The Effects of Semantic Category and Knowledge Typeon Lexical-Semantic Access: A PET Study

S. F. Cappa,* D. Perani,†,‡ T. Schnur,‡ M. Tettamanti,‡ and F. Fazio†,‡,§†Institute of Neuroscience and Bioimaging–CNR, 20132 Milan; ‡Scientific Institute H. S. Raffaele, Milan; §University of Milan, Milan;

and *University of Brescia, Brescia, Italy

Received October 6, 1997

Neuropsychological studies of patients with category-specific recognition disorders, as well as PET investiga-tions of semantic category effects in visual recognitiontasks, have led some authors to the hypothesis thatvisual-perceptual knowledge plays a crucial role in therecognition of natural items, such as animals, whilefunctional-associative information is more importantfor the recognition of man-made tools. To study thecerebral correlates of the retrieval of different types ofsemantic knowledge about living and nonliving enti-ties, we performed a PET experiment in which normalsubjects were required to access visual- and functional-associative information related to visually presentedwords corresponding to animals and tools. The experi-mental conditions were the following: (1) Rest. (2)Baseline: letter detection in pseudo-words. (3) Animal,visual knowledge: decide whether the animal has along or short tail with respect to the body. (4) Animal,associative knowledge: decide whether the animal istypically found in Italy. (5) Tool, visual knowledge:decide whether the object is longer than wider or viceversa. (6) Tool, functional knowledge: decide whetherthe object is typically used as a kitchen tool. Lexical-semantic access (all lexical conditions pooled) acti-vated the prefrontal cortex on the left and the parietal–occipital junction and posterior cingulate cortexbilaterally. An analysis of the individual experimentalconditions in comparison with the nonword baselineshowed that accessing visual versus associative knowl-edge was associated with different activation patterns:predominantly frontal in the case of visual features,temporoparietal for associative knowledge. While theactivation patterns involved similar areas for livingand nonliving entities, in the case of the latter theywere restricted to the left hemisphere. The analysis ofmain effects confirmed these findings: there were sev-eral significant differences in the visual-associativecomparison, while category-related differences wereless prominent. These findings indicate that the re-trieval of different types of knowledge is associatedwith distinct patterns of brain activation; on the other

hand, category-related differences were less evidentthan in picture matching and naming tasks. r 1998

Academic Press

INTRODUCTION

According to information processing models, bothpicture identification and word understanding requireaccess to semantic knowledge from, respectively, visual(‘‘structural’’) and lexical representations. While someauthors have claimed that a single, amodal semanticsystem is responsible for the identification and under-standing of both pictorial and lexical stimuli (Riddochet al., 1988; Caramazza et al., 1990), others haveargued for the existence of multiple, modality-specificsystems (Shallice, 1988). The debate between propo-nents of these two opposing views is largely based onneuropsychological findings, coming from individualpatients with modality-specific disorders of naming,such as optic aphasia (Beauvois, 1982), and with selec-tive disorders of picture and word comprehension (War-rington, 1975). There is ample evidence that the under-standing of words and pictures can be differentiallyimpaired by brain damage. Some patients, clinicallylabeled associative agnosics, are selectively impaired inrecognizing visual objects or pictures, without a disor-der of word comprehension, while the reverse dissocia-tion (i.e., a selective impairment in word comprehen-sion with preserved picture recognition) has also beenobserved (Bub et al., 1988; McCarthy and Warrington,1988). These dissociations have been interpreted asevidence for modality-specific (i.e., visual and verbal)semantic systems (Shallice, 1988) or alternatively at-tributed, respectively, to a subtle impairment of visualperception (in the case of patients with word comprehen-sion superior to picture recognition) or to the intrinsi-cally superior difficulty in understanding a word incomparison to a picture (Organized Unitary ContentHypothesis—OUCH model—Caramazza et al., 1990).Category-specific effects in naming and comprehension(Warrington and Shallice, 1984) have also been consid-

NEUROIMAGE 8, 350–359 (1998)ARTICLE NO. NI980368

3501053-8119/98 $25.00Copyright r 1998 by Academic PressAll rights of reproduction in any form reserved.

ered to speak for or against the unique/multiple seman-tic system debate. If the disorder in the recognition ofliving entities is related to defective perceptual knowl-edge or defective access to it, damage to a visualsemantic system might be considered a possible expla-nation for selective disorders in animal recognition; theconverse dissociation could be ascribed to a selectiveimpairment of ‘‘verbal’’ semantics, which is consideredto be the repository of functional-associative knowledge(Warrington and Shallice, 1984). It must be underlined,however, that this type of evidence is also compatiblewith a partial damage within an amodal unitary seman-tic system in which category-related items are repre-sented contiguously (Hillis and Caramazza, 1991) orwith focal damage to domain-specific knowledge sys-tems (Caramazza and Shelton, 1998).

