Brain Mechanisms for Reading Words and Panagiotis G. Simos1, Joshua I. Breier1, Jack M.

Fletcher2,
Barbara R. Foorman2, Eduardo M. Castillo1 and Andrew C.
Pseudowords: an Integrated Approach Papanicolaou1

1
Vivian L. Smith Center for Neurologic Research, Department
of Neurosurgery and 2Department of Pediatrics, The University
of Texas–Houston, Medical School, Houston, TX 77030, USA

The present study tested two predictions of dual-process models of study, we obtained evidence that the two purported mech-
reading: (i) that the brain structures involved in sublexical anisms for reading may differ in at least one important feature.
phonological analysis and those involved in whole-word phono- The mechanism that subserves assembled phonology depends
logical access during reading are different; and (ii) that reading of on the activity of the posterior part of the left superior temporal
meaningful items, by means of the addressed phonology process, is gyrus (STGp), whereas the mechanism that is responsible for
mediated by different brain structures than reading of meaningless addressed phonology does not necessarily involve this region
letter strings. We obtained brain activation profiles using Magnetic (Simos et al., 2000a). Unfortunately, the extent of the cortical
Source Imaging and, in addition, pronunciation latencies during area that can be examined in the operating room, with cortical
reading of: (i) exception words (primarily involving addressed stimulation techniques, is very limited. Functional brain imaging
phonology and having meaning), (ii) pseudohomophones (requiring
techniques, on the other hand, do not share this limitation and
assembled phonology and having meaning), and (iii) pseudowords
(requiring assembled phonology but having no meaning). Reading of are therefore capable of providing a more complete picture of
meaningful items entailed a high degree of activation of the left the cerebral mechanism of complex linguistic functions such as
posterior middle temporal gyrus (MTGp) and mesial temporal lobe reading.
areas, whereas reading the meaningless pseudowords was Previous approaches to this problem using functional imaging
associated with much reduced activation of these two regions. methods have relied on data from a single source, namely the
Reading of all three types of print resulted in activation of the spatial profile of brain areas that show increased levels of activity
posterior superior temporal gyrus (STGp), inferior parietal and basal during reading tasks, obtained using a particular functional
temporal areas. In addition, pronunciation speed of exception words imaging technique (Pugh et al., 1996; Price et al., 1996; Rumsey
correlated significantly with the onset of activity in MTGp but not et al., 1997). To be successful, however, this undertaking
STGp, whereas the opposite was true for pseudohomophones and requires additional information on: (i) how different activated
pseudowords. These findings are consistent with the existence of
brain areas may operate together during reading tasks, and
two different brain mechanisms that support phonological process-
(ii) the degree and type of contribution of each area to reading
ing in word reading: one mechanism that subserves assembled
phonology and depends on the posterior part of STGp, and a second performance.
mechanism that is responsible for pronouncing words with rare The present report describes an attempt to integrate data
print-to-sound correspondences and does not necessarily involve derived from two noninvasive techniques. These techniques are,
this region but instead appears to depend on MTGp. first, magnetic source imaging (MSI) performed in neuro-
logically intact volunteers during reading tasks and, second,
monitoring of reading performance (naming speed) in the same
Introduction
participants. MSI is unique among other functional imaging
It has been proposed that reading is subserved by two
techniques for its ability to provide brain activation profiles with
independent mechanisms [for a review, see Coltheart et al.
high spatiotemporal resolution. It can be used to determine not
(Coltheart et al., 1993)]. This ‘dual-route’ model is based mainly
only which areas participate in reading, but also how these areas
on evidence for a double dissociation in the ability to read
might interact with each other in real time to enable this
pseudowords and real words in patients with acquired dyslexia
function (Breier et al., 1998, 1999a). In addition, the capacity of
secondary to brain damage (Marshal and Newcombe, 1973;
Shallice and Warrington, 1980). According to this model, an MSI to provide accurate and detailed maps of language-specific
addressed or lexical mechanism mediates the conversion of cortex has been validated in the context of two large clinical
visual input to a whole-word phonological representation by studies against invasive cortical mapping techniques (Breier et
means of access to a word-specific, lexical representation. al., 1999b, 2001; Papanicolaou et al., 1999; Simos et al., 1999,
Presumably, this route can only be used for reading aloud real 2000a).
words, and is required for pronouncing words with peculiar This study addresses two issues: first, the dual-process
print-to-sound correspondences, such as PINT and COLONEL hypothesis that the brain mechanism for reading words that
(i.e. exception words). In contrast, the assembled or sublexical require assembled phonology, in experienced readers, is
mechanism involves mapping of individual orthographic different from the mechanism for reading words that do not; and
segments onto the appropriate phonological elements to arrive second, whether different brain structures mediate reading of
at a complete phonological representation (a process also meaningful versus meaningless words. If such distinct regions
referred to as ‘phonological decoding’). Assembled phonology is exist, they may be involved in lexical access, which according
required for reading unfamiliar letter strings (pseudowords), as to dual-route models mediate pronunciation of exception words.
well as pseudohomophones (i.e. letter strings, such as BURTH, To address both issues we obtained MSI-derived brain activation
which are pronounced the same as their real-word counterparts, profiles during reading of three types of print: exception words
but have different spellings). (relying more on addressed phonology and having meaning),
Recently, in the context of an electrocortical stimulation pseudohomophones (requiring assembled phonology and also

