ed., Vol. 4, pp. 139–174). New York:
Wiley.
Coltheart, M. (2004a). Brain imaging, connectionism, and cognitive neuropsychology. Cognitive Neuropsychology, 21, 21–25.
Coltheart, M. (2004b). Are there lexicons? Quarterly Journal of Experimental Psychology A, 57, 1153–1171.
Copland, D. A., de Zubicaray, G. I., McMahon, K. L., Eastburn, M. M.,
Wilson, S. J., & Chenery, H. J. (2003). Brain activity during automatic
semantic priming revealed by event-related fMRI. Neuroimage, 20,
302–310.
de Zubicaray, G. I., McMahon, K. L., Eastburn, M. E., Finnigan, S., &
Humphreys, M. S. (2005a). fMRI evidence of word frequency and
strength eVects during episodic memory encoding. Cognitive Brain
Research, 22, 439–450.
de Zubicaray, G. I., McMahon, K. L., Eastburn, M. E., Finnigan, S., &
Humphreys, M. S. (2005b). fMRI evidence of word frequency and
strength eVects during recognition memory. Cognitive Brain Research,
24, 587–598.
de Zubicaray, G. I., McMahon, K. L., Eastburn, M. M., & Wilson, S. J.
(2002). Orthographic/phonological facilitation of naming responses in
the picture-word task: An event-related fMRI study using overt vocal
responding. Neuroimage, 16, 1084–1093.
de Zubicaray, G. I., Wilson, S. J., McMahon, K. L., & Muthiah, S. (2001).
The semantic interference eVect in the picture-word paradigm: An
event-related fMRI study employing overt responses. Human Brain
Mapping, 14, 218–227.
Dell, G. S., & Sullivan, J. M. (2004). Speech errors and language production: Neuropsychological and connectionist perspectives. In B. H. Ross
(Ed.), The psychology of learning and motivation (pp. 63–108). San
Diego: Elsevier.
Dell, G. S., & O’Seaghdha, P. G. (1994). Inhibition in interactive activation
models of linguistic selection and sequencing. In D. Dagenbach & T. H.
Carr (Eds.), Inhibitory processes in attention, memory, and language.
San Diego: Academic Press.
Dennett, D. C. (1991a). Granny’s campaign for safe science. In B. Loewer
& G. Rey (Eds.), Meaning in mind: Fodor and his critics (pp. 87–94).
Cambridge, MA: Blackwell.
Dennett, D. C. (1991b). Consciousness explained. Toronto: Little, Brown &
Company.
Dennis, S., & Humphreys, M. S. (2001). A context noise model of episodic
word recognition. Psychological Review, 108, 452–477.
Dobbins, I. G., Rice, H. J., Wagner, A. D., & Schacter, D. L. (2003). Memory orientation and success: Separable neurocognitive components
underlying episodic recognition. Neuropsychologia, 41, 318–333.
Donaldson, W. (1996). The role of decision processes in remembering and
knowing. Memory & Cognition, 24, 523–533.
Dunn, J. C. (2004). Remember-know: A matter of conWdence. Psychological Review, 111, 524–542.
Eldridge, L. L., Knowlton, B. J., Furmanski, C. S., Bookheimer, S. Y., &
Engel, S. A. (2000). Remembering episodes: A selective role for the hippocampus during retrieval. Nature Neuroscience, 3, 1149–1152.
Ferreira, V., & Pashler, H. (2002). Central bottleneck inXuences on the processing stages of word production. Journal of Experimental Psychology: Human Learning and Memory, 28, 1187–1199.
Fodor, J. (1981). Representations. Cambridge, MA: MIT Press.
Fuster, J. M., Van Hoesen, G. W., Morecraft, R. J., & Semendeferi, K.
(2000). Executive systems. In B. S. Fogel, R. B. SchiVer, & S. M. Rao
(Eds.), Synopsis of neuropsychiatry (pp. 229–242). Philadelphia: Lippincott, Williams and Wilkins.
Glanzer, M., Adams, J. K., Iverson, G. J., & Kim, K. (1993). The regularities of recognition memory. Psychological Review, 100, 546–567.
Glanzer, M., Hilford, A., & Kim, K. (2004). Six regularities of source recognition. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 30, 1176–1195.
Harley, T. A. (1993). Phonological activation of semantic competitors during lexical access in speech production. Language and Cognitive Processes, 8, 291–309.
