You are on page 1of 10

Neuropsychologia 46 (2008) 409–418

Categorical and thematic knowledge representation in the brain: Neural


correlates of taxonomic and thematic conceptual relations
Olga Sachs a,∗ , Susanne Weis b , Timo Krings c , Walter Huber b , Tilo Kircher a
a Department of Psychiatry and Psychotherapy, RWTH Aachen University, Pauwelstr. 30, D-52074 Aachen, Germany
b Department of Neurology, RWTH Aachen University, Pauwelstr. 30, D-52074 Aachen, Germany
c Department of Neuroradiology, RWTH Aachen University, Pauwelstr. 30, D-52074 Aachen, Germany

Received 24 April 2007; received in revised form 8 August 2007; accepted 18 August 2007
Available online 25 August 2007

Abstract
Most current models of knowledge organization in the brain are based on hierarchical or taxonomic categories (animals, tools). Another important
organizational pattern is thematic categorization, i.e. categories held together by external relations, a unifying scene or event (car and garage).
We used fMRI to examine neural activation patterns as subjects performed a category construction task where these two category types were
contrasted. Subjects were visually presented with a target word followed by the presentation of two match words and had to choose by button press
one match that goes best with the target word. In the balanced or cross-categorization condition (Car/Garage Bus) both match words fit the target;
in the biased conditions only one match word fit the target either thematically (Car/Garage Brush) or taxonomically (Car/Bus Eraser).
We found that in the biased conditions, thematic and taxonomic categories recruited very similar cortical regions: left inferior frontal, middle
temporal and occipital regions. In the balanced condition subjects showed no behavioral preference for either thematic or taxonomic matches.
However, contrasting signal changes during a subjective taxonomic choice in the presence of a thematic alternative vs. a subjective thematic choice
in the presence of a taxonomic alternative required the additional recruitment of right middle frontal gyrus, left precuneus and left thalamus.
Our results suggest that thematic relations between objects are processed similarly to taxonomic relations, but require less cerebral processing
demand, providing validation for thematic categories as an alternative principle of conceptual organization.
© 2007 Elsevier Ltd. All rights reserved.

Keywords: Categorization; Thematic categories; Taxonomic categories; fMRI; Cross-categorization

How knowledge is represented and stored in the brain is still brain areas involved in the processing of isolated taxonomic
an open question in cognitive neuroscience. The basic princi- categories, such as animals or tools. The intuitive idea that
ple of organizing semantic information is categorization. Two distinct conceptual categories should also have their distinct
category types can be distinguished: (1) taxonomic categories domain-specific localization in the brain (living vs. non-living
that include items of the same kind (e.g., superordinate category things, animals vs. tools) has gained support from the studies
vehicles with members such as car, bus, train, etc.) (2) the- of brain-damaged patients who have category-specific deficits
matic categories that include externally related items interacting (e.g., Caramazza & Shelton, 1998; Warrington & McCarthy,
within scenes or events (e.g., car and garage, dog and leash).1 1983) and functional imaging studies of healthy subjects that
Most lesion and neuroimaging studies of knowledge rep- showed distinct areas of activation for natural kinds and artifacts
resentation and categorization have focused on identifying (e.g., Perani, Schnur, et al., 1999; Thompson-Schill, Aguirre,
D’Esposito, & Farah, 1999). For example, Perani, Schnur, et al.
(1999) have found that the left fusiform gyrus was involved in
∗ Corresponding author. Tel.: +49 241 8089721; fax: +49 241 8082422. the processing of pictures and words of living entities and the
E-mail address: osachs@ukaachen.de (O. Sachs). left middle temporal gyrus was involved in the processing of
1 Members of taxonomic categories are often referred to in the literature as
words and pictures of tools. Leube, Erb, Grodd, Bartels, and
“categorically” or “semantically” related concepts. Members of thematic cate-
gories are often said to share an “associative” relationship and therefore referred
Kircher (2001) have even found a hemispheric differentiation
to as “semantic associates” or “associatively” related items. In this paper we between living and non-living entities: right frontal, temporal
predominantly use the terms “taxonomic” and “thematic” categories. and fusiform areas were activated when subjects judged words

0028-3932/$ – see front matter © 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.neuropsychologia.2007.08.015
410 O. Sachs et al. / Neuropsychologia 46 (2008) 409–418