© Oxford University Press 2002. All rights reserved. Cerebral Cortex Mar 2002;12:297–305; 1047–3211/02/$4.00
having meaning) and pseudowords (requiring assembled other source, two simultaneous source solutions were retained only if the
phonology but having no meaning). The prediction was that corresponding dipoles were at least 5 cm apart. Using this method, no
activity in posterior temporal and inferior parietal structures more than two sources in different anatomical regions can be computed
in each hemisphere at each 4 ms time bin. In this way a maximum
known to be involved in reading (Breier et al., 1998, 1999a;
number of 1000 ms/4 = 250 sources can be computed for each hemi-
Pugh et al., 1996; Simos et al., 2000b,c) would differentiate
sphere during the entire recording epoch. Reliably localized activity
processing of (i) meaningful versus meaningless items and (ii) sources [i.e. those passing a 0.90 best-fit correlation criterion, see (Breier
letter strings that require assembled phonology versus those that et al., 1999b; Simos et al., 1999)] were co-registered on structural MRI
do not. scans, and the anatomical location of each source was determined using a
standard MRI atlas (Damasio, 1995).
The sum of all acceptable sources localized in a particular area (i.e. left
Experiment 1 STGp), starting at ∼150 ms after stimulus onset, when the first ‘wave’ of
activity in the primary visual cortex has subsided, and ending 1 s later,
Materials and Methods served as a metric of the degree of stimulus-locked activation of that area.
The validity of this measure as an index of regional activation has been
Participants established in several studies involving neurologically intact volunteers
Sixteen adults (10 males, mean age: 29, range: 25–42 years), who had no and patients (Simos et al., 1998, 1999; Breier et al., 1999b, 2001;
history of neurological or psychiatric disorder, learning disability or visual Papanicolaou et al., 1999). On the basis of previous MSI studies on
impairment were studied. In addition, all participants were right-handed reading (Breier et al., 1998; 1999a; Simos et al., 2000b,c), we examined
with English as their primary language. They were paid $30.00 for their the following areas in each hemisphere: posterior third portion of the
participation. This study (as well as the one reported below in Experi- superior temporal gyrus (STGp), posterior third portion of the middle
ment 2) had been approved by the University of Texas Institutional temporal gyrus (MTGp), supramarginal gyrus (SMG), angular gyrus
Review Board. All participants were asked to sign a consent form after the (A NG), mesial temporal lobe (MTL) including the hippocampus and
nature of the procedures involved had been explained to them. parahippocampal gyrus, and basal temporal cortex (BTC) comprising the
fusiform and lingual gyri. Activity sources were also noted in the inferior
frontal gyrus (IFG), predominantly in the left hemisphere (Broca’s area)
Stimuli and Tasks
and in sensorimotor cortex, bilaterally. The proportion of subjects who
Each participant was tested on three word pronunciation tasks involving
showed activity in either of the two areas in at least one condition and
(i) exception words, (ii) pseudohomophones and (iii) pseudowords. Each
hemisphere was not significant (9/16 subjects, P < 0.81 for inferior frontal
list contained 80 monosyllabic letter strings ranging in length from four to
and 10/16 subjects, P < 0.45 according to the binomial test), and data
six letters. Pseudowords and pseudohomophones were the same stimuli
pertaining to these regions will be mentioned separately in the Results
used by McCann and Besner in their Experiment 1 (McCann and Besner,
section.
1987). Words in the exception list were adapted from Pugh et al. (Pugh et
al., 1997) and Glushko’s (Glushko, 1979) exception inconsistent word
lists and were generally items with very rare print-to-sound correspond-
ences. Mean frequency of occurrence for the exception words was 138 Results
(range: 3–1700) per million in the Kucera and Francis corpus (1967) and Reading error rates were too low to allow further analyses: group
100 (range: 2–1264 occurrences) per million for the words from which mean error rates were 2.3% (range: 1–4%), 4.2% (range: 2–5%)
pseudohomophones were derived (P > 0.61). and 5.1% (range: 3–7%) for exception words, pseudohomo-
phones and pseudowords, respectively.
Procedure Individual brain activation profiles obtained during reading of
The MSI scan was performed during all three tasks within a single session all three types of stimuli closely resembled those observed in
in a different random order across participants. Printed stimuli were previous MSI studies in the context of silent reading tasks (see
presented for one second in order to prevent potential contamination of Fig. 1). They feature initial activation of the mesial occipital
the event-related field (ERF) record by visual offset responses. The cortices bilaterally (within the first 150 ms after stimulus onset),
interstimulus interval varied randomly between 3 and 4 s across trials.
followed by activity in basal temporal cortices predominantly in
The stimuli were presented in lowercase letters through a Sharp LCD
the left hemisphere (starting within 200 ms post-stimulus
projector (Model XG-E690U, Sharp Electronics Corporation, Mahwah, NJ,
USA) controlled by a Macintosh G3 portable computer running SuperLab onset). In the next several hundred milliseconds, the profiles
Pro. They were projected on a white screen located ∼1.5 m in front of the entailed activation of posterior temporal and inferior parietal
participant and subtended 1.0–2.0° and 0.5° of horizontal and vertical and, in some cases, frontal areas as well as of mesial temporal
visual angle, respectively. regions. With the notable exception of MTGp, the degree of
The principles underlying the MSI method as well as MSI data activity in all other temporal areas was strongly left-hemisphere
collection and analysis methods are described in detail elsewhere dominant.
(Papanicolaou et al., 1999) and will only be brief ly outlined here. MSI data Across reading tasks, the most striking pattern that could be
were recorded in a magnetically shielded room with a whole-head discerned by mere visual inspection of brain activation profiles
neuromagnetometer (WH2500, 4D Neuroimaging, San Diego, CA)
in each participant was the near complete absence of activity
consisting of 148 magnetometer coils. The precise location of the
intracranial sources of the observed evoked fields were computed at sources in MTGp and MTL during pseudoword reading. In
successive 4 ms intervals for a period of 1 s after the onset of the stimuli contrast, activity in these regions was detected in every
using standard algorithms (Sarvas, 1987). Source estimation was participant during reading of exception words. In the left MTGp
performed separately for each hemisphere and was attempted only when and MTL, greater activation was found for exception as
the surface distribution of magnetic f lux was dipolar, i.e. consisted of a compared with pseudowords in 15/16 participants. In both
single region of magnetic outf lux and a single region of magnetic inf lux. areas the degree of activity was greater during reading of
This surface map configuration usually indicates the presence of a single pseudohomophones as compared with pseudowords in 14/16
underlying active cortical patch that can be modeled as an equivalent
participants.
current dipole (ECD). Occasionally, two distinct dipolar distributions
The data were analyzed using a multivariate approach to
were discerned, typically, over the left hemisphere, one over anterior
frontal regions and the second over temporo-occipital areas. In that A NOVA with three within-subject variables [Type of Letter
case, source estimation was performed for both dipolar distributions String (exception, pseudohomophone, pseudoword), Area
independently. To avoid localization errors produced by smearing of (STGp, MTGp, SMG, A NG, BTC, and MTL) and Hemisphere
the magnetic f lux produced by one source by the f lux induced by the (left, right)]. The Bonferroni method was used for maintaining

