OUP user menu

Identification of famous faces and buildings
A functional neuroimaging study of semantically unique items

M. L. Gorno-Tempini, C. J. Price
DOI: http://dx.doi.org/10.1093/brain/124.10.2087 2087-2097 First published online: 1 October 2001

Summary

Several functional imaging experiments have clearly established that the fusiform gyri are preferentially responsive to faces, whereas the parahippocampal/lingual gyri are more responsive to buildings. Other studies have demonstrated that famous faces additionally activate the anterior temporal cortex relative to unfamiliar faces, animals, tools, body parts and maps. One explanation for this apparent specialization for known people could be that famous faces are `semantically unique items'. In other words, they carry unique semantic associations that are not shared by other perceptually similar category members. If this hypothesis is correct, the anterior temporal cortex should also respond to other semantically unique items, such as famous buildings. In this PET study, we investigated the effect of fame (famous relative to non-famous) on activation elicited by famous and non-famous faces and buildings during a same–different matching task. We found that, when the task was held constant, category-specific activations in the fusiform and parahippocampal/lingual areas were not modulated by fame. In contrast, in the left anterior middle temporal gyrus there was an effect of fame that was common to faces and buildings. These results suggest that the identification of famous faces and buildings involves category-specific perceptual processing in the fusiform and parahippocampal/lingual regions, respectively, and shared analysis of unique semantic attributes in the left anterior temporal cortex.

  • faces
  • buildings
  • semantic and lexical processes
  • PET
  • fusiform
  • BA = Brodmann area
  • FB = famous buildings
  • FF = famous faces
  • FFA = fusiform face area
  • MTG = middle temporal gyrus
  • NFB = non-famous buildings
  • NFF = non-famous faces
  • RT = reaction time
  • SB = scrambled buildings
  • SF = scrambled faces

Introduction

Psychological studies have suggested that the task of fully identifying and naming a famous person is achieved by a cascade of sequential processing stages (Bruce and Young, 1986). Neuropsychological evidence supports this view, and patients have been described with impairments at various stages of the identification process, including: (i) the pre-semantic stage, when recognition of famous faces is impaired only in the visual domain, i.e. prosopagnosia; (ii) the semantic stage, when loss of biographical information about known people (person-specific semantics) occurs regardless of the stimulus modality; and (iii) the post-semantic lexical retrieval stage, when name retrieval is impaired but semantic information is retrieved correctly, i.e. proper name anomia.

The issue of whether these deficits reflect the existence of face- or person-specific cognitive modules has been debated since the earliest reports of prosopagnosia. In some cases, the impairment appeared to be so strictly selective for either faces (De Renzi, 1986; De Renzi et al.1991; Sergent and Signoret, 1992; McNeil and Warrington, 1993; Farah et al.1995), person-specific semantics (Hanley et al.1989; Evans et al.1995) or people's proper names (McKenna and Warrington, 1980; Lucchelli and De Renzi, 1992) that the existence of dedicated cognitive modules seemed possible. In other cases, prosopagnosia extended to impaired recognition of specific objects that, like faces, have many visually similar neighbours, e.g. breeds of dogs, types of flowers or cars (Lhermitte and Pillon, 1975; Damasio et al.1982, 1990), individual animals (Bornstein et al.1969; Assal et al.1984), buildings and landmarks (Pallis, 1955; Landis et al.1986). Similarly, patients with deficits in retrieving person-specific semantics or proper names can also have difficulties with other objects that, in addition to being part of a perceptually homogeneous category, also have unique semantic and lexical associations. These stimuli are referred to as `semantically unique items'. For example, a patient might be able to recognize a stimulus as a dog but be unable to identify it as Lassie, the collie that starred in the popular TV series (Ellis et al.1989). Other examples of semantically unique items include famous buildings and landmarks. These stimuli have often been reported as impaired not only in prosopagnosia (see above) but also in patients with person-specific deficits at the semantic (Ellis et al.1989; Kartsounis and Shallice, 1996; McCarthy et al.1996) and lexical (Semenza and Zettin, 1988, 1989) levels. Such co-occurring deficits are not consistent with the existence of person-specific modules, but do support the hypothesis that specific deficits can arise from the different demands that faces and proper names place on the object- and word-processing systems. First, they need to be distinguished from other, perceptually similar category members. Then they need to be linked to unique semantic and lexical features.