Functional imaging methods have been recently ap-plied to the investigation of the neural correlates ofpicture and word recognition (see Bly and Kosslyn,1997), for a review). There is evidence for differentialactivations in the brain while the subjects are process-ing pictures representing living and nonliving entitiesin order to perform a same–different matching task(Perani et al., 1995) or to name them (Damasio et al.,1996; Martin et al., 1996). The observation of differentregional patterns of brain activation involving, respec-tively, the inferior temporal lobe for animal recognitionand frontotemporal associative cortex for tools can beconsidered to reflect separate semantic representationsaccording to category. However, the localization offunctional activation might also be compatible with thehypothesis of a crucial role of perceptual knowledge inanimal recognition and of functional-associative knowl-edge for artifacts (Warrington and Shallice, 1984).There is ample evidence for the participation of inferiortemporo-occipital areas in higher-order visual process-ing; the prevalent left frontotemporal activation foundwith artifact discrimination, on the other hand, mightreflect the close connection between linguistic andfunctional-associative knowledge. These two hypoth-eses are not mutually exclusive, as separate representa-tions might have developed in the neighborhood of thebrain areas preferentially involved in processing theclass of information which plays the most importantrole, crucial for the identification of that category.

In order to investigate directly the cerebral corre-lates of accessing different types of semantic informa-tion from lexical stimuli we have performed a PETexperiment, using visually presented words. In thistask, the subjects were required to access the meaningof visually presented words in order to be able toanswer questions related either to perceptual or tofunctional-associative characteristics of the stimuli.The same tasks were used for both animals and tools, inorder to allow a direct comparison of the differences in

brain activation related to, respectively, semantic cat-egory and class of knowledge.

MATERIALS AND METHODS

Subjects

The experimental subjects were 13 right-handedmale volunteers (age range 22–26 years), who gavewritten informed consent prior to the experiment. Allsubjects had no history of neurological or psychiatricdisorders. Right-handedness was verified using theEdinburgh Inventory (Oldfield, 1971). The experimen-tal protocol was approved by the local hospital ethicscommittee.

Experimental Design and Materials

Subjects underwent 12 consecutive rCBF measure-ments for this experiment, 2 for each of the followingsix conditions.

(a) Rest condition (silence): subjects were instructedto close their eyes and ‘‘empty their minds.’’

(b) Legal pseudo-word task (baseline): pronounceablepseudo-words were presented one at a time; subjectshad to decide whether the pseudo-word contained theletter ‘‘e.’’

(c) Animal/visual knowledge task: words referring toanimals were presented one at a time; subjects had todecide whether the tail of the animal was long or shortwith respect to the body.

(d) Animal/associative knowledge task: words refer-ring to animals, as above; subjects had to decidewhether the animal was typically found in Italy.

(e) Tool/visual knowledge task: words referring totools, as above; subjects had to decide whether theobject was longer than wider or wider than longer.

(f) Tool/functional knowledge task: words referring tononliving objects, as above; subjects had to decidewhether the object was typically used for food prepara-tion activities (see Table 1 for examples).

Subjects were asked to respond to each task bypressing the right or left button on a response box withthe right hand. The stimuli for the animal and artifactdiscrimination tasks were selected from a pilot study innormal subjects. The stimuli were identical for the twoanimal tasks and for the two tool tasks and weredisplayed at a 2.3-presentation rate. Mean reactiontimes during the scanning sessions were measured.The legal pseudo-word control task was chosen tocontrol for the visual input and motor response associ-ated with each task. A preliminary investigation wasconducted in which given features of an object oranimal (length vs width, frequency of object use in foodpreparation, long/short tail, habitat of the animal) wererated on a scale of 1 to 5 by 25 volunteers (age range18–30 years). Words that were selected as most typical

351SEMANTIC ACCESS: VISUAL WORD PROCESSING

of their category were checked for frequency and length.Only items which elicited consistent responses (morethan 90% of subjects) were used. From this list, foursets of stimuli were constructed containing two blocksof 20 words each.

PET Data Acquisition

rCBF was measured by recording the distribution ofradioactivity following an intravenous injection of 15O-labeled water (H2

15O) with a GE-Advance scanner(General Electric Medical System, Milwaukee, WI),which has an axial field of view of 15.2 cm, thusallowing a sampling of both the entire brain and thecerebellum. Data were acquired by scanning in 3Dmode. A 5-mCi bolus of H2

15O was injected as a tracer ofblood flow while scans were acquired (Mazziotta et al.,1985). After attenuation correction (measured by atransmission scan using a pair of rotating pin sourcesfilled with 68Ge), the data were reconstructed as 35transaxial images (4.25 mm each). The reconstructedtransaxial FOV was 25 cm on an image matrix of 128 3128 (pixel size of 1.95 mm). The reconstruction programwas a three-dimensional filtered back projection with aHanning filter (cutoff 4 mm filter width) in the trans-axial plane and a Ramp filter (cutoff 8.5 mm) in theaxial direction. The integrated counts collected for 70 s,starting 20 s after injection time, were used as an index

of rCBF; 12 scans related to six different experimentalconditions were acquired for each subject.