298 Imaging Word and Pseudoword Reading • Simos et al.

 family-wise Type I error under 0.05 when pairwise means
comparisons were performed. Pairwise comparisons that
followed a significant Stimulus Type by Area interaction were
based on data collapsed across hemispheres for each area and
task. Pairwise comparisons that followed a significant Area by
Hemisphere interaction were based on data collapsed across
tasks for each area and hemisphere. The main ANOVA yielded a
significant Type by Area interaction, F(10,150) = 2.06, P < 0.031,
and a significant Area by Hemisphere interaction, F(5,75) = 6.24,
P < 0.0001. Pairwise comparisons revealed that activity in MTGp
was significantly increased during reading of exception words,
t(15) = 4.27, P < 0.001, and pseudohomophones, t(15) = 3.95, P <
0.001, as compared with pseudowords. Reading of exception
words and pseudohomophones was associated with approxi-
mately twice or three times as much activity in MTGp as
compared with pseudoword reading. A similar (nearly two-fold)
increase in activity was noted in MTL, although the results of
the pairwise comparisons did not satisfy our stringent P value
criterion of 0.0028 [t(15) = 2.02, P < 0.061 for pseudohomo-
phones versus pseudowords, and t(15) = 3.06, P < 0.008 for
exception words versus pseudowords]. These differences in
degree of activation across reading tasks are presented graphic-
ally in Figure 2.
A closer look at hemisphere differences at each area revealed
a significant left hemisphere predominance in the degree of
activation for STGp [t(15) = 4.46, P < 0.0001)], BTC [t(15) = 3.60,
P < 0.003] and MTL [t(15) = 5.88, P < 0.0001)] regardless of
stimulus type. These differences were highly consistent across
participants: greater left than right STGp activation was found in
94% of participants for exception words, in 82% for pseudo-
homophones and in 92% for pseudowords. Corresponding
figures for MTL were 94, 82 and 90%, and for BTC, 75, 67 and
67%, respectively.
Given that the proportion of participants who showed activity
sources in IFG and sensorimotor areas was not significant, data
from these areas were examined separately and resulting find-
ings should be considered as preliminary. A two-way ANOVA
conducted on data from IFG with Task (3) and Hemisphere (2) as
the within-subject factors revealed a significant Hemisphere
main effect, F(1,15) = 5.19, P < 0.038 and a marginally significant
Task main effect, F(2,30) = 3.38, P < 0.071. These results
ref lected the predominantly left hemisphere IFG activation and
a trend for stronger activation during the pseudohomophone
and pseudoword tasks compared with the exception word
reading task. None of the ANOVA effects approached signifi-
cance for activity sources in sensorimotor areas (P > 0.1).
The results outlined above supported our first prediction,
namely that meaningful items entailed high degree of activation
of the MTGp and MTL regions, whereas reading meaningless
letter strings entailed greatly reduced activation of these two
regions. However, the degree of activity in superior temporal
and temporoparietal areas did not differentiate between
exception words and pseudohomophones (i.e. stimuli that are
meaningful, yet differ in the amount of assembled phonology
processing that they require). This was somewhat unexpected
given the results of our previous electrocortical stimulation
Figure 1. Activity sources from one representative case projected on a 3-D rendering
of the participant’s MRI (Experiment 1). Clusters of activity sources computed at 4 ms study (Simos et al., 2000a), which suggested that at least one
intervals after the presentation of each printed stimulus were projected on the brain temporal lobe area (STGp) is a key component of the mechanism
surface for easier visualization. Sources occurring between 100 and 300 ms after for reading aloud items that require assembled phonology,
stimulus onset (shown in yellow) were typically localized in basal temporal cortices. although pronunciation of exception words may not depend on
Posterior temporal and inferior parietal sources (shown in orange) usually became active STGp. Therefore, it appeared likely that STGp activation
later between 300 and 1000 ms after stimulus onset. Note the abundance of activity
sources in the left middle temporal gyrus and mesial temporal cortex during exception observed in the present study during exception word reading
word and pseudohomophone reading and the lack of activity sources in these areas may indicate automatic engagement of this region. This may be
during reading of meaningless letter strings. part of an attempt to apply assembled phonology operations or,

Cerebral Cortex Mar 2002, V 12 N 3 299

 and pseudohomophone reading, its onset should correlate with
the speed of articulation of these stimuli. Further, if addressed
phonology operations depend upon access to whole-word repre-
sentations, and if MTGp and MTL are involved in the retrieval
of such representations, onset of activity in these areas should
correlate with speed of articulation of exception words. Finally,
if MTGp activation is only secondary to the process of reading
aloud pseudohomophones, then its onset latency should not
correlate with articulation speed for this type of letter string.
To assess these predictions, we obtained pronunciation latencies
for each of the three types of stimuli used in Experiment 1 from
a subgroup of the participants in that experiment.

Experiment 2

Materials and Methods

Participants
Fourteen of the participants in Experiment 1 agreed to visit the laboratory
for a second testing session (nine males, mean age: 29, range: 25–42
years). They were paid $30.00 for their participation.