as belonging to the “living” category, whereas the processing (2001) suggest that besides sharing a functional relationship,
of the “non-living” category members was largely limited to thematic categories can often share other types of rela-
the left frontal and occipital areas. Several hypotheses were tions, e.g., spatial (roof–house), temporal (meal–bill) or causal
proposed to account for the discovered anatomical segregation (electricity–light). However, in most of such cases a functional
of natural kinds and artifacts, ranging from the evolutionary relation remains strong also in the presence of other types of
pressure to distinguish between living and non-living things to relations.
the role of perceptual and functional properties in distinguish- The neural correlates of thematic categories are little under-
ing individual members of these categories. For example, the stood, although their importance and salience has been shown in
sensory-motor account of conceptual representation suggests many different areas of cognitive psychology. Cognitive devel-
that perceptual features play a greater role in the processing opment research suggests that children often show preference
of natural kinds, while functional features are more important for thematic categories (Nguen & Murphy, 2003; Smiley &
for artifacts. This difference is reflected in distinct brain areas Brown, 1979). Anthropological research and studies of cate-
involved in the representation of natural kinds and artifacts found gorization and inductive reasoning among expert groups, such
in neuroimaging studies (Cappa, Perani, Schnur, Tettamanti, & as tree experts or commercial fishermen, indicate that people
Fazio, 1998; Damasio, Grabowski, Tranel, Hichwa, & Damasio, often make knowledge-based thematic groupings that are not
1996; Martin, Wiggs, Ungerleider, & Haxby, 1996). based exclusively on similarity, but also on causal relations
The role of functional information in the conceptual represen- between entities (Lopez, Atran, Coley, Medin, & Smith, 1997;
tation of artifacts or “the manner in which an object interacts with Medin, Lynch, Coley, & Atran, 1997; Shafto & Coley, 2003).
the environment and with the observer” is also emphasized in the Finally, even the most typical research subjects, U.S. college
distributed conceptual structure theory of categorical represen- students, make thematic categories and use them to make induc-
tation (Tyler, Moss, Durrant-Peatfield, & Levy, 2000). Here it is tive inferences (Lin & Murphy, 2001; Ross & Murphy, 1999).
suggested that we know what something is only when we know For example, Lin and Murphy (2001) have demonstrated that
what it does and that the strong relationship between appear- when given both a taxonomic and a thematic match (e.g. lion
ance and function is particularly important for the conceptual and litter box) and asked which one of them forms a better cate-
representation of artifacts. Tyler et al.’s theoretical claim was gory with cat, young educated adults performed thematic (litter
supported in a later fMRI study where they found that objects and box–cat) rather than taxonomic categorization (lion–cat) most
their associated actions activated the same neural regions—left of the time. These results held even when subjects were given
fusiform gyrus, superior and middle temporal cortex (Tyler et a taxonomic, similarity-based definition of a category as part of
al., 2003). An earlier study conducted by Grabowski, Damasio, the instructions for constructing categories. This shows that they
and Damasio (1998) has also found a common area of activa- formed thematic categories fully aware of the fact that they were
tion for the retrieval of words denoting objects and words for supposed to form categories and not to sort objects according
actions. These findings extend the traditional understanding of to some other principle. This set of results also highlights the
what a category is by showing that categories are held together increasing recognition in the current behavioral categorization
not just by the internal correlations between their features, but research that items can be cross-classified into more than one
also by the external functional relationship between an object category type (Lin & Murphy, 2001; Nguen & Murphy, 2003;
and its environment. Ross & Murphy, 1999). Besides taxonomic categories (e.g. ani-
We would like to take this idea a step further and investigate mals, plants, vehicles), people use a variety of other category
the neural representation of categories that are held together types that may include the same items, such as thematic cate-
by strong external or complementary relations between objects gories (e.g. dog and leash, car and garage, picture and frame),
that co-occur or interact together in space and time (Lin & as well as ad hoc or goal-derived categories (e.g. things to carry
Murphy, 2001). These categories are called thematic. They can out of the burning house; Barsalou, 1983), or script categories
be formed when thematically related items, such as car and (e.g. breakfast foods; Ross & Murphy, 1999).
garage, are grouped together in a conceptual sorting situation. However, it is important to note that classical taxonomic
Unlike members of traditional similarity-based taxonomic cat- categories that combine “things of the same kind” differ from
egories that are usually organized in hierarchies or taxonomies thematic categories in many important ways. One crucial differ-
(e.g., artifacts → vehicles → cars → BMW), members of the- ence lies in how they are labeled: we use the same word to refer to
matic categories do not share any perceptual features. However, members of taxonomic categories and not to things that co-occur
despite the absence of any perceptual similarity, members of together. One way to interpret the differences between thematic
thematic categories share a very strong and salient functional and taxonomic conceptual links is to only call things of the
relationship. Members of taxonomic categories may also be same kind “categories”, and to refer to items that are related the-
functionally related: for example, train and car are both used to matically as “semantic associates” (accepted terminology in the
drive. However, their category membership is based on their per- neuroimaging literature). Another possibility is to treat thematic
ceptual or functional similarity, whereas members of thematic and taxonomic relations as two different, perhaps independent,
categories do not share any functional similarity, but comple- but equally important ways of organizing semantic knowledge
ment each other through their function. In the absence of any into groups that both can be called “categories” (more com-
perceptual or functional similarity such functional relationship mon in the cognitive literature). We use the terminology of the
is, in fact, the only connection that they have. Lin and Murphy cognitive literature because thematic relations have been inves-
O. Sachs et al. / Neuropsychologia 46 (2008) 409–418 411