Data Analysis

Image manipulations and statistical analysis wereperformed in MATLAB 4.2 (Math Works, Natick, MA)using statistical parametric mapping (SPM96, Well-come Department of Cognitive Neurology, London,UK). PET images were transformed into a standardstereotactic space (Talairach and Tournoux, 1988; Fris-ton et al., 1995a) (final voxel size, 2 3 2 3 4 mm); globaldifferences in cerebral blood flow were covaried out forall voxels, and comparisons across conditions weremade using t statistics with appropriate linear con-trasts (Friston et al., 1995b). In order to increasesignal-to-noise ratio and accommodate normal variabil-ity in functional gyral anatomy each image wassmoothed in three dimensions with a low-pass Gauss-ian filter (15 3 15 3 15 mm). A repeated-measuresANCOVA was used for the comparison of differenttasks, in which every subject was studied under allconditions. The SPM package allows the comparison ofone experimental condition relative to a pair or a groupof control conditions, specifying a set of weights thatare used to add the means together with a sum set tozero. The contrast is divided by the adjusted root meansquare error and the resulting value has the t distribu-tion under the null hypothesis of no differences. Thisset of Z values comprises the statistical parametricmap (SPM)5t6). Only regional activations significant atP # 0.001 (uncorrected) were considered.

Statistical Comparisons

The distribution of rCBF during the different taskswas investigated according to a subtractive (1, 2, 3)as well as a factorial (4) design: (1) Pseudo-wordsversus rest condition (silence) (b 2 a), (2) lexical-semantic access versus pseudo-words [(c 1 d 1 e 1 f ) 2b], (3) lexical-semantic conditions (individually)versus pseudo-words [(c 2 b), (d 2 b), (e 2 b), ( f 2 b)],and (4) main effects of class of knowledge [visualsemantics, (c 1 e) 2 (d 1 f ); associative semantics,(d 1 f ) 2 (c 1 e)] and category [living, (c 1 d) 2 (e 1 f );nonliving, (e 1 f ) 2 (c 1 d)]. Interaction effects be-tween class of knowledge and semantic category werealso sought with the appropriate formulas.

RESULTS

Behavioral Results

Measurement of reaction time (RT) demonstratedthat the pseudo-word letter-detection task was theeasiest of all tasks, while making visual semantic

TABLE 1

Examples of Experimental Stimuli

Living

Task 1Native (Î)/

not native (X) to Italy(associative)

Task 2Long tail (Î)/short tail (X)

(visual)

Lucertola (lizard) Î ÎTopo (mouse) Î ÎConiglio (rabbit) Î XCoccodrillo (croccodile) X ÎIppopotamo (hippopotamus) X XFoca (seal) X X

Nonliving

Task 3Kitchen (Î)/

nonkitchen tool (X)(associative)

Task 4Longer (Î)/wider (X)(visual)

Coltello (knife) Î ÎMattarello (rolling pin) Î ÎPentola (frying pan) Î XFucile (revolver) X ÎDivano (couch) X XPianoforte (piano) X X

Note. Stimuli were identical for both associative and visual tasksrespectively for living and nonliving stimuli. All tasks were repeatedtwice but stimuli were not identical between the first and the secondtime the same task was performed (e.g., a subject saw the samestimuli when responding about visual or associative information butthe second time he had to perform the task the stimuli differed).

352 CAPPA ET AL.

judgments took more time than making associativesemantic judgments (mean RT in milliseconds: pseudo-words 685.3, visual/living 1031.5, visual/nonliving1042.1, associative/living 791.1; associative/nonliving809.2) (ANOVA) [F(4, 12) 5 55.21, P , 0.0001]. This isa well-known effect, which prevents a matching of thetask according to difficulty. Within each task, for thevisual semantic task as well as the functional semantictask, no differences were seen in reaction times be-tween living and nonliving stimuli. Between tasks,subjects were significantly slower in visual semanticjudgments than in associative semantic judgments(P , 0.01) for judgments on either living or nonlivingstimuli (Tukey Honest Difference posthoc comparison).The stimuli had been selected in order to yield a highaccuracy in normal subjects; accordingly, the perfor-mance was above 80% correct for all tasks, withoutsignificant differences.

PET Results

Pseudo-words versus Rest

Detecting a letter in pseudo-words, in comparison tothe rest with eyes closed, activated bilaterally thelingual gyrus/calcarine sulcus (Ba 18/17) and the fusi-form gyrus (Ba 37). Separate foci were present in theleft cuneus (Ba 18), left inferior parietal lobule (Ba 40),and right superior parietal lobule (Ba 7). There werealso right cerebellar activations (see Table 2).

Lexical-Semantic Access versus Pseudo-words

All four lexical conditions were combined in order tocompare lexical-semantic access as a whole to thereading of pseudo-words (Table 2). Left hemisphericactivations were seen in the inferior (Ba 45) andsuperior (Ba 8) frontal gyrus and bilaterally in the

parieto-occipital junction (Ba 19/39) and posterior cin-gulate cortex (Ba 31/23) (see Table 3, Fig. 1).

Visual Semantic Tasks versus Pseudo-words

When subjects were required to give judgmentsconcerning respective dimensions of an item, a lefthemispheric pattern of activation was seen for nonliv-ing stimuli. Activation foci were located in the dorsolat-eral frontal cortex (Ba 45, 6, 8, and 9) and in the inferiorand middle temporal cortex (Ba 37/21 and 21). Forliving items there were right and left foci in theprefrontal cortex (Ba 45, 47, 6, 8, 10) and posteriorcingulate cortex (Ba 31/23) and a focus in the rightparieto-occipital junction (Ba 39/19) (see Table 4A).