Stimuli and Tasks

A subset of 30 stimuli was randomly selected to form each of the three
lists used in this session. Mean word frequencies for these shorter lists
were not significantly different from those used in the corresponding MSI
sessions (mean = 128 and 119 for exception words and pseudohomo-
phones, respectively, P < 0.48). To avoid practice effects, this part of the
study was conducted at least 1 month after the MSI scan.
Each participant was seated in a quiet room at a distance of 30 cm in
front of the computer screen and asked to read aloud letter strings
presented once every 3–4 s as rapidly as possible without neglecting
accuracy. The letter strings remained on the screen until the computer
registered the participant’s vocal response. Pronunciation errors were
also recorded by an experimenter seated next to the participant. The
order of task presentation was again counterbalanced across participants.

Results
Again, reading errors were too low to allow further analyses (1.8,
Figure 2. Mean number of activity sources in six posterior temporal and inferior parietal 3.0 and 4.5% for exception words, pseudohomophones and
areas for each of the three stimulus types in Experiment 1. Vertical bars represent pseudowords, respectively). Pronunciation latencies are
standard error values. Significant task differences were found in the MTGp and MTL. A presented in Table 1. On average, exception words were
near three-fold increase was found in the degree of MTGp activation (bilaterally) pronounced 60 ms faster than pseudohomophones, which were
between exception words and pseudowords. This pattern was apparent in all but one pronounced 22 ms faster than pseudowords. The latter
participants. On average, the left MTL showed approximately twice as much activity
difference is essentially identical to that reported by McCann and
during reading of exception words and pseudohomophones than during pseudoword
reading. Again, this pattern was observed in the vast majority of individual brain Besner for the same stimulus lists (McCann and Besner, 1987). A
activation profiles (i.e. in 14/16 participants). one-way ANOVA, computed on mean pronunciation latencies
from each participant, with Type of Letter String as a within
subjects variable, was significant, F(2,26) = 6.88, P < 0.004.
alternatively, ref lect a process that is secondary to whole-word Pairwise comparisons revealed a significant difference between
phonological access. In a similar manner, MTGp activation exception words and both pseudohomophones [t(13) = 2.17,
during pseudohomophone reading would appear to ref lect P < 0.049] and pseudowords [t(13) = 3.57, P < 0.003]. The
lexical access that is secondary to phonological assembly. Activ- difference between pseudohomophones and pseudowords,
ity in IFG, although present in only a subset of the participants, although highly consistent across participants (with 11/14 cases
was clearly lateralized to the left hemisphere, and there were showing the effect), did not reach significance.
indications that Broca’s area showed increased activation in tasks A lthough differences in mean word frequency between
that required explicit phonological decoding operations. This is exception words and pseudohomophones were small, we
in agreement with previous reports using MSI (Breier et al., wanted to ensure that they did not exert a significant inf luence
1999a) and functional magnetic resonance imaging (fMRI) on naming speed. For this purpose, we computed, for each
studies (Pugh et al., 1996). participant, the correlation between naming latency and word
If the above explanation regarding the relative roles of STGp frequency for exception words and pseudohomophones
and MTGp in reading is correct, it should follow that the onset of according to the Kucera and Francis norms (Kucera and Francis,
STGp activation would not correlate with the speed of articula- 1967). For the latter, the frequency of the real word from
tion of exception words. Moreover, if STGp activation indeed which the pseudohomophone was derived was used. Pearson
ref lected operations of assembled phonology for pseudoword correlation coefficients ranged between –0.16 and 0.14 across

300 Imaging Word and Pseudoword Reading • Simos et al.

 areas (in all cases P < 0.0001). There were no significant
Table 1
Mean pronunciation latencies and mean latencies of onset of regional activation in Experiments 1 differences across participants in the onset of activation in the
and 2 (in ms after stimulus onset; SD values in parentheses) following areas: STGp, MTGp, SMG, ANG and MTL.
To summarize, the results from Experiment 2 were in accord
Left hemisphere Pronunciation STGp MTGp SMG ANG MTL with electrocortical stimulation data reported previously (Simos
latency
et al., 2000a) in that: (i) there was no significant relation
Exception 771 467 410 580 420 500 between STGp activation onset and pronunciation latency for
(106) (170) (138) (118) (170) (140) exception words; (ii) there was a substantial correlation be-
Pseudohomophone 831 470 420 480 460 521
(115) (134) (120) (66) (190) (148)
tween the onset of activity in STGp and pronunciation latency
Pseudoword 853 450 409 600 400 560 for pseudowords and pseudohomophones; (iii) there was a
(118) (135) (83) (117) (180) (123) significant correlation between onset of MTGp activation and
STGp: posterior portion of the superior temporal gyrus, MTGp: posterior portion of the middle pronunciation latency of exception words; and (iv) there was no
temporal gyrus, SMG: supramarginal gyrus, ANG: angular gyrus, MTL: mesial temporal lobe significant relation between onset of MTGp activation and
regions (hippocampus and parahippocampal gyrus), BTC: basal temporal cortices (fusiform and pronunciation latency for pseudohomophones.
lingual gyrus).