tigated there much more extensively than in the neuroimaging set on the evaluation of taxonomic categories when they compete
categorization research. with the thematic categories.
Relatively good predictions for our study can be derived
from a large body of semantic priming studies that investigated 1. Materials and methods
the structure of the lexical–semantic network by contrasting
the processing of categorical relations between prime target 1.1. Participants
pairs (what we refer to as taxonomic relations between cate-
Fourteen healthy male subjects participated in the study. All of them were
gory members) with associative relations between prime target right-handed according to the Edinburgh Inventory of Handedness (Annett,
pairs (what we refer to as thematic relations). For example, 1970). Their mean age was 28 (S.D. 7.5) years and they all spoke German as
Kotz, Cappa, von Cramon, and Friederici (2002) have compared their native language. All participants had normal or corrected-to-normal vision,
the neuroanatomical correlates of categorical and associative gave informed consent and were paid 10 Euros for participation in the study.
The study was approved by the Local Ethics Committee.
relations between prime target pairs using an auditory primed
lexical decision task. They found that compared to the asso-
1.2. Stimuli and design
ciative relations (e.g., key–chain), the processing of categorical
relations (e.g. cow–dog) recruited additional brain areas in the Subjects performed a forced-choice category construction task that is com-
right precuneus, the isthmus gyrus cinguli and the cuneus, monly used in categorization research (e.g., Lin & Murphy, 2001) and in
thus suggesting a more effortful processing of categorical rela- developmental classification studies (e.g., Smiley & Brown, 1979). In this task
subjects are first presented with a target word in the center of the computer
tions in the right hemisphere. Evidence from a number of
screen (e.g., Car), followed by the simultaneous presentation of two matches
other semantic priming studies that used either visual fields (e.g., Garage, Bus). Subjects then have to decide which one of the two matches
or ERPs to establish the hemispheric difference in the prim- goes best with the first target word to form a category. We used a categorical
ing of different semantic relationships (Beeman, Friedman, design where all subjects were tested in three main experimental conditions
Grafman, & Perez, 1994; Chiarello & Richards, 1992) also with 30 trials in each condition: (1) balanced condition where both matches
fit the target—Car/Garage Bus (one thematically, e.g., Car–Garage, and one
points toward the right-hemisphere advantage for the processing
taxonomically, e.g., Car–Bus); (2) thematically biased condition where only
of categorical or taxonomic relations between concepts. Finally, one match fits the target thematically—Car/Garage Brush; (3) taxonomically
a lesion study by Hagoort, Brown, & Swaab, 1996 showed biased condition where only one match fits the target taxonomically—Car/Bus
that patients with right hemisphere lesions produced a nor- Eraser. The balanced condition was used for addressing the cross-categorization
mal N400 priming effect while listening to associatively related question and the two biased conditions were used to investigate thematic and
taxonomic categories in isolation.
prime-target pairs (bread–butter), but a trend toward a reduced
For each of the 30 German target words a thematic and a taxonomic match
N400 priming effect for categorically related prime-target pairs were found that formed either a thematic or a taxonomic relation with the tar-
(church–villa). get. A pretest was conducted on a separate group of subjects to ensure that
In the current study our goal is to (1) compare the neural corre- thematic and taxonomic matches had an equally strong and salient relationship
lates of thematic and taxonomic categories and (2) to investigate with a target. Fifteen pretest participants were asked to rate the relationship
between potential thematic matches and their targets on a scale from 1 to 7
the issue of cross-categorization: what happens when the same
based on the definition of thematic categories in Lin and Murphy (2001), i.e.
object is an equally good member of both a thematic and a tax- how well they form an external or complementary relationship and co-occur
onomic category. When we combine both lines of research that and interact together in space and time. A definition of taxonomic categories
were discussed above—neuroimaging categorization research was used to rate the relationship between potential taxonomic matches and their
and semantic priming studies, we can develop fairly specific pre- targets, i.e. how similar they are in their perceptual, biological or functional
properties. All taxonomic matches were related to the target on the superor-
dictions regarding our research questions. Based on the results
dinate level (e.g., bus and car are both vehicles). Based on the results of this
of neuroimaging categorization studies that involved non-living pretest, we selected only those thematic and taxonomic matches that scored
objects or artifacts (Perani, Schnur, et al., 1999) we hypothe- 5 or higher. The selected thematic and taxonomic matches did not differ in
size that both thematic and taxonomic categories made up of their average rating (Mthem = 6.69, S.D. = 0.32; Mtax = 6.60, S.D. = 0.49, t-test,
such concepts should activate similar brain areas in the left mid- p = 0.13). The unrelated matches for the biased condition scored 2 or less on the
same test (Munr = 1.35, S.D. = 0.52). All selected stimuli belonged to the same
dle temporal gyrus, as well as in the left frontal and occipital
overall conceptual domain (all words depicted only artifacts), were concrete
areas. However, differences in the processing of taxonomic and and imageable (all words were concrete objects that are easy to visualize). The
thematic relations between concepts found in semantic prim- CELEX lexical database (Baayen, Piepenbrock, & van Rijn, 1993) was used
ing studies (Beeman et al., 1994; Chiarello & Richards, 1992; to control for the average word frequency of German words used in taxonomic
Kotz et al., 2002) and in a lesion study by Hagoort et al. (1996) and thematic trials (Mthem = 34.53, S.D. = 62.06; Mtax = 23.07, S.D. = 46.91, t-
test, p = 0.13). All stimuli were also controlled for word length (Mthem = 5.80,
suggest that we may also predict that taxonomic categories will
S.D. = 1.24; Mtax = 5.63, S.D. = 1.30, t-test, p = 0.31). See Table 1 for examples
recruit additional right-hemispheric areas since they may require of the stimuli used in the experiment.
a more effortful processing than thematic categories.
Besides investigating taxonomic and thematic categories in 1.3. Procedure
isolation, our design allows us most importantly to distinguish
between cerebral activation during the subjective classification An event-related design was used to present balanced, thematically biased,
into one or the other category (cross-categorization). We hypoth- taxonomically biased and baseline trials in a randomized order to participants.
Four randomized versions of the experiment were counterbalanced across sub-
esize that when thematic and taxonomic categories compete, jects to avoid a systematic effect of conditions on each other. Presentation of
there will be a greater additional activation during subjective tax- stimuli was controlled by a computer using the Presentation 10.1 software pack-
onomic vs. thematic categorization reflecting increased demands age (Neurobehavioral Systems, http://www.neurobs.com/). Subjects lying in the
412 O. Sachs et al. / Neuropsychologia 46 (2008) 409–418