Associative Semantic Tasks versus Pseudo-words

When subjects were required to give judgmentsconcerning functional-associative knowledge related toliving and nonliving items, bilateral activations in theparieto-occipital junction (Ba 39/19) and posterior cin-gulate cortex (Ba 23/31) were observed for living stimuliwith an additional focus in the left medial frontal gyrus(Ba 8). The activation was prevalently left hemisphericin the case of nonliving items, in the inferior frontalgyrus (Ba 45), and in the parieto-occipital junction (Ba39/19). The posterior cingulate cortex (Ba 23/31) wasalso activated bilaterally (see Table 4B).

Main Effects: Task

Several regions were activated in the comparisonbetween the visual semantic tasks and the associativesemantic tasks for living and nonliving stimuli: thesupramarginal gyri (Ba 40), inferior temporal cortex(Ba 37/20), and several prefrontal areas (Ba 6, 9, 10),bilaterally. The reverse comparison yielded a signifi-cant difference only in the posterior cingulate cortex(Ba 31/23) (see Table 5A and Figs. 2A and 2B).

Main effects: Category

Some significant differences were also observed forsemantic category: they were right-sided for living,predominantly left-sided for nonliving. In these compari-

TABLE 2

Pseudo-words versus Baseline

x y z Z

L lingual gyrus/calcarine sulcus(Ba 18/17) 220 296 212 8.6

L lingual gurus/calcarine sulcus(Ba 18/17) 216 298 28 8.3

L inferior parietal lobule (Ba 40) 234 232 52 7.8L fusiform gyrus (Ba 37) 246 252 220 6.4L cuneus (Ba 18) 222 276 12 4.3L/R lingual gyrus (Ba 18) 2 274 0 7.8R lingual gyrus/calcarine sulcus

(Ba 18/17) 16 298 28 8.3R fusiform gyrus (Ba 37) 46 252 220 5.1R superior parietal lobule (Ba 7) 28 270 40 4.9R superior parietal lobule (Ba 7) 32 260 44 4.4R cerebellar nuclei 4 262 228 7.1R cerebellum 20 252 220 6.8

TABLE 3

Combined Lexical Access Conditions versus Pseudo-words

x y z Z

L inferior frontal gyrus (Ba 45) 244 22 20 4.7L parieto-occipital junction (Ba 39/19) 242 272 24 3.8L superior frontal gyrus (Ba 8) 212 20 48 3.9L superior frontal gyrus (Ba 8) 214 32 40 3.5L/R posterior cingulate cortex

(Ba 31/23) 0 258 12 4.1R parieto-occipital junction (Ba 39/19) 42 276 28 4.0

353SEMANTIC ACCESS: VISUAL WORD PROCESSING

sons, the living condition activated the right middlefrontal gyrus (Ba 10) and the right fusiform gyrus (Ba37/20), whereas the nonliving condition activated theleft temporo-occipital junction (Ba 37/19), the left supra-marginal gyrus (Ba 40), the right superior temporalgyrus (Ba 22), and the right thalamus. There were nosignificant interactions between task and category (seeTable 5B and Figs. 2C and 2D).

DISCUSSION

The first part of this study was analyzed according toa classical subtractive design, with a ‘‘rest’’ condition(silence) as a baseline for the letter detection task. Thelatter task was then used as baseline for prelexicalprocessing of visual words.

Detection of letters in pseudo-words activated bilat-eral striate and extrastriate brain areas. These activa-tions are in agreement with the results of previousinvestigations and can be attributed to the processingof visual forms and specifically of elementary letterfeatures (Petersen et al., 1988; Bookheimer et al., 1995)(Table 2). A ventral occipital pathway activation duringpseudo-word reading has already been reported (Frithet al., 1995). The activation in the left inferior parietallobule might be related to phonological processing assuggested by previous PET studies (Paulesu et al.,1993). Indeed, this activation may reflect the subjectstranslating visual letters within the pseudo-words into

articulatory patterns and sound code. On the otherhand, the activation of the right superior parietallobule may be explained by the attentional demands ofletter detection in comparison to the rest condition(Corbetta et al., 1991).

The comparison between the four lexical conditionscombined and pseudo-words provides the functionalcorrelates of lexical-semantic information retrieval(Table 3): a left-hemispheric activation which includedthe prefrontal cortex and the parieto-occipital–tempo-ral junction (39/19). Activation in the same areas wasfound with different semantic tasks by Demonet et al.(1992) and by Vandenberghe et al. (1996). Similar areasare activated by semantic encoding in verbal memorytasks (Shallice et al., 1994). Petersen and Fiez (1993)have suggested that prefrontal activation may be re-lated to the active manipulation of semantic informa-tion. Activations near the temporoparietal junctionhave been found bilaterally, with a left hemisphericprevalence by other authors investigating differingaspects of lexical access (Petersen et al., 1989; Demonetet al., 1992; Zatorre et al., 1992; Vandenberghe et al.,1996). In addition, a focus was present in the posteriorcingulate cortex. This area has widespread corticalconnections (Pandya et al., 1981) and is part of anetwork related to directed attention for visual input(Mesulam, 1990). A comparable activation was alsoreported by Demonet et al. (1992) during a lexical-semantic task requiring the monitoring of a series of

FIG. 1. Combined lexical access conditions versus pseudo-words. Brain areas activated during the tasks rendered onto a MRI brain atlasin stereotaxic space. Red areas correspond to voxels significant at P , 0.001, Z 5 3.09 (see Tables for the stereotaxic coordinates of activationfoci).