The debate over the extent to which face-processing engages specific modules has recently extended into the functional neuroimaging literature. Most studies of face-specificity have focused on the perceptual level of processing and have used unfamiliar faces that cannot be linked to specific semantic or lexical representations. Viewing and matching unfamiliar faces relative to other categories of objects consistently activate a region of the lateral fusiform gyrus bilaterally, but more consistently on the right, that has recently been labelled the `fusiform face area' [FFA; mean Talairach coordinates across subjects: x = 40, y = –55, z = –10 (Kanwisher et al.1997)]. However, the response in the FFA is not exclusive to faces (Ishai et al.1999) and the same area also responds to animals (Chao et al.1999) and animal faces (Maguire et al., 2001) and is activated when visually similar objects are categorized at the subordinate level, e.g. when distinguishing different types of birds and cars, especially when the subject is an expert (Gauthier et al.1999, 2000). Interestingly, pictures of buildings, which also need a subordinate level of perceptual categorization, activate a more medial region in the ventral temporo-occipital cortex when compared with other categories of objects. This region has been reported to be located in the right anterior lingual sulcus (coordinates in Talairach space: x = 20.6 ± 5.0, y = 53.8 ± 6.7, z = –9.2 ± 3.9) (Aguirre et al.1998), bilateral medial fusiform (x = –26, y = –57, x = –14 and x = 28, y = –57, z = –13) (Ishai et al.1999) and parahippocampal gyri (x = 20, y = –39, z = –5 and x = –28, y = –39, z = –6) (Epstein et al.1999). Even if the peak activations in these studies were relatively close, it has recently been suggested that, while the more anterior parahippocampal region might be more involved in processing spatial layouts, such as rooms and scenes (Epstein and Kanwisher, 1998; Epstein et al.1999), the more posterior fusiform/lingual region might be more involved in processing single buildings (Haxby et al.2000; Maguire et al., 2001). We will refer to this building-responsive area as the `parahippocampal/lingual' region.

To summarize, the currently available imaging and neuropsychological data indicate that faces and buildings place different demands on perceptual, presemantic processing, although the specificity of these responses is not absolute. The neuroimaging study reported in this paper focuses on the identification of famous faces and buildings. This is an area that has received little attention in the neuroimaging literature and few studies have investigated brain responses to either familiar (friends or associates), newly learned (seen once or twice before scanning) or famous (well-known celebrities) faces or buildings. Therefore, many issues are still debated. For instance, the effect of fame on the FFA and parahippocampal/lingual activations has not yet been fully established. Differential responses to famous or familiar faces in the FFA have been confounded by the task and/or attentional demands (Sergent et al.1992; Kim et al.1999; Leveroni et al.2000). When the cognitive task was held constant, we found no effect of fame on the fusiform face activation (Gorno-Tempini et al.1998). Similarly, the effect of fame on the parahippocampal/lingual activation remains unclear. Responses to famous buildings have never been investigated and only one study has compared familiar (campus landmarks for students) and unfamiliar landmarks (Epstein et al.1999). However, in this study, the analysis was confined to the `parahippocampal place area' (a region that responds to scenes more than objects and buildings) and no other brain areas were investigated.