Discussion
Table 2 The spatiotemporal activation profiles associated with reading
Pearson correlation coefficients between onset latency of activation in areas STGp, MTGp and aloud each of the three different types of print displayed a
MTL, and pronunciation latency number of common features: first, the regular progression of
activation from occipital to basal temporal areas within the
Left hemisphere STGp MTGp MTL
first 150–200 ms after stimulus onset; second, the subsequent
Exception –0.37 0.57* –0.40 ‘spread’ of activation to postero-lateral temporal regions; and
Pseudohomophone 0.61* –0.07 –0.35
third, the strong left hemisphere predominance in the degree of
Pseudoword 0.55* –0.21 –0.16
activation in both basal and lateral temporal regions. One should
STGp: posterior portion of the superior temporal gyrus, MTGp: posterior portion of the middle note that in all three tasks used in the present study subjects
temporal gyrus, MTL: mesial temporal lobe regions (hippocampus and parahippocampal gyrus).
were asked to produce a vocal response to each printed word
*P < 0.05.
stimulus. Therefore, the tasks involved speech as well as reading
(i.e. conversion of graphemic input into a phonological
participants, indicating a negligible relation between these representation). However, the close similarity between the
variables. activation maps obtained in the context of these tasks and those
obser ved in previous studies in our laborator y using silent
reading of real words as well as pseudowords with several
Relation between Naming Latency and Onset of Regional groups of non-impaired readers (ranging in age between 7 and
Activation 45 years) (Breier et al., 1998; 1999a; Simos et al., 2000b,c)
Onset latency of activation of a particular area was defined at the suggest that these profiles are specific to reading, regardless of
earliest latency (4 ms time bin), after stimulus onset, in which the particulars of the experimental task used to obtain the
the first of at least two consecutive activity sources was found activation profiles. Direct comparisons between silent and
in a particular area. The mean onset latency of the activation of reading aloud tasks, in the same group of subjects, are now
each area is also shown in Table 1 for comparison. The relation under way to examine subtle differences in the spatiotemporal
between onset latency of activity in each area and naming activation profiles that may ref lect the engagement of neuro-
latency was examined by computing the Pearson r correlation physiological processes specific to the vocal response and
coefficient separately for each type of letter string. As shown in unrelated to reading per se.
Table 2, the onset latency of activity in STGp accounted for a The stimuli used in the present study allowed us to identify
moderate proportion of the variability in naming latency for additional features of these profiles that are specific to two
pseudowords and pseudohomophones (R2 = 0.31, and R2 = 0.37, important attributes of print: (i) meaningfulness and (ii) depend-
respectively). In contrast, a significant proportion of the vari- ence on assembled phonology operations. Specifically, it is
ance in naming speed of exception words (R2 = 0.32) was generally assumed that meaningful stimuli are associated with
accounted for by the onset latency of activation in MTGp. Onset word-specific (lexical) mental representations. It has been
of activity in MTGp and naming latency for the other two types proposed that the mechanism responsible for pronouncing real
of letter strings were negatively correlated (see Fig. 3). Cor- words, especially those with rare print-to-sound correspond-
relations between naming latency and onset of activity in mesial ences, initially involves access to a lexical representation that
and basal temporal areas were either negligible or even negative. subsequently mediates the retrieval of the word’s name
Also negligible were correlations between the onset of activity in (Coltheart et al., 1993). Both exception words and pseudohomo-
homologous right hemisphere areas and pronunciation latencies. phones (by virtue of their phonological similarity to real words)
possess entries in the hypothetical ‘mental lexicon’, whereas
Onset of Activation Across Regions pseudowords do not. Based on this premise, we hypothesized
An ANOVA conducted on the onset of activation with three that the mechanisms involved in reading exception words and
within-subject variables [Type of Letter String (Exception, pseudohomophones would share at least one common com-
Pseudohomophone, Pseudoword), Area (STGp, MTGp, SMG, ponent, namely a process related to lexical access. This process
ANG, BTC and MTL) and Hemisphere (Left, Right)] revealed a would not be part of the mechanism for reading pseudowords.
significant Area main effect, F(4,60) = 44.75, P < 0.0001. Our findings were consistent with this notion, showing that a
Pairwise comparisons confirmed our initial observation (see prominent feature of the activation profile associated with
Table 1) that activity in basal temporal regions was detectable reading aloud both exception words and pseudohomophones
earlier than activity in all other temporal and inferior parietal involved the left middle temporal gyrus and mesial temporal

Cerebral Cortex Mar 2002, V 12 N 3 301

Figure 3. Regression plots that demonstrate the relation between onset latency in the left STGp and MTGp and pronunciation latency for each of the three types of letter strings used
in the study (n = 14).

regions. However, reading aloud pseudowords involved very logical decoding, but they are also orthographically unfamiliar.
sparse activity in these regions. Thus, MTGp shows reduced This feature is what makes their pronunciation unequivocally
neurophysiological activity as well as reduced regional cerebral dependent upon phonological decoding. In principle this
blood f low during pseudoword as compared to real word difference could account for the relation between onset latency
reading (Hagoort et al., 1999). Moreover, it appears that the left of activity in the left STGp and reading speed of these two
MTGp plays a special role in exception word reading. The types of letter strings. However, given that pseudohomophones
significant correlation between onset of activity in the left MTGp possess familiar phonological representations, this assumption
and naming latency indicated that the earlier the engagement would imply that the left STGp is primarily involved in visual/
of this area following word presentation, the faster the pro- orthographic processing of unfamiliar graphemic patterns. To
nunciation of the letter strings. The fact that MTGp activity did our knowledge, there is no evidence to support this claim.
not predict pronunciation speed for pseudohomophones In contrast to pseudowords and pseudohomophones, reading
suggests that engagement of this area may be a byproduct of aloud words that contain rare print-to-sound correspondences
phonological access achieved through the assembled route does not require these operations. In experienced readers,
for nonwords that sound like real words. Involvement of the pronunciation of exception words that occur with a relatively
MTGp in lexical/semantic analysis is suggested by several inde- high frequency in print is likely to be highly automatized and
pendent sources of evidence, including noninvasive functional depend little on assembled phonology. In a previous study
imaging investigations (Mummery et al., 1998; Hart et al., 2000; (Simos et al., 2000a), we obtained evidence that the brain
Kuperberg et al., 2000) and lesion studies (Damasio and mechanism for reading engages different areas depending upon
Damasio, 1989). the regularity (or frequency) of print-to-sound correspondences
Both pseudowords and pseudohomophones, on the other displayed by the words used. Specifically, we ascertained, using
hand, require assembled phonology operations. By definition, an invasive technique (electrocortical stimulation mapping), that
pseudohomophones and pseudowords not only require phono- the left STGp plays a crucial role in pseudoword pronunciation,