Table 1 for their different acquisition times. Volumes were then normalized into standard
Sample stimuli used in the experiment stereotaxic anatomical MNI-space by using the transformation matrix calculated
from the first EPI-scan of each subject and the EPI-template. The default settings
Target word Thematic match Taxonomic match
for normalization in SPM2 with 16 non-linear iterations and the standard EPI-
German (English German (English German (English
template supplied with SPM2 were used. Afterwards, the normalized data with
translation) translation) translation)
a resliced voxel size of 4 mm × 4 mm × 4 mm were smoothed with a 10 mm
Auto (Car) Garage (Garage) Bus (Bus) FWHM isotropic Gaussian kernel to accommodate intersubject variation in brain
Boot (Boat) Segel (Sail) Schiff (Ship) anatomy. The time series data were high-pass filtered with a high-pass cut-off
Faden (Thread) Nadel (Needle) Seil (Rope) of 1/128 Hz. The first-order autocorrelations of the data were estimated and
Topf (Pot) Herd (Stove) Kessel (Kettle) corrected for.
For each subject individually, responses during the balanced conditions were
sorted into taxonomic matches and thematic matches. These individual data were
scanner viewed the computer screen through MR-compatible video goggles used for modeling the expected hemodynamic response at stimulus onset for each
(VisuaStim XGA, Resonance Technology, Inc., http://www.mrivideo.com/). event-type (taxonomic match in the balanced condition, thematic match in the
Each individual trial started with a fixation cross in the middle of the computer balanced condition, taxonomic match in the bias condition, thematic match in
screen presented for 500 ms followed by the presentation of the target word that the biased condition, as well as baseline trials). The hemodynamic response was
lasted 1000 ms. As soon as the target word disappeared two match words were modeled by two response functions, a canonical hemodynamic response function
presented in the middle of the computer screen for 2000 ms. Match words were (HRF; Friston et al., 1998) and its temporal derivative. The temporal derivative
presented one above the other to avoid saccadic eye movements. The presenta- was included in the model to account for the residual variance resulting from
tion of match words was immediately followed by the # sign in the center of small temporal differences in the onset of the hemodynamic response, which is
the screen that lasted 2500–7500 ms to create jitter between individual trials. not explained by the canonical HRF alone. The functions were convolved with
Therefore, the average ISI was 5000 ms. Subjects could make their response at the event-train of stimulus onsets to create covariates in a general linear model.
any time point after the presentation of the two matches and before the next trial Subsequently, parameter estimates of the HRF regressor for each of the different
began. They were instructed to press the top button of the parallel port device conditions were calculated from the least mean squares fit of the model to the
when the top word was a better match to the target and to press the bottom time series. Parameter estimates for the temporal derivative were not further
button when the bottom word was a better match. The pressing was done with considered in any contrast.
the middle and index finger of the left hand to avoid the motor-related activation An SPM2 random-effects group analysis was performed by entering param-
in the left hemisphere. The upper or lower position of thematic and taxonomic eter estimates for all conditions into a within-subject one-way ANOVA. We
matches in the balanced condition and the position of the related match in both expected the differences between processing of taxonomic as opposed to the-
biased conditions were counterbalanced across trials. matic categories to be quite small. Still, we were interested in studying effects
A baseline task in this experiment was a letter categorization task used in across the whole brain instead of restricting the search volume to pre-defined
other categorization studies because it shares the same stimulus and response regions of interest. Therefore, we chose to employ Monte-Carlo simulation of
characteristics as the main categorization task, but has no semantic component the brain volume to establish an appropriate voxel contiguity threshold (Slotnick,
(Devlin et al., 2002; Tyler et al., 2003). Participants were first presented with a Moo, Segal, & Hart, 2003). This correction has the advantage of higher sensitiv-
target letter string (e.g., Ccc), followed by the simultaneous presentation of two ity to smaller effect sizes, while still correcting for multiple comparisons across
matches (e.g., Gggggg, Cccc). Subjects then had to decide which one of the two the whole brain volume. Assuming an individual voxel type I error of p < 0.001,
matches goes best with the target letter string. All letter strings were matched to a cluster extent of 10 contiguous resampled voxels was indicated as neces-
word stimuli in letter number. sary to correct for multiple voxel comparisons at p < 0.05. The reported voxel
coordinates of activation peaks were transformed from MNI space to Talairach
and Tournoux (1988) atlas space by non-linear transformations (www.mrc-
1.4. MRI acquisition cbu.cam.ac.uk/Imaging/mnispace.html). This was done to be able to use the
Talairach and Tournoux (1988) atlas to identify the anatomical brain region for
All scanning was performed on a 1.5 T scanner (Gyroscan Intera, Philips the activation peaks.
Medical Systems, Best, The Netherlands) using standard gradients and a cir-
cular polarized phase array head coil. For each subject, we acquired two series
of functional volumes of T2*-weighted axial EPI-scans including five initial 2. Results
dummy scans parallel to the AC/PC line with the following parameters: number
of slices (NS) = 31; slice thickness (ST) = 3.5 mm; interslice gap (IG) = 0.35 mm; 2.1. Behavioral data
matrix size (MS) = 64 × 64; field of view (FOV) = 240 mm × 240 mm; voxel
size = 3.75 mm × 3.75 mm; echo time (TE) = 50 ms; flip angle = 90; repetition
2.1.1. Choice data
time (TR) = 2.8 s. After the functional run, for each participant an anatomical
scan was acquired using a high-resolution T1-weighted 3D-sequence consist- In biased conditions where only one of the two matches could
ing of 140 transveral slices (TR = 7776 ms; TE = 3.56 ms; FoV = 256 mm; slice form a category with the target, subjects made this choice cor-
thickness = 1 mm; inter slice gap = 1 mm, flip angle = 8). Three hundred twenty rectly 97% (S.D. = 3.8) of the time for the thematically biased
two whole head scans were performed per subject divided into two runs. Each run condition and 95% (S.D. = 4.3) of the time for the taxonomi-
contained 158–166 whole head scans depending on the randomization version
cally biased condition with no significant difference between
the subject received.
the two (p = 0.09), as revealed by a t-test (t-tests were also
used for other comparisons in the behavioral data section). Only
1.5. fMRI data analysis
correct choices were included in the analysis of the imaging
MR images were analyzed using Statistical Parametric Mapping software data. In the balanced condition where each one of the matches
(SPM2; www.fil.ion.ucl.ac.uk) implemented in MATLAB 6.5 (Mathworks Inc., could form either a thematic or a taxonomic category with
Sherborn, MA). the target, subjects made a thematic choice just as often as
After discarding the first five volumes, all images were realigned to the they made a taxonomic choice (Mthem = 50.02%, S.D. = 10.19;
first image to correct for head movement. Unwarping was used to correct for
the interaction of susceptibility artifacts and head movement. After realignment Mtax = 49.98%, S.D. = 10.19, p = 0.49). All but three subjects
and unwarping, the signal measured in each slice was shifted relative to the have shown this choice pattern (within one standard deviation
acquisition time of the middle slice using a sinc interpolation in time to correct above and below 50%), two out of the three remaining sub-
O. Sachs et al. / Neuropsychologia 46 (2008) 409–418 413

Table 2
Activation peaks for the biased conditions with their localization
Anatomical region BA Coordinates z-Value No. voxels

x y z

Thematically biased condition > baseline


Inferior frontal gyrus (L)* 47, 9 −51 [−52] 21 [20] 25 [28] 6.26 275
Middle temporal gyrus (L)* 21 −59 [−60] −35 [−36] 2 [0] 5.04 88
Cuneus (L) 18 −8 [−8] −84 [−88] 23 [20] 3.85 59
Angular gyrus (L)* 39 −32 [−32] −56 [−60] 36 [36] 4.42 40
Precentral gyrus (L) 6 −48 [−48] −2 [−4] 44 [48] 4.69 29
Parahippocampal gyrus (L)* 18 −12 [−12] −47 [−48] 2 [0] 3.94 22
Cuneus (R)* 18 16 [16] −97 [−100] 9 [4] 4.00 21
Medial frontal gyrus (L)* 6 0 [0] −8 [−12] 63 [68] 3.94 15
Lingual gyrus (R)* 18 12 [12] −50 [−52] 3 [100] 3.61 11
Taxonomically biased condition > baseline
Inferior frontal gyrus (L)* 47, 45 −51 [−52] 20 [20] 21 [24] 6.94 400
Angular gyrus (L)* 39, 7 −32 [−32] −56 [−60] 36 [36] 4.41 59
Medial frontal gyrus (L)* 6 0 [0] 6 [4] 51 [56] 3.91 55
Cuneus (R)* 18 20 [20] −97 [−100] 9 [4] 4.91 40
Parahippocampal gyrus (L)* 19 −12 [−12] −47 [−48] 2 [0] 4.43 38
Lingual gyrus (L) 17 −16 [−16] −94 [−96] −5 [−12] 4.66 35
Middle temporal gyrus (L)* 21 −63 [−64] −43 [−44] 2 [0] 3.99 27
Middle frontal gyrus (R) 9 51 [52] 21 [20] 25 [28] 3.82 26
Lingual gyrus (R)* 18 16 [16] −50 [−52] 3 [0] 3.83 14
Thalamus (L) −24 [−24] −27 [−28] −2 [−4] 4.05 13
Insula (R) 13 40 [40] 16 [16] 3 [4] 3.86 12