Harley, T. A. (2004a). Does cognitive neuropsychology have a future?
Cognitive Neuropsychology, 21, 3–16.
Harley, T. A. (2004b). Promises, promises. Cognitive Neuropsychology, 21,
51–56.
Henson, R. N. A. (2003). Neuroimaging studies of priming. Progress in
Neurobiology, 70, 53–81.
Henson, R. N. A. (2004). What can functional neuroimaging tell the experimental psychologist? Quarterly Journal of Experimental Psychology A,
58, 193–233.
Henson, R. N. A., Cansino, S., Herron, J. E., Robb, W. G. K., & Rugg, M.
D. (2003). A familiarity signal in human anterior medial temporal cortex? Hippocampus, 12, 301–304.
Henson, R. N. A., Shallice, T., & Dolan, R. J. (2000). Neuroimaging evidence for dissociable forms of repetition priming. Science, 287, 1269–
1272.
Humphreys, M. S., Dennis, S., Chalmers, K. A., & Finnigan, S. (2000).
Dual processes in recognition: Does a focus on measurement operations provide a suYcient foundation? Psychonomic Bulletin & Review,
7, 593–603.
Indefrey, P., & Levelt, W. J. M. (2000). The neural correlates of language
production. In M. Gazzaniga (Ed.), The new cognitive neurosciences
(pp. 845–865). Cambridge, MA: MIT Press.
Indefrey, P., & Levelt, W. J. M. (2004). The spatial and temporal signatures
of word production components. Cognition, 101–144.
Joordens, S., & Hockley, W. E. (2000). Recollection and familiarity
through the looking glass: When old does not mirror new. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 26, 1534–
1555.
Kensinger, E. A., Clarke, R. C., & Corkin, S. (2003). What neural processes
underlie successful encoding and retrieval? A functional magnetic resonance imaging study using a divided attention paradigm. Journal of
Neuroscience, 23, 2407–2415.
Kinder, A., & Shanks, D. R. (2003). Neuropsychological dissociations
between priming and recognition: A single-system connectionist
account. Psychological Review, 110, 728–744.
Landauer, T. K. (1975). Memory without organization: Properties of a
model with random storage and undirected retrieval. Cognitive Psychology, 7, 495–531.
G.I. de Zubicaray / Brain and Cognition 60 (2006) 272–281
Levelt, W. J. M. (1999). Models of word production. Trends in Cognitive
Sciences, 3(6), 223–232.
Levelt, W. J. M. (2001). Spoken word production: A theory of lexical
access. Proceedings of the National Academy of Sciences of the United
States of America, 98, 13464–13471.
Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1999). A theory of lexical
access in speech production. Behavioral and Brain Sciences, 22, 1–75.
Levy, D. A., Bayley, P. J., & Squire, L. R. (2004). The anatomy of semantic
knowledge: Medial vs. lateral temporal lobe. Proceedings of the
National Academy of Sciences of the United States of America, 101,
6710–6715.
Logothetis, N. K., & PfeuVer, J. (2004). On the nature of the BOLD fMRI
contrast mechanism. Magnetic Resonance Imaging, 22, 1517–1531.
Lupker, S. J. (1982). The role of phonetic and orthographic similarity in
picture-word interference. Canadian Journal of Psychology, 36, 349–
367.
MacLeod, C. M. (1991). Half a century of research on the Stroop eVect:
An integrative review. Psychological Bulletin, 109, 163–203.
MacLeod, C. M., & MacDonald, P. A. (2000). Inter-dimensional interference in the Stroop eVect: Uncovering the cognitive and neural anatomy
of attention. Trends in Cognitive Sciences, 4, 383–391.
Maddox, W. T., & Estes, W. K. (1997). Direct and indirect stimulus-frequency eVects in recognition. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 23, 539–559.
Malmberg, K. J., Holden, J. E., & ShiVrin, R. M. (2004). Modeling the
eVects of repetitions, similarity, and normative word frequency on
judgments of frequency and recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30, 319–331.
Mandler, G. (1980). Recognizing: The judgment of previous occurrence.
Psychological Review, 87, 252–271.
Marr, D. (1982). Vision. San Francisco: W.H. Freeman.
Medler, D. A. (1998). A brief history of connectionism. Neural Computing
Surveys, 1, 61–101.