302 Imaging Word and Pseudoword Reading • Simos et al.

but may not be an indispensable component of the mechanism Automatic activation of STGp in tasks that do not require
responsible for reading aloud exception words (Simos et al., phonological decoding, but tap primarily into word recognition
2000a). However, inspection of the brain activation profiles processes, has been reported in other imaging studies as well
obtained from the same patients during reading of both regular (Hart et al., 2000). This possibility is consistent with the view
and exception words (using MSI) indicated that the left STGp that even words with rare sound–spelling correspondences
showed significant activation in all participants. In agreement (such as LAUGH) can be pronounced via a non-lexical route as
with this finding, activity in the left STGp was an invariable postulated by connectionist models of reading (Seidenberg and
feature of the activation profiles obtained in the present study McClelland, 1989). Finally, with respect to MTL, the results are
during reading of all three types of print. We did not observe consistent with findings from other imaging modalities and
reliable differences in either the degree or the onset latency of support the notion that activation of this area may occur even in
STGp activation across the three types of letter strings. Closer tasks that do not pose explicit demands for semantic analysis or
inspection of the data, however, indicated that the contribution even for encoding for subsequent retrieval (Martin et al., 1997).
of STGp in the brain mechanism of reading may change as a A final note is in order regarding the interpretation of timing
function of the presumed amount of sublexical phonological data. The source modeling approach adopted in this and our
processing required to pronounce items in each list. Thus, the previous investigations can only accommodate a maximum of
onset of activity in STGp was the only reliable predictor of two (usually only one) sources at each 4 ms bin per hemisphere.
naming speed for pseudowords and pseudohomophones, a Theoretically, several, simultaneously active, sources can be
finding consistent with the purported role of this area in distinguished on the basis of MSI data. The technique we
assembled phonology. In contrast, no significant relation routinely use in our lab is the standard method employed in
between the onset of MTGp activity and naming speed for these all clinical applications of MSI worldwide. Further, we have
items was found. Taken together our data demonstrate that, at ascertained the validity of this procedure in the context of a
least in experienced readers, the posterior portion of STGp, series of combined MSI–electrocortical stimulation studies
although routinely activated during reading of real words, is not (Papanicolaou et al., 1999; Simos et al., 1999, 2000a; Castillo et
an indispensable component of the mechanism for reading aloud al., 2001). Although it is in principle possible to differentiate
words that do not contain common print-to-sound corres- sources located at a smaller distance from each other, we do not
pondences. An alternative mechanism that could support access have cross-validation data that support this approach. Based on
to phonological representations for pronouncing real words may our combined MSI–electrocortical stimulation studies, we have
involve engagement of the middle temporal gyrus. ascertained that the location of clusters of activity sources
Data regarding the progression of activation across different represent cortical patches that play a crucial role in certain
regions provides information regarding the temporal course of component operations, such as phonological analysis. Taking
regional engagement in the experimental task that, in turn, these studies into account we can safely conclude that, at any
ref lects the functional connectivity among the areas that point in time, the activity sources that meet the criteria adopted
compose the mechanism of reading. A closer inspection of the in our source modeling procedure represent the most
spatiotemporal activation profiles obtained in the present study prominently active cortical patch in a given hemisphere.
reveal the following regarding regional interactions that may Using this method, we observed, in a given subject, clusters of
occur during reading. First, visual association areas located in temporally contiguous activity sources in a particular area
the basal surface of the left hemisphere appear to serve as an running for 8–100 ms at a time, which were preceded and
intermediate station between modality-specific visual cortices followed by source clusters in different areas. Accordingly, in the
(areas 17 and 18) and tertiary association regions in the temporal spatiotemporal maps constructed for each subject, anatomical
lobe. Second, posterior temporal, inferior parietal, and mesial areas within each hemisphere appeared to become active
temporal areas appear to become engaged either in parallel, or at sequentially. Collapsing spatiotemporal profiles across partici-
least in very close temporal proximity to each other, so that a pants is necessary in order to derive a more realistic repre-
consistent temporal succession of activation is not discernible. sentation of the temporal course of regional activation associated
Third, areas that may not be directly involved in the operations with a given task. The group spatiotemporal profiles displayed in
required for a particular function may nevertheless show Figure 4 were derived using this procedure. The observation
increased levels of activity during performance of this function. that the three main cortical areas discussed above (STGp, MTGp,
Three areas showed automatic activation although their engage- MTL) became active at one time or another during the same
ment did not appear to be a key component in the mechanism of 400 ms latency window (300–700 ms) may have two alternative,
reading a particular type of print: STGp during reading of but not mutually exhaustive, explanations: first, that these
exception words, MTGp during pseudohomophone reading, regions were all active simultaneously but activity in only one
and MTL during reading of exception words and pseudo- area at a time could be modeled in each participant; and second,
homophones. Assuming that MTGp is directly involved in that these regions became engaged strictly sequentially. It is
lexical/semantic access, engagement of this region during word difficult to distinguish between the two alternatives solely on the
reading may ref lect automatic semantic analysis of meaningful basis of functional imaging data. Methods that rely on measures
stimuli. Regional increases in blood f low/metabolism in this of regional blood f low or metabolism simply lack the temporal
region, associated with implicit word processing, have also been resolution necessary to monitor neurophysiological activity in
found in Positron Emission Tomography (PET) and fMRI studies real time. MSI, on the other hand, possesses adequate temporal
(Price et al., 1996; Hart et al., 2000). Automatic engagement resolution, but it is currently limited by validity considerations,
of semantic representations is consistent with parallel and to the use of source modeling techniques that permit reliable
distributed models of reading (Seidenberg and McClelland, identification of no more than two simultaneously active cortical
1989). Further, assuming that STGp is directly involved in patches. One way to resolve this issue would be to combine data
assembled phonology, then activation of this region during from MSI and electrocortical stimulation in the same patients.
exception word reading may ref lect phonological decoding This approach is generally very promising as an adjunct to
that occurs in parallel with addressed phonology operations. any non-invasive functional imaging method, but is difficult to