354 CAPPA ET AL.

concrete nouns for semantic content. They suggest thatmultimodal representations of the words were accessedespecially via mental imagery. Indeed, this area mightbe involved in visual imagery (Rudge and Warrington,1991) and in memory (Valenstein et al., 1987; Rudgeand Warrington, 1991). Activations in the posteriorcingulate cortex, near or including the precuneus, havebeen found in other functional neuroimaging studiesimplicating both visual imagery and memory retrieval(Grasby et al., 1993; Fletcher et al., 1995).

The main aim of this study was to investigate class ofknowledge and category effects in semantic access fromlexical stimuli. Different patterns of cerebral activationwere found comparing retrieval of visual and associa-tive semantic knowledge from words. Both the indi-vidual comparisons of each of the semantic access taskswith the non-word baseline and the analysis of maineffects indicated significant differences in brain activa-tion related to the type of knowledge. Accessing visualperceptual knowledge was associated with bilateral

activation in inferior temporal, inferior parietal, andprefrontal areas. The latter finding is particularlyinteresting, as the prefrontal cortex has been repeat-edly suggested to play a crucial role in the retrieval andmanipulation of semantic knowledge. Not all semantictasks are associated with frontal activation; Thompson-Schill et al. (1997) have recently provided evidence thatthe left inferior prefrontal cortex is dependent of theselection of information among competing alternatives.The presence of longer RT during the visual tasks maybe compatible with the idea that a more complexselection of information from semantic memory mayhave been associated with the visual task. Posteriorcerebral activations have been associated with thestorage of semantic information (Fiez, 1997). Differ-ences in the activation of these areas became moreevident when considering category-related differences.Both the individual comparisons with baseline and thedirect contrast between living and nonliving entitiesshowed activations which were bilateral for the former,predominantly left hemispheric for the latter.

The present findings must be considered in theframework provided by neuropsychological studies ofsemantic disorders. Studies of patients with semanticmemory impairment are the main source of evidence(Warrington, 1975). Schematically, three main issues,

TABLE 4

Each Condition Separately Compared to Pseudo-words

x y z Z

A. Visual semantics

LivingL inferior frontal gyrus (Ba 45) 242 24 16 4.8L medial frontal gyrus (Ba 8) 216 28 40 4.0L precentral gyrus (Ba 6) 238 2 32 3.5L superior frontal gyrus (Ba 10) 222 54 16 3.3R parieto-occipital junction

(Ba 39/19) 40 278 28 4.0R inf frontal gyrus (Ba 47) 32 40 0 3.3R/L medial frontal gyrus (Ba 8) 22 14 44 4.1R/L posterior cingulate cortex

(Ba 31/23) 22 256 12 3.4Nonliving

L inferior frontal gyrus (Ba 45) 244 22 20 4.8L precentral gyrus (Ba 6) 240 8 28 3.9L medial frontal gyrus (Ba 8) 214 30 36 3.6L inferior temporal cortex (Ba 21/37) 246 260 28 3.4L middle temporal gyrus (Ba 21) 246 248 24 3.2L superior frontal gyrus (Ba 9) 218 50 20 3.4

B. Associative semantics

LivingL parieto-occipital junction

(Ba 39/19) 240 274 24 4.3L parieto-occipital junction(Ba 39/19) 234 280 32 3.6L medial frontal gyrus (Ba 8) 214 20 48 3.4R parieto-occipital junction

(Ba 39/19) 42 276 28 3.8R/L posterior cingulate (Ba 23/31) 22 262 20 4.5

NonlivingL inf frontal gyrus (Ba 45) 246 22 20 3.7L parieto-occipital junction

(Ba 39/19) 240 278 24 3.3L/R posterior cingulate (Ba 23/31) 22 262 20 3.4

TABLE 5

Main Effects

x y z Z

A. Task effects: Visual and associative semantics

Visual vs associativeL supra marginal gyrus (Ba 40) 244 250 36 5.7L inferior temporal cortex (Ba 37/20) 246 260 28 4.2L middle frontal gyrus (Ba 10) 226 52 12 4.4L middle frontal gyrus (Ba 9) 238 24 20 3.6L precentral gyrus (Ba 6) 240 22 32 4.5L anterior cingulate gyrus (Ba 32) 218 26 36 3.7L putamen 224 6 0 3.1R middle frontal gyrus (Ba 10) 32 46 8 4.3R supra marginal gyrus (Ba 40) 44 248 40 3.7R inferior temporal cortex (Ba 37/20) 56 260 212 3.5R precentral gyrus (Ba 6) 42 6 28 4.0R precentral gyrus (Ba 6) 44 14 24 3.8

Associative vs visualR/L posterior cingulate cortex

(Ba 31/23) 0 250 32 3.6

B. Category effects: Living and nonliving

Living vs nonlivingR middle frontal gyrus (Ba 46/10) 30 50 12 3.2R fusiform gyrus (Ba 37) 34 238 224 3.2

Nonliving vs livingL inferior temporal cortex (Ba 21/37) 248 264 28 4.0L supra marginal gyrus (Ba 40) 250 240 24 3.3R sup temporal gyrus (Ba 22) 52 216 16 3.3R thalamus 24 222 0 3.7

355SEMANTIC ACCESS: VISUAL WORD PROCESSING

all originating from clinical observations (Shallice,1988), have been the focus of investigation:

1. Category effects, i.e., differences related to thecategorical nature of the visual or verbal stimuluspromoting semantic access.