Significance level and the size of the respective activation cluster (number of voxels) at p < 0.05, corrected. Coordinates are listed in Talairach and Tournoux (1988)
atlas space with MNI coordinates in []. BA is the Brodmann area nearest to the coordinate and should be considered approximate. Areas that are present in both
contrasts are marked with asterisk (*).

jects had a taxonomic preference (72% vs. 28% and 63% vs. included left middle temporal gyrus, left and right cuneus, left
37%) and one subject showed a thematic preference (66% vs. angular and precentral gyrus, left parahippocampal gyrus and
34%). These subjects were not excluded from further analy- left medial frontal gyrus. Finally, there was also a small area of
ses since it was expected that some participants may be biased activation in the right lingual gyrus.
in one or other direction, reflecting a normal variability in the Very similar areas were recruited for correct taxonomic
population. choices in the taxonomically biased condition. The largest clus-
ter of activation (400 voxels) was again in the left inferior frontal
2.1.2. Reaction time data gurus spreading into left medial frontal gyrus. Left angular
Reaction time was measured from the moment the two gyrus, right cuneus, left parahippocampal gyrus and left mid-
matches were presented till the subject made a correct response. dle temporal gyrus were activated similarly to the thematically
In the biased conditions it took subjects on average 1154 ms biased condition. However, lingual gyrus was activated in this
(S.D. = 179 ms) to make a choice in the thematically biased condition bilaterally, and there were two more significant clus-
condition and 1191 ms (S.D. = 184 ms) to make a choice in ters of activation in the right hemisphere—one in the right middle
the taxonomically biased condition. As in the choice data, frontal gyrus and one in right insula. Finally, left thalamus was
there was no significant difference between the two (p = 0.29). activated as well. For exact coordinates, z-statistics and cluster
A similar pattern was observed in the balanced condition. It sizes see Table 2.
took subjects on average 1189 ms (S.D. = 165 ms) to make a None of the differences in the areas recruited in the processing
thematic choice and 1186 ms (S.D. = 172 ms) to make a tax- of isolated thematic and taxonomic categories remained signifi-
onomic choice. Again, there was no significant difference cant in the subtraction of thematically biased condition from the
between the time it took to make the two types of choices taxonomically biased condition or the other way round.
(p = 0.49).
2.2.2. Balanced condition, thematic choice
2.2. Imaging data For the analysis of the imaging data from the balanced tri-
als all responses were grouped according to the type of the
2.2.1. Thematically and taxonomically biased condition response—thematic or taxonomic and analyzed as separate con-
Comparing the activation corresponding to the correct the- ditions.
matic choices in the thematically biased condition against the Comparing the activation corresponding to the thematic
baseline, we found the largest cluster of activation (275 vox- choice in the balanced condition to the baseline, we found five
els) in the left inferior frontal gyrus. Other areas of activation clusters of activation again located in the left inferior frontal
414 O. Sachs et al. / Neuropsychologia 46 (2008) 409–418

Fig. 1. The areas activated in the balanced cross-categorization condition. Panel A: Regions activated for the contrast of thematic choices in the balanced condition
minus baseline. Panel B: Regions activated for the contrast of taxonomic choices in the balanced condition minus baseline. Panel C: Regions activated for the contrast
of taxonomic choices in the balanced condition minus thematic choices in the balanced condition. The panels have different scaling and scaling steps. The activation
map is shown overlaid onto selected sections of a canonical single-subject T1-weighted volume. The position of selected sections corresponds to the biggest clusters
of activation corrected at p < 0.05 across the whole brain. Scale bar shows t-values for contrasts.

gyrus, spreading to the left superior frontal gyrus. Left middle biggest clusters were again in left inferior frontal gyrus (459
temporal gyrus and right cuneus were activated as in the themat- voxels) and in the left middle temporal gyrus (170). However,
ically biased condition. However, there was also a small area of the area of activation in the right inferior and middle frontal
activation in the right hemisphere–right inferior frontal gyrus. gyrus was larger than in the thematic choice (118 voxels com-
See Fig. 1A and Table 3 for a summary of activations in this pared to 11 voxels). Except for the activation of right thalamus,
condition. all other areas of activation corresponding to the taxonomic
choice in the balanced trials were limited to the left hemisphere:
2.2.3. Balanced condition, taxonomic choice left medial frontal gyrus, left precuneus, left cuneus, left lin-
Comparing the activation corresponding to the taxonomic gual gyrus and left parahippocampal gyrus (see Fig. 1B and
choice in the balanced condition to the baseline, the main and Table 3).
O. Sachs et al. / Neuropsychologia 46 (2008) 409–418 415

Table 3
Activation peaks for the balanced condition with their localization
Anatomical region BA Coordinates z-Value No. voxels

x y z

Thematic choice > baseline


Inferior frontal gyrus (L)* 45, 47 −52 [−52] 24 [24] 21 [24] 6.71 478
Middle temporal gyrus (L)* 21 −59 [−60] −39 [−40] −1 [−4] 5.52 144
Superior frontal gyrus (L) 6 −4 [−4] 18 [16] 47 [52] 5.45 121
Cuneus (R) 18 20 [20] −97 [−100] 1 [−4] 4.41 13
Inferior frontal gyrus (R)* 47 48 [48] 19 [20] −4 [−4] 3.82 11
Taxonomic choice > baseline
Inferior frontal gyrus (L)* 47, 45 −51 [−52] 24 [24] 21 [24] 7.41 459
Middle temporal gyrus (L)* 21 −59 [−60] −35 [−36] 2 [0] 6.12 170
Inferior frontal gyrus (R)* 9, 45 48 [48] 24 [24] 21 [24] 4.81 118
Medial frontal gyrus (L) 6 0 [0] 14 [12] 47 [52] 5.01 114
Precuneus (L) 7, 19 −36 [−36] −68 [−72] 40 [40] 4.68 101
Cuneus (L) 17, 18 −16 [−16] −77 [−80] 15 [12] 3.86 38
Thalamus (R) 4 [4] −8 [−8] 4 [4] 3.76 37
Lingual gyrus (L) 17 −16 [−16] −94 [−96] −5 [−12] 5.52 26
Parahippocampal gyrus (L) 19 −16 [−16] −47 [−48] 2 [0] 4.55 20

Significance level and the size of the respective activation cluster (number of voxels) at p < 0.05, corrected. Coordinates are listed in Talairach and Tournoux (1988)
atlas space with MNI coordinates in []. BA is the Brodmann area nearest to the coordinate and should be considered approximate. Areas that are present in both
contrasts are marked with asterisk (*).