Cerebral Cortex Mar 2002, V 12 N 3 303

 the ability to read different types of print. This approach meets a
number of critical requirements for any investigation of the brain
mechanisms of complex cognitive or linguistic functions. First, it
utilizes a noninvasive functional imaging technique (MSI) that
has the capacity to provide reliable images of the working brain
of individual subjects. Secondly, this technique captures critical
aspects of brain activation (i.e. neuronal signaling, as opposed to
secondary delayed effects of neuronal activity such as regional
blood f low or metabolism). Thirdly, the functional significance
of activated brain areas is verified in patients undergoing
functional mapping using invasive techniques (Simos et al.,
2000a), thereby ascertaining the external validity of activation
profiles obtained non-invasively through MSI. Fourthly, MSI
captures the spatial as well as the temporal features of regional
activation in real time.
As discussed above, MSI (like any other functional imaging
method) may fail to detect certain details of the activation
profiles, it appears to be capable of capturing the essential
features of this profile, as indicated by the results of direct
comparisons with invasive mapping techniques (Breier et al.,
1999b; Simos et al., 1999, 2000a). A comparison of the results
presented here with those from functional imaging studies that
use measures of cerebral blood f low or metabolism reveals many
similarities, but some notable differences as well. The present
data are consistent with reports of increased activation in MTG
during reading of real words, implicating this region in
whole-word (i.e. lexical) processing (Price et al., 1996; Hart et
al., 2000) and with reports of automatic activation of STGp even
in tasks that do not require addressed phonology (Hart et al.,
2000). Activation in the left frontal operculum, which has been
found in several PET and fMRI studies (Herbster et al., 1997;
Hagoort et al., 1999; Fiez et al., 1999; Pugh et al., 1996), was not
as consistent in the present study as activation in temporal
and temporoparietal regions. This discrepancy may ref lect a
peculiarity of the functional imaging modality used in the
present study: in our experience MSI appears to be more
sensitive to neurophysiological activity produced in the
temporal (including mesial temporal), parietal and occipital
lobes than in the frontal lobe. When detected, prefrontal
activity showed the expected left hemisphere lateralization and
modulation by task demands, i.e. increased activity in tasks that
require phonological decoding, in agreement with previous
Figure 4. Temporal course of activation in each of the three brain regions (left
hemisphere) that showed significant print-type related effects and in basal temporal reports (Pugh et al., 1996; Fiez and Petersen, 1998; Breier et al.,
areas for the group of 16 participants. Note the clear temporal distinction between 1999a; Hagoort et al., 1999).
activation of basal temporal and all other areas, and the significant overlap in the course Future studies should examine more closely the role of
of activation among the latter. stimulus (such as degree of regularity, consistency, and relative
frequency) and subject variables (such as age and reading skill)
on the degree and timing of neuronal activity in the brain
implement, mostly due to time constraints and patient safety
regions identified in this report. A mong other issues, such
considerations associated with direct cortical stimulation
studies would address, in a more systematic manner, the neuro-
studies. In this approach, MSI data could be used to identify
logical validity of theories postulating single versus dual
cortical patches that appear to be essential for the performance
mechanisms for reading words.
of a particular experimental task. Subsequently, during the
electrocortical stimulation study, MSI-derived cortical patches
may be stimulated with brief pulses delivered at different delays
after stimulus onset to determine the critical time window Notes
We wish to thank three anonymous reviewers for their constructive
that stimulation of a particular cortical region disrupts task
comments and suggestions that led to significant improvements over
performance.
earlier versions of the manuscript. This work was supported in part
To summarize, the present investigation integrates three types by NSF Grant REC-9979968 and NIH Grant NS37941-01 to A.C.
of evidence to provide unique insights into the role of various Papanicolaou.
temporal lobe regions in reading, namely information on: (i) the Address correspondence to Panagiotis G. Simos, Department of
degree; (ii) the temporal course of the engagement of various Neurosurger y, The University of Texas—Houston Medical School,
brain areas during tasks that exemplify word reading; and (iii) 1133 MD Anderson Blvd, Suite 304, Houston, TX 77030, USA. Email:
the effects of transient interference with one of these regions on [email protected].