2. Class of knowledge effects, i.e., differences relatedto the type of semantic information accessed, pertain-ing to associative/functional vs perceptual (usuallyvisual) semantics.

3. Modality effects, i.e., differences in accessing se-mantic knowledge from pictorial or verbal material.

As discussed in the introduction, the debate withininformation processing models has been between single(‘‘amodal’’) semantics and multiple semantic systems.The former class of models (Riddoch et al., 1988;Caramazza et al., 1990) attributes the ‘‘modality ef-fects’’ to presemantic processing impairments, whilecategory and class of knowledge effects are supposed toreflect differences in the internal organization of theunique semantic system (Hillis and Caramazza, 1991).The multiple semantics models (McCarthy and Warring-ton, 1992), on the other hand, interpret all three effectsas the consequence of selective damage to a verbal orvisual semantic system. Another open issue is therelationship between effects 2 and 3, as the categoryeffect has been considered a by-product of a class ofknowledge effect, with the living category recognitionimpairment associated with a disproportionate impair-ment of visual semantics (Farah et al., 1996).

The effort to relate these issues to the neural sub-strate has been approached by means of clinical andanatomical correlations in patients and also functionalneuroimaging. The former line of investigation has metwith a limited success. The anatomical correlates ofmodality effects can be inferred from the differentiallesion sites in patients with optic aphasia and associa-tive agnosia (usually medial temporo-occipital lesionsin the left hemisphere—De Renzi, 1996), in comparisonwith patients with severe word comprehension disor-ders, such as transcortical sensory aphasics (temporo-parietal areas of the left convexity and underlyingwhite matter—Alexander et al., 1989). Semantic cat-egory effects in naming and comprehension have beenusually observed in patients with different pathologies,which are typically associated with distinct lesionlocations. The cases with impairment for biologicalentities have been generally affected by herpes simplexencephalitis and have shown the characteristic involve-ment of the limbic areas, extending to inferior temporalneocortex, usually in both hemispheres (see Gainottiand Silveri, 1996, for a discussion). Patients with thereverse dissociation have generally been aphasics withextensive lesions, affecting the perisylvian areas of theleft hemispheric convexity (Saffran and Schwartz, 1994).More precise correlations have been suggested by Dama-sio et al. (1996). Patients more impaired for animal

naming have been shown to have lesions involving theleft anterior inferior temporal cortex, while patientsmore impaired with tools had posterolateral inferiortemporal lesions extending to the temporo-occipitopari-etal junction. In a second study on the same population,Tranel et al. (1997) reported that bilateral lesions werenecessary for the occurrence of defective animal recog-nition, while left hemispheric damage was sufficient toimpair the recognition of tools. Since the issue of classof knowledge effects has been strictly intertwined withthat of category-specific disorders, similar correlationshave been suggested for these aspects, with inferiortemporal neocortical damage being associated with lossof visuoperceptual knowledge (Gainotti and Silveri,1996).

The results of this study, as well as of other func-tional imaging investigations, provide further relevantevidence. The present findings cannot be considered tospeak for or against modality effects, because thedifferences in the paradigm used in the present studyprevent a direct comparison with the investigationswhich have used pictures as an input for matching ornaming tasks. Concerning category effects, clear-cutdifferences according to semantic category have beenfound by Perani et al. (1995) with a picture-matchingtask and by Martin et al. (1996) and Damasio et al.(1996) with picture-naming tasks. Damasio et al. (1996)reported temporal lobe activations from a PET study inwhich subjects named from pictures famous people,animals, and tools. Naming famous people activated anarea in the left ventrolateral temporal pole, whilenaming animals and tools activated different areas inthe left posterior inferotemporal and left temporal pole,indicating separate cerebral representations involvedin the lexical retrieval of words from different catego-ries. The category-related differences were less promi-nent in the present study, suggesting that categorydifferences are more easily observed with pictorialrather than with lexical input and in conditions whichrequire lexical retrieval. However, the results were inagreement with the findings reported by Tranel et al.(1997), indicating a participation of both hemispheresin animal recognition and a predominant role of the lefthemisphere in the recognition of tools. Class of knowl-edge effects have been previously addressed only byVandenberghe et al. (1996). Only marginal differenceswere found in the latter study (left hippocampal regionand cerebellum being more active in the associativetask). The more extensive differences found in thepresent study suggest that the class of knowledge effectmight be amplified by tasks requiring self-generated(effortful) lexical access rather than in externally drivenmatching conditions. It is noteworthy that complemen-tary evidence for a fractionation of knowledge in thebrain is provided by a study by Martin et al. (1995) in

356 CAPPA ET AL.

which a word-generation task was used. Associating toa named object a color name activated the fusiformgyri, with a left-sided prevalence, while associating anaction name activated the left posterior middle andsuperior temporal gyri.