Table 4
Activation peaks for taxonomic choices in balanced condition > thematic choices in balanced condition with their localization
Anatomical Region BA Coordinates z-Value No. voxels

x y z

Precuneus (L) 7 −8 [−8] −68 [−72] 40 [40] 4.01 24


Middle frontal gyrus (R) 46 48 [48] 28 [28] 24 [28] 3.96 22
Thalamus (L) −20 [−20] −16 [−16] 4 [4] 4.14 14

Significance level and the size of the respective activation cluster (number of voxels) at p < 0.05, corrected. The opposite contrast (thematic choices > taxonomic
choices) revealed no difference in activation. Coordinates are listed in Talairach and Tournoux (1988) atlas space with MNI coordinates in []. BA is the Brodmann
area nearest to the coordinate and should be considered approximate.

2.2.4. Comparing taxonomic and thematic choice in the across conditions in the thalamus is qualitatively (if not quan-
balanced condition titatively) very similar to the patterns in the right MFG and the
When balanced trials where the thematic choice was made precuneus. All of these regions have increased activation for the
were subtracted from balanced trials where the taxonomic choice taxonomic choice and a deactivation for the thematic choice.
was made, three significant clusters of activation were found: in
left thalamus, right middle frontal gyrus and left precuneus (see 3. Discussion
Fig. 1C and Table 4). All of these areas were also present in
the activations for the taxonomic choice of the balanced condi- The main objective of this study was to (1) compare the neural
tion compared to the baseline. The opposite contrast (thematic correlates of thematic and taxonomic categories and (2) to inves-
choice > taxonomic choice) revealed no differences. tigate the issue of cross-categorization. In response to the first
Since the activations that were found in the taxo- objective we found that, when investigated in isolation, thematic
nomic > thematic contrast were relatively small, we investigated categories recruited the same cortical network as the processing
them more thoroughly to rule out that the differences that we of “classical” similarity-based taxonomic categories: left infe-
found are not due to deactivations in the thematic condition rior frontal, middle temporal and occipital regions that many
compared to the baseline, but represent true increases of the other studies have found to be implicated in the processing of cat-
BOLD response in the taxonomic condition. By plotting param- egories (Devlin et al., 2002; Perani, Schnur, et al., 1999; Spitzer
eter estimates for each one of the conditions we established that et al., 1998; Tyler et al., 2003; Vitali et al., 2005). It is worth men-
in all three clusters of activation there was an increase of sig- tioning that unlike thematic categories, constructing taxonomic
nal in the taxonomic condition compared to the baseline and to categories involved activation in the right middle frontal gyrus
the thematic condition (see Fig. 2). It is worth noting that the and in the right insula, although this difference was not signifi-
parameter estimate for the taxonomic condition in the thalamus cant after the subtraction of thematic trials from the taxonomic
is weaker than in other regions. However, the overall pattern ones.
416 O. Sachs et al. / Neuropsychologia 46 (2008) 409–418

tive as they were to form a taxonomic category in the presence


of a thematic alternative. This pattern of results fits with the
imaging data from the biased conditions showing no differences
between thematic and taxonomic categories when they have no
competing alternatives. However, the brain activation data sug-
gest that making a subjective taxonomic choice in the presence
of a thematic alternative requires the recruitment of an additional
cortical network, i.e. thalamus, right middle frontal gyrus and
left precuneus. In contrast, making a thematic choice in the pres-
ence of a taxonomic alternative did not lead to the recruitment
of this network or any additional brain regions. This finding
supports our hypothesis of increased effort in the processing of
taxonomic categories that can be reflected in the involvement
of the right-hemispheric regions. Moreover, in line with our
predictions, this difference between brain areas involved in the
construction of taxonomic and thematic categories became sig-
nificant only when the two categories were pitted against each
other in the cross-categorization condition. Our results are all
the more interesting considering that the stimuli that subjects
saw while making both types of categories were identical. The
only difference is in the choice they made—either matching
objects thematically or taxonomically according to their subjec-
tive preference. The fact that matching objects taxonomically
required additional recruitment of the thalamus which is asso-
ciated with rule-based categorization (Grossman, Smith, et al.,
2002) or selective attention (Frith & Friston, 1996) suggests that
constructing taxonomic categories might require more effort.
Indeed, thematic concepts that are highly associated might be
activated with less effort than taxonomic concepts that are related
at the superordinate level and share a more abstract relationship.
Several categorization studies have also proposed that detec-
tion of similarity between taxonomically related entities takes a
deeper level of analysis (Lin & Murphy, 2001; Mandler, 1983).
The activation of right frontal cortex that was found to be impli-
cated in processing abstract nouns and verbs (Grossman, Kenig,
et al., 2002; Perani, Cappa, et al., 1999) supports the idea that
forming taxonomic categories requires assessing a more abstract
relationship between the target and the match word. This process
can also be more effortful and require the additional recruitment
of the right hemisphere. The greater activation of precuneus in
the taxonomic choice condition may indicate that additional
mental imagery processing was necessary to decide if taxo-
nomic matches were similar enough to the target to make up
a category. The activation of precuneus could also be associated
Fig. 2. Parameter estimates for the contrast of taxonomic choices in the bal-
anced condition minus thematic choices in the balanced condition. Plots of
with increased demands set on the retrieval of episodic mem-
mean parameter estimates with ± standard error showing response amplitude in ory and contextual associations that were necessary to make the
precuneus (top), middle frontal gyrus (center) and thalamus (bottom) for each taxonomic choice (Fletcher et al., 1995; Lundstrom, Ingvar, &
condition: thematically biased, taxonomically biased, thematic choice in the Petersson, 2005).
balanced condition, taxonomic choice in the balanced condition and baseline. Outside of the competition or cross-categorization context,
Labeling of the y-axis is in arbitrary units.
processing of taxonomic and thematic categories in isolation
involved markedly similar brain regions—left inferior frontal
Investigating the issue of cross-categorization, the second gyrus, left middle temporal gyrus, cuneus, precuneus, lingual
objective of our study, has shown that there is no difference and angular gyri. Many previous studies have found these areas
in the reaction time or choice patterns when subjects had to to be implicated in explicit semantic tasks, semantic processing
pick either a taxonomic or a thematic match to make a category. and categorization. More specifically, the left inferior frontal
When subjects had the free choice they were just as likely to cortex is typically associated with lexical retrieval of seman-
form a thematic category in the presence of a taxonomic alterna- tic knowledge, executive aspects of semantic processing, such
O. Sachs et al. / Neuropsychologia 46 (2008) 409–418 417