304 Imaging Word and Pseudoword Reading • Simos et al.

References neuroanatomy of the semantic system: divisible by what? J Cogn
Neurosci 10:766–777.
Breier JI, Simos PG, Zouridakis G, Papanicolaou AC (1998) Relative timing
of cortical activation during a word recognition task. J Clin Exp Papanicolaou AC, Simos PG, Breier JI, Zouridakis G, Willmore LJ,
Neuropsychol 20:782–790. Wheless JW, Constantinou JE, Maggio W W, Gormley WB (1999)
Magnetoencephalographic mapping of the language-specific cortex.
Breier JI, Simos PG, Zouridakis G, Papanicolaou AC (1999a) Temporal
J Neurosurg 90:85–93.
course of regional activation associated with phonological decoding.
J Clin Exp Neuropsychol 21:465–476. Papanicolaou AC, Simos PG, Breier JI, Fletcher JM, Foorman BR, Francis
Breier JI, Simos PG, Papanicolaou AC, Zouridakis G, Wilmore LJ, Wheless DJ, Castillo EM, Davis R (2001) Brain mechanisms for reading in
JW, Constantinou JC, Maggio W W (1999b) Cerebral laterality children with and without dyslexia: a review of studies of normal
assessment using magnetoencephalography: a comparison with the development and plasticity. J Child Neurol 16:241–251.
Intracarotid Amobarbital Procedure. Neurology 22:938–945. Price CJ, Wise RJ, Frackowiak RS (1996) Demonstrating the implicit
Breier JI, Simos PG, Wheless JW, Constantinou JEC, Papanicolaou AC processing of visually presented words and pseudowords. Cereb
(2001) Hemispheric language dominance in children determined by Cortex 6:62–70.
magnetic source imaging. J Child Neurol 16:124–130. Pugh KR, Shaywitz BA, Shaywitz SE, Constable RT, Skudlarski P, Fulbright
Castillo EM, Simos PG, Venkataraman, V, Breier JI, Papanicolaou AC RK, Bronen R A, Shankweiler DP, Katz L, Fletcher JM, Gore JC (1996)
(2001) Mapping of expressive language cortex using Magnetic Source Cerebral organization of component processes in reading. Brain
Imaging. Neurocase 7:419–422. 119:1221–1238.
Coltheart M, Curtis B, Arkins P, Haller M (1993) Models of reading aloud: Pugh KR, Shaywitz BA, Shaywitz SE, Shankweiler DP, Katz L, Fletcher JM,
dual-route and parallel-distributed-processing approaches. Psychol Skudlarski P, Fulbright RK, Constable RT, Bronen R A, Lacadie C, Gore
Rev 100:589–608. JC (1997) Predicting reading performance from neuroimaging
Damasio H (1995) Human brain anatomy in computerized images. New profiles: the cerebral basis of phonological effects in printed word
York: Oxford University Press. identification. J Exp Psychol 23:299–318.
Damasio H, Damasio A (1989) Lesion analysis in neuropsychology. New Rumsey JM, Horwitz B, Donohue BC, Nace K, Maisog JM, Andreason P
York: Oxford University Press. (1997) Phonological and orthographic components of word reading: a
Fiez JA, Petersen SE (1998) Neuroimaging studies of words reading. Proc PET–rCBF study. Brain 120:739–759.
Natl Acad Sci USA 95:914–921. Sarvas J (1987) Basic mathematical and electromagnetic concepts of the
Fiez JA, Balota DA, Raichle ME, Petersen SE (1999) Effects of lexicality, biomagnetic problem. Phys Med Biol 32:11–22.
frequency, and spelling-to-sound consistency on the functional Seidenberg MS, McClelland JL (1989) A distributed, developmental model
anatomy of reading. Neuron 24:205–218. of word recognition and naming. Psychol Rev 96:523–568.
Glushko RJ (1979) The organization and activation of orthographic Shallice T, Warrington EK (1980) Single and multiple component central
knowledge in reading aloud. J Exp Psychol 5:674–691. dyslexic syndromes. In: Deep dyslexia (Coltheart M, Patterson K,
Haggort P, Indefrey P, Brown C, Herzog H, Steinmetz H, Seitz RJ (1999) Marshall JC, eds). London: Routledge and Kegan Paul.
The neural circuitry involved in the reading of German words and Simos PG, Breier JI, Zouridakis G, Papanicolaou AC (1998) Assessment of
pseudowords: a PET study. J Cogn Neurosci 11:383–398. cerebral dominance for language using magnetoencephalography.
Hart J, Kraut MA, Kremen S, Soher B, Gordon B (2000) Neural substrates J Clin Neurophysiol 15:364–372.
of orthographic lexical access as demonstrated by functional brain Simos PG, Papanicolaou AC, Breier JI, Wilmore LJ, Wheless JW,
imaging. Neuropsychiatry Neuropsychol Behav Neurol 13:1–7. Constantinou JC, Gormley W, Maggio W W (1999) Localization of
Herbster AN, Mintum MA, Nebes RD, Becker JT (1997) Regional cerebral language-specific cortex using MEG and intraoperative stimulation
blood f low during word and nonword reading. Human Brain Mapp mapping. J Neurosurg 91:787–796.
5:84–92. Simos PG, Breier JI, Wheless JW, Maggio W W, Fletcher JM, Castillo EM,
Kuperberg GR, McGuire PK, Bullmore ET, Brammer MJ, Rabe-Hesketh S, Papanicolaou AC (2000a) Brain mechanisms for reading: the role
Wright IC, Lythgoe DJ, Williams SC, David AS (2000) Common and of the superior temporal gyrus in word and pseudoword naming.
distinct neural substrates for pragmatic, semantic, and syntactic NeuroReport 11:2443–2447.
processing of spoken sentences: an fMRI study. J Cogn Neurosci Simos PG, Breier JI, Fletcher JM, Bergman E, Papanicolaou AC (2000b)
12:321–341. Cerebral mechanisms involved in word reading in dyslexic children: a
Marshal JC, Newcombe F (1973) Patterns of paralexia: a psycholinguistic magnetic source imaging approach. Cereb Cortex 10:809–816.
approach. J Psycholinguist Res 2:175–199. Simos PG, Papanicolaou AC, Breier JI, Fletcher JM, Foorman BR, Bergman
Martin A, Wiggs CL, Weisberg J (1997) Modulation of human medial E, Fishbeck K, Papanicolaou AC (2000c) Brain activation profiles in
temporal lobe activity by form, meaning, and experience. dyslexic children during nonword reading: a magnetic source imaging
Hippocampus 7:587–593. study. Neurosci Lett 290:61–65.
McCann RS, Besner D (1987) Reading pseudohomophones: implications Simos PG, Breier JI, Fletcher JM, Foorman BR, Mouzaki A, Papanicolaou
for models of pronunciation assembly and the locus of word- AC (2001) Age-related changes in regional brain activation during
frequency effects in naming. J Exp Psychol 13:14–24. phonological decoding and printed word recognition. Dev
Mummer y CJ, Patterson K, Hodges JR, Price CJ (1998) Functional Neuropsychol 19:191–210.

Cerebral Cortex Mar 2002, V 12 N 3 305