In conclusion, the overall picture of the neural corre-lates of access to semantic knowledge is not unexpect-edly complex. However, several general principles ap-pear to emerge, both from clinical investigations inpatients with selective disorders and from functional

investigations in normal subjects. Both semantic cat-egory and type of semantic information have beenshown to be associated with different patterns of cere-bral activation. However, the characteristics of the taskplay an important role in producing different pattern ofactivation: not only modality of input (picture or word),but also task requirements (matching or lexical re-trieval) are relevant variables, which appear to modu-late the network of cerebral areas involved in therecognition and naming of meaningful objects.

FIG. 2. (A and B) Main effects: task (visual and associative). (C and D) Main effects: category (living and nonliving). For details, see legendto Fig. 1.

357SEMANTIC ACCESS: VISUAL WORD PROCESSING

ACKNOWLEDGMENTS

This work was supported by grants from HFSP and CNR. We aregrateful to Felice Neutro and Pasquale Petta for technical assistance.

REFERENCES

Alexander, M. P., Hildbrunner, B., and Fisher, R. S. 1989. Distributedanatomy of transcortical sensory aphasia. Arch. Neurol. 46:885–892.

Beauvois, M. F. 1982. Optic aphasia: A process of interaction betweenperception and language. Philos. Trans. R. Soc. London B 298:35–47.

Bly, B. M., and Kosslyn, S. M. 1997. Functional anatomy of objectrecognition in humans: Evidence from positron emission tomogra-phy and functional magnetic resonance imaging. Curr. OpinionNeurol. 10:5–9.

Bookheimer, S. Y., Zeffiro, T. A., Blaxton, T., Gaillard, W., andTheodore, W. 1995. Regional cerebral blood flow during objectnaming and word reading. Hum. Brain Mapping 3:93–106.

FIG. 2—Continued

358 CAPPA ET AL.

Bub, D., Black, S. Hampson, E. and Kertesz, A. 1988. Semanticencoding of pictures and words: Some neuropsychological observa-tions. Cognit. Neuropsychol. 5:27–66.

Caramazza, A., Hillis, A. E., Rapp, B., and Romani, C. 1990. Themultiple semantics hypothesis: Multiple confusions? Cognit. Neuro-psychol. 8:161–190.

Caramazza, A. and Shelton, J. R. 1998. Domain-specific knowledgesystems in the brain: The animate–inanimate distinction. J.Cognit. Neurosci. 10:1–34..

Corbetta, M., Miezin, F. M., Dobmeyer, S., Shulman, G. L., andPetersen, S. E. 1991. Selective and divided attention during visualdiscrimination of shape, color, and speed: Functional anatomy bypositron emission tomography. J. Neurosci. 13:2383–2402.

Damasio, H., Grabowski, T. J., Tranel, D., Hichwa, R. D., andDamasio, A. R. 1996. A neural basis for lexical retrieval. Nature380:499–505.

De Renzi, E. 1996. Le agnosie. In. Manuale di Neuropsicologia (G. F.Denes and L. Pizzamiglio, Eds.), pp. 497–541. Zanichelli, Bologna.

Demonet, J. F., Chollet, F., Ramsay, S., Cardebat, D., Nespoulous, J.L., Wise, R., and Frackowiak, R. S. J. 1992. The anatomy ofphonological and semantic processing in normal subjects. Brain115:1753–1768.

Farah, M. J., Meyer, M. M., and McMullen, P. A. 1996. The living/nonliving dissociation is not an artifact: Giving an a priori implausiblehypothesis a strong test. Cognit. Neuropsychol. 13:137–154.

Fiez, J. A. 1997. Phonology, semantics and the role of the left inferiorprefrontal cortex. Hum. Brain Mapping 5:79–83.

Fletcher, P. C., Frith, C.D., Baker, S. C., Shallice, T., Frackowiak,R. S. J., and Dolan, R. J. 1995. The mind’s eye: Precuneusactivation in memory-related imagery. NeuroImage 2:195–200.

Friston, K. J., Ashburner, J., Frith, C. D., Poline, J.-B., Heather, J. D.,and Frackowiak, R. S. J. 1995a. Spatial registration and normaliza-tion of images. Hum. Brain Mapping 2:165–189.

Friston, K. J., Holmes, A. P., Worsley, K. J., Poline, J.-B., Frith, C. D.,and Frackowiak, R. S. J. 1995b. Statistical parametric maps infunctional imaging: A general linear approach. Hum. Brain Map-ping 2:189–210.

Frith, C. D., Kapur, N., Friston, K. J., Liddle, P. F., and Frackowiak,R. S. J. 1995. Regional cerebral activity associated with the inci-dental processing of pseudo-words. Hum. Brain Mapping 3:153–160.

Gainotti, G., and Silveri, M. C. 1996. Cognitive and anatomical locusof lesion in a patient with a category-specific semantic impairmentfor living beings. Cognit. Neuropsychol. 13:357–389.

Grasby, P. M., Frith, C. D., Friston, K. J., Bench, C., Frackowiak, R. S.J., and Dolan, R. J. 1993. Functional mapping of brain areasimplicated in auditory–verbal memory function. Brain 116:1–20.