as semantic working memory, semantic search, or comparing and by the START-Program of the Faculty of Medicine, RWTH
semantic concepts in working memory (Bookheimer, 2002). Aachen.
Since most of our stimuli were artifacts, the involvement of mid-
dle temporal gyrus, usually activated for tools or tool-associated References
actions is not surprising either (Beauchamp, Lee, Haxby, &
Martin, 2002; Martin, Ungerleider, & Haxby, 2000; Tyler et al., Annett, M. (1970). Classification of hand preference by association analysis.
2003). Finally, the activation of such visual areas as left and right British Journal of Psychology, 61, 303–321.
cuneus, precuneus and lingual gyrus has been previously asso- Baayen, R. H., Piepenbrock, R., & van Rijn, H. (1993). The CELEX Lexi-
ciated with the involvement of a visualization strategy during cal Database (CD-ROM). Philadelphia, PA: Lingusitic Data Consortium,
University of Pennsylvania.
semantic tasks and inspecting visual images (Fletcher, Shallice, Barsalou, L. W. (1983). Ad hoc categories. Memory & Cognition, 11, 211–227.
Frith, Frakowiak, & Dolan, 1996; Perani, Schnur, et al., 1999; Beauchamp, M. S., Lee, K. E., Haxby, J. V., & Martin, A. (2002). Parallel visual
Vitali et al., 2005). This function can be equally important for motion processing streams for manipulable objects and human movements.
forming and evaluating both thematic and taxonomic categories Neuron, 34, 149–159.
made up of artifacts. Beeman, M., Friedman, R. B., Grafman, J., & Perez, E. (1994). Summation
priming and coarse semantic coding in the right hemisphere. Journal of
Several caveats must be kept in mind regarding our results. Cognitive Neuroscience, 6, 26–45.
First, our results are necessarily driven by the semantic pro- Bookheimer, S. (2002). Functional MRI of language: New approaches to under-
cesses involved in completing the cognitive task that we used. standing the cortical organization of semantic processing. Annual Review
These task demands can interact with the semantic knowl- Neuroscience, 25, 151–188.
edge and influence the neural activation pattern that we found Cappa, S. F., Perani, D., Schnur, T., Tettamanti, M., & Fazio, F. (1998). The
effects of semantic category and knowledge type on lexical–semantic access:
(Grossman et al., 2006). Second, thematic categories often share A PET study. NeuroImage, 8, 350–359.
more than one type of thematic relations and are often con- Caramazza, A., & Shelton, J. (1998). Domain-specific knowledge systems in
founded by a spatial relationship or a part-whole relationship the brain: The animate-inanimate distinction. Journal of Cognitive Neuro-
(that is also functional—otherwise there would be no need for science, 10(1), 1–34.
the part). Although in our stimuli development we controlled Chiarello, C., & Richards, L. (1992). Another look at categorical priming in
cerebral hemispheres. Neuropsychology, 30, 381–392.
for this factor and made sure that our thematic categories share Damasio, H., Grabowski, T. J., Tranel, D., Hichwa, R. D., & Damasio, A. R.
only either a functional or a part-whole relationship, there is (1996). A neural basis for lexical retrieval. Nature, 380, 499–505.
still a possibility that not all functional relations are 100% Devlin, J. T., Russell, R. P., Davis, M. H., Price, C. J., Moss, H. E., Fadili, M. L.,
pure. et al. (2002). Is there an anatomical basis for category-specificity? Semantic
In conclusion, our results show that on the neural level exter- memory studies in PET and fMRI. Neuropsychologia, 40, 54–75.
Fletcher, P. C., Frith, C. D., Baker, S. C., Shallice, T., Frakowiak, R. S., &
nal correlations that build thematic relations between objects Dolan, R. J. (1995). The minds’s eye—Precuneus activation in memory-
are processed in a way that is very similar to the process- related imagery. NeuroImage, 2, 195–200.
ing of similarity-based feature correlations. External correlation Fletcher, P. C., Shallice, T., Frith, C. D., Frakowiak, R. S. J., & Dolan, R. J.
between objects like closet and dress has obviously compen- (1996). Imagery and semantic cueing in episodic memory: The influence of
sated for their perceptual dissimilarity to lead to the same imagery and semantic cueing. Brain, 119, 1587–1596.
Friston, K. J., Fletcher, P., Josephs, O., Holmes, A., Rugg, M. D., & Turner,
neural activation pattern as the one elicited by closet and R. (1998). Event-related fMRI: Characterizing differential responses. Neu-
dresser—members of the same taxonomic category furniture. roImage, 7, 30–40.
Thus, our results have implications for the general question Frith, C. D., & Friston, K. J. (1996). The role of the thalamus in “top-down”
of semantic knowledge organization in the brain. Taxonomic modulation of attention to sound. NeuroImage, 4, 210–215.
and thematic groupings seem to be two equally important ways Grabowski, T. J., Damasio, H., & Damasio, A. R. (1998). Premotor and prefrontal
correlates of category-related lexical retrieval. NeuroImage, 7, 232–243.
of recognizing commonalities among objects. It can be taken Grossman, M., Koenig, P., DeVita, C., Glosser, G., Alsop, D., Detre, J., et al.
as evidence that classical taxonomic categorization is not the (2002). The neural basis for category-specific knowledge: An fMRI study.
only way of organizing semantic knowledge and that thematic NeuroImage, 15, 936–948.
associations create equally salient conceptual groups. Whether Grossman, M., Smith, E. E., Koenig, P., Glosser, G., DeVita, C., Moore, P., et al.
such groups can be called categories in neuroimaging research (2002). The neural basis for categorization in semantic memory. NeuroIm-
age, 17, 1549–1561.
is an open question—our results suggest that there is no rea- Grossman, M., Koenig, P., Kounios, J., McMillan, C., Work, M., & Moore, P.
son why they should not. While the current study only begins (2006). Category-specific effects in semantic memory: Category-task inter-
to tackle this issue, further research of thematic and other actions suggested by fMRI. NeuroImage, 30, 1003–1009.
ways of semantic organization can improve our understanding Hagoort, P., Brown, C., & Swaab, T. Y. (1996). Lexical-semantic event-related
of the fundamental issue motivating this research–knowledge potential effects in patients with left hemisphere lesions and aphasia, and
patients with right hemisphere lesions without aphasia. Brain, 119, 627–649.
representation in the brain and the principles of conceptual Kotz, S. A., Cappa, S. F., von Cramon, D. Y., & Friederici, A. D.
organization. (2002). Modulation of the lexical–semantic network by auditory seman-
tic priming: An event-related functional MRI study. NeuroImage, 17,
Acknowledgements 1761–1772.
Leube, D. T., Erb, M., Grodd, W., Bartels, M., & Kircher, T. T. J. (2001). Acti-
vation of right fronto-temporal cortex characterizes the ‘living’ category in
The study was supported by a grant from the Interdisciplinary semantic processing. Cognitive Brain Research, 12(3), 425–430.
Center for Clinical Research “BIOMAT” within the Faculty of Lin, E. L., & Murphy, G. L. (2001). Thematic relations in adults’ concepts.
Medicine at the RWTH Aachen University (IZKF VV N68) Journal of Experimental Psychology: General, 130(1), 3–28.
418 O. Sachs et al. / Neuropsychologia 46 (2008) 409–418