Medler, D. A., Dawson, M. R. W., & Kingstone, A. (2005). Functional
localization and double dissociations: The relationship between internal structure and behavior. Brain and Cognition, 57, 146–150.
Menon, V., Boyett-Anderson, J. M., Schatzberg, A. F., & Reiss, A. L.
(2002). Relating semantic and episodic memory systems. Cognitive
Brain Research, 13, 261–265.
Miozzo, M., & Caramazza, A. (2003). When more is less: A counterintuitive eVect of distractor frequency in the picture-word interference paradigm. Journal of Experimental Psychology: General, 132, 228–258.
Morsella, E., & Miozzo, M. (2002). Evidence for a cascade model of lexical
access in speech production. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 28, 555–563.
Nickels, L. A. (2002). Theoretical and methodological issues in the cognitive
neuropsychology of spoken word production. Aphasiology, 16, 3–19.
Norman, K. A., & O’Reilly, R. C. (2003). Modeling hippocampal and neocortical contributions to recognition memory: A complementary learning systems approach. Psychological Review, 110, 611–646.
Page, M. (2000). Connectionist modelling in psychology: A localist manifesto. Behavioral and Brain Sciences, 23, 443–467.
Peterson, R. R., & Savoy, P. (1998). Lexical selection and phonological
encoding during language production: Evidence for cascaded process-
281
ing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24, 539–557.
Pylyshyn, Z. W. (1984). Computation and cognition: Towards a foundation
for cognitive science. Cambridge, MA: MIT Press.
Raichle, M. E. (1997). Brain imaging. In M. S. Gazzaniga (Ed.), Conversations in the cognitive neurosciences. Cambridge, MA: MIT Press.
Rapp, B., & Goldrick, M. (2004). Feedback by any other name is still interactivity: A reply to Roelofs’ comment on Rapp & Goldrick (2000).
Psychological Review, 111, 573–578.
Rissman, J., Eliassen, J. C., & Blumstein, S. E. (2003). An event-related
fMRI investigation of implicit semantic priming. Journal of Cognitive
Neuroscience, 15, 1160–1175.
Roelofs, A. P. A. (2003). Goal-referenced selection of verbal action: Modeling attentional control in the Stroop task. Psychological Review, 110,
88–125.
Roelofs, A. P. A. (2004). Comments—Comprehension-based versus production-internal feedback in planning spoken words: A rejoinder to
Rapp and Goldrick (2004). Psychological Review, 111, 579–580.
Rugg, M. D., & Yonelinas, A. P. (2003). Human recognition memory: A
cognitive neuroscience perspective. Trends in Cognitive Sciences, 7,
313–319.
Rumelhart, D. E., McClelland, J. L., & the PDP Research Group (1986).
Parallel distributed processing: Explorations in the microstructure of
cognition (Vol. 1 and 2). Cambridge, MA: MIT Press.
Seidenberg, M. (1993). Connectionist models and cognitive science. Psychological Science, 4, 228–235.
Shallice, T. (1988). From neuropsychology to mental structure. Cambridge:
Cambridge University Press.
ShiVrin, R. M., & Steyvers, M. (1997). A model for recognition memory:
REM: Retrieving EVectively from Memory. Psychonomic Bulletin &
Review, 4, 145–166.
Sikström, S. (2001). The variance theory of the mirror eVect in recognition
memory. Psychonomic Bulletin & Review, 8, 408–438.
Starreveld, P. A., & La Heij, W. (1996). Time course analysis of semantic
and orthographic context eVects in picture naming. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 896–918.
Steyvers, M., & Malmberg, K. J. (2003). The eVect of normative context
variability on recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 29, 760–766.
Stretch, V., & Wixted, J. T. (1998). On the diVerence between strengthbased and frequency-based mirror eVects in recognition memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24,
1379–1396.
Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, 12, 643–662.
Weis, S., Specht, K., Klaver, P., Tendolkar, I., Willmes, K., Ruhlmann, J., et
al. (2004). Process-dissociation between contextual retrieval and item
recognition. Neuroreport, 15, 2729–2733.
Wixted, J. T., & Stretch, V. (2004). In defense of the signal-detection interpretation of remember/know judgments. Psychonomic Bulletin &
Review, 11, 616–641.
Yonelinas, A. P. (2002). The nature of recollection and familiarity: A
review of 30 years of research. Journal of Memory and Language, 46,
441–517.