Hillis, A. E., and Caramazza, A. 1991. Category-specific naming andcomprehension impairment:Adouble dissociation. Brain 114:2081–2094.

Martin, A., Haxby, J. V., Lalonde, F. M., Wiggs, C. L., and Ungerleider,L. G. 1995. Discrete cortical regions associated with knowledge ofcolor and knowledge of action. Science 270:102–105.

Martin, A., Wiggs, C. L., Ungerleider, L. G., and Haxby, J. V. 1996.Neural correlates of category-specific knowledge. Nature 379:649–652.

Mazziotta, J. C., Huang, S. C., Phelps, M. E., Carson, R. E.,MacDonald, N. S., and Mahoney, K. 1985. A non-invasive positroncomputed tomography technique using oxygen-15-labeled waterfor the evaluation of neurobehavioral task batteries. J. Cereb.Blood Flow Metab. 5:70–78.

McCarthy, R. A., and Warrington, E. K. 1988. Evidence for modality-specific meaning systems in the brain. Nature 334:428–430.

McCarthy, R. A., and Warrington, E. K. 1992. Cognitive Neuropsychol-ogy: A Clinical Introduction. Academic Press, New York.

Mesulam, M.-M. 1990. Large-scale neurocognitive networks anddistributed processing for attention, language, and memory. Ann.Neurol. 28:597–613.

Oldfield, R. C. 1971. The assessment and analysis of handedness: TheEdinburgh Inventory. Neuropsychologia 9:97–113.

Pandya, D. N., Van Hoesen, G. W., and Mesulam, M.-M. 1981.Efferent connections of the cingulate gyrus in the rhesus monkey.Exp. Brain Res. 42:319–330.

Paulesu, E., Frith, C. D., and Frackowiak, R. S. J. 1993. The neuralcorrelates of the verbal component of working memory. Nature362:342–345.

Perani, D., Cappa, S. F., Bettinardi, V., Bressi, S., Gorno Tempini, M.L., Matarrese, M., and Fazio, F. 1995. Different neural networks forthe recognition of biological and man-made entities. Neuroreport6:1637–1641.

Petersen, S. E., and Fiez, J. A. 1993. The processing of single wordsstudied with positron emission tomography. Annu. Rev. Neurosci.160:509–530.

Petersen, S. E., Fox, P. T., Posner, M. I., Mintun, M., and Raichle,M. E. 1988. Positron emission tomography studies of the corticalanatomy of single-word processing. Nature 331:585–589.

Petersen, S. E., Fox, P. T., Posner, M. I., Mintun, M., and Raichle,M. E. 1989. Positron emission tomographic studies of the process-ing of single words. J. Cognit. Neurosci. 1:153–170.

Riddoch, M. J., Humphreys, G. W., Coltheart, M., and Funnel, E.1988. Semantic system or systems? Neuropsychological evidencere-examined. Cognit. Neuropsychol. 5:3–25.

Rudge, P., and Warrington, E. K. 1991. Selective impairment ofmemory and visual perception in splenial tumours. Brain 114:349–360.

Saffran, E. M., and Schwartz, M. F. 1994. Of cabbages and things:Semantic memory from a neuropsychological perspective—A tuto-rial review. In Attention and Performance. XV. Conscious andUnconscious Processes C. Umilta and M. Moscovitch, Eds.). MITPress, Cambridge, MA.

Shallice, T. 1988. From Neuropsychology to Mental Structure. Cam-bridge Univ. Press, Cambridge, UK.

Shallice, T., Fletcher, P., Frith, C. D., Grasby, P., Frackowiak, R. S. J.,and Dolan, R. J. 1994. Brain regions associated with acquisitionand retrieval of verbal episodic memory. Nature 368:633–635.

Talairach, J., and Tournoux, P. 1988. Co-planar Stereotaxic Atlas ofthe Human Brain. Thieme, Stuttgart.

Thompson-Schill, S., D’Esposito, M., Aguirre, G. K., and Farah, M. J.1997. Role of left inferior prefrontal cortex in retrieval of semanticknowledge: A reevaluation. Proc. Natl. Acad. Sci. USA 94:14792–14797.

Tranel, D., Logan, C. G., Frank, R. J., and Damasio, A. R. 1997.Explaining category-related effect in the retrieval of conceptualand lexical knowledge for concrete entities: Operationalization andanalysis of factors. Neuropsychologia 35:1329–1339.

Valenstein, E., Bowers, D., Verfaellie, M., Heilman, K. M., Day, A.,and Watson, R. T. 1987. Retrosplenial amnesia. Brain 110:1631–1646.

Vandenberghe, R., Price, C., Wise, R., and Josephs, O., and Frackow-iak, R. S. J. 1996. Functional anatomy of a common semanticsystem for words and pictures. Nature 383:254–256.

Warrington, E. K. 1975. The selective impairment of semanticmemory. Q. J. Exp. Psychol. 27:635–657.

Warrington, E. K., and Shallice, T. 1984. Category-specific semanticimpairment. Brain 107:829–854.

Zatorre, R. J., Evans, A. C., Meyer, E., and Gjedde, A. 1992.Lateralization of phonetic and pitch discrimination in speechprocessing. Science 256:846–849.

359SEMANTIC ACCESS: VISUAL WORD PROCESSING