Lopez, A., Atran, S., Coley, J. D., Medin, D. L., & Smith, E. E. (1997). The Shafto, P., & Coley, J. D. (2003). Development of categorization and reason-
tree of life: universal and cultural features of folkbiological taxonomies and ing in the natural world: Novices to experts, naive similarity to ecological
inductions. Cognitive Psychology, 32, 251–295. knowledge. Journal of Experimental Psychology: Learning, Memory and
Lundstrom, B. N., Ingvar, M., & Petersson, K. M. (2005). The role of pecuneus Cognition, 29, 641–649.
and left inferior frontal cortex during source memory episodic retrieval. Slotnick, S. D., Moo, L. R., Segal, J. B., & Hart, J.Jr. (2003). Distinct prefrontal
NeuroImage, 27, 824–834. cortex activity associated with item memory and source memory for visual
Mandler, J. M. (1983). Representation. In J. H. Flavell & E. M. Markman (Eds.), shapes. Brain Research: Cognitive Brain Research, 17, 75–82.
Handbook of child psychology Cognitive development (pp. 420–494). New Smiley, S. S., & Brown, A. L. (1979). Conceptual preference for thematic or
York: Wiley. taxonomic relations: A non-monotonic age trend from preschool to old age.
Martin, A., Wiggs, C., Ungerleider, L., & Haxby, J. (1996). Neural correlates of Journal of Experimental Child Psychology, 28, 249–257.
category-specific knowledge. Nature, 379, 649–652. Spitzer, M., Kischka, U., Gückel, F., Belleman, M., Kammer, T., Seyyedi, S.,
Martin, A., Ungerleider, L. G., & Haxby, J. V. (2000). Category-specificity and et al. (1998). Functional magnetic resonance imaging of category-specific
the brain: The sensory-motor model of semantic representations of objects. cortical activation: Evidence for semantic maps. Cognitive Brain Research,
In M. S. Gazzaniga (Ed.), The cognitive neurosciences (2nd ed., Vol. 3, pp. 6, 309–319.
1023–1036). Cambridge: MIT Press. Talairach, J., & Tournoux, P. (1988). Co-planar stereotaxic atlas of the human
Medin, D. L., Lynch, E. B., Coley, J. D., & Atran, S. (1997). Categorization brain. New York: Thieme.
and reasoning among tree experts: Do all roads lead to Rome? Cognitive Thompson-Schill, S. L., Aguirre, G. K., D’Esposito, M., & Farah, M. J. (1999).
Psychology, 32, 49–96. A neural basis for category and modality specificity of semantic knowledge.
Nguen, S. P., & Murphy, G. L. (2003). An apple is more than just a Neuropsychologia, 37, 671–676.
fruit: Cross-classification in children’s concepts. Child Development, 74(6), Tyler, L. K., Moss, H. E., Durrant-Peatfield, M. R., & Levy, J. P. (2000). Con-
1783–1806. ceptual structure and the structure of concepts: A distributed account of
Perani, D., Cappa, S., Schnur, T., Tettamanti, M., Collins, S., Rosa, M. M., et category-specific deficits. Brain and Language, 75, 195–231.
al. (1999). The neural correlates of verb and noun processing: A PET study. Tyler, L. K., Stamatakis, E. A., Dick, E., Bright, P., Fletcher, P., & Moss,
Brain, 122, 2337–2344. H. (2003). Objects and their actions: Evidence for a neurally distributed
Perani, D., Schnur, T., Tettamanti, M., Gorno-Tempini, M., Cappa, S. F., & Fazio, semantic system. NeuroImage, 18, 542–557.
F. (1999). Word and picture matching: A PET study of semantic category Vitali, P., Abutalebi, J., Tettamanti, M., Rowe, J., Scifo, P., Fazio, F., et al. (2005).
effects. Neuropsychologia, 37, 293–306. Generating animal and tool names: An fMRI study of effective connectivity.
Ross, B. H., & Murphy, G. L. (1999). Food for thought: Cross-classification and Brain and Language, 93, 32–45.
category organization in a complex real-world domain. Cognitive Psychol- Warrington, E., & McCarthy, R. (1983). Category specific access dysphasia.
ogy, 38, 495–553. Brain, 106, 859–878.

You might also like