Other studies investigating activation for famous faces have highlighted the anterior temporal cortices, either bilaterally (Sergent et al.1992; Damasio et al.1996; Leveroni et al.2000) or on the left only (Gorno-Tempini et al.1998, 2000; Henson et al.2000). For instance, the left anterior middle temporal gyrus (MTG) was more active for: (i) naming pictures of famous faces (well-known celebrities) relative to five other categories of objects (animals, artefacts, body parts, maps and colours) (Gorno-Tempini et al.2000); and (ii) matching famous proper names (e.g. Marilyn Monroe) relative to object names constructed from two words (e.g. wheel- barrow) (Gorno-Tempini et al.1998). However, this region has not been observed for familiar (friends and associates) relative to unknown faces, which activated the right temporal pole (Nakamura et al.2000).

Several questions therefore remain. First, are the fusiform and parahippocampal/lingual regions modulated by fame? Secondly, does the left anterior MTG respond to famous faces (e.g. Prince Charles) more than famous buildings (e.g. Buckingham Palace)? Thirdly, is the right anterior temporal cortex responsive to faces and buildings and is it also influenced by fame? To address these questions, we engaged subjects in a same–different judgement on visually presented pairs of pictures depicting famous, non-famous and scrambled faces and buildings. We chose universally famous items rather than personally familiar stimuli because semantic and lexical associations should be more stable and less influenced by personal experience and individual differences. We chose the same–different matching task because it allowed us to keep the task constant while famous, non-famous and scrambled stimuli were compared with each other. The same task and face stimuli were used in our previous study (Gorno-Tempini et al., 1998). Therefore, when investigating the effect of fame, which was manipulated in both experiments, we combined the old and new data in a single statistical analysis (see Methods).

We predicted that (i) the FFA and the parahippocampal/lingual regions would be more active for faces than buildings (Kanwisher et al.1997; Aguirre et al.1998; Chao et al.1999; Ishai et al.1999; Maguire et al., 2001); (ii) the anterior temporal cortex would be more responsive to famous faces than to non-famous faces and scrambled stimuli (Gorno-Tempini et al.1998); and (iii) the FFA response would not be affected by fame (Gorno-Tempini et al.1998).

Our open questions were (i) whether the parahippocampal/lingual area would respond differently to famous and non-famous buildings; and (ii) whether the left and/or right anterior temporal cortex would show a common or differential response to famous faces and famous buildings, when compared with their non-famous counterparts.

Methods

Subjects

Fifteen male subjects (age range 18–38 years) took part in the experiment. They were all right-handed, native English speakers, healthy, on no medication and free from any history of neurological or psychiatric illness. Subjects gave informed, written consent to the study, which was approved by the local hospital ethics committee and the Administration of Radioactive Substances Advisory Committee (UK).

Experimental design and procedure

Each subject underwent 12 PET relative perfusion scans while deciding whether two visual stimuli, displayed simultaneously as a pair in the centre of a computer screen, were the same or different. Whereas the task was constant across conditions, the stimuli were pairs of famous or non-famous faces or buildings. The experiment was therefore of a 2 × 2 factorial design, with category (faces and buildings) and fame (famous and non-famous) as the two factors. A low-level visual control condition was also introduced, in which subjects had to perform a same–different task on pairs of scrambled stimuli. Therefore, six experimental conditions were created: famous faces (FF), non-famous faces (NFF), scrambled faces (SF), famous buildings (FB), non-famous buildings (NFB) and scrambled buildings (SB). Subjects saw 10 stimulus pairs (six same and four different) per scan and responded with a key-press: right button for same and left button for different pairs. Reaction times (RTs) and accuracy were recorded. Each pair of stimuli was presented for 5 s at the rate of one stimulus every 6 s. Each of the six conditions was repeated twice with different stimuli. There were no stimulus repetitions within subjects and condition order was counterbalanced between and within subjects. Since all stimuli were novel to the subjects at the time of scanning, after the PET session, the famous faces and buildings were shown again to ensure that correct identification had occurred.

Stimuli

The famous faces came from a pool of 200 black-and-white photographs of celebrities in different professional categories. Their familiarity was determined by a behavioural study conducted on 20 normal male subjects (age range 18–33 years) who were shown each face on a computer screen for 5 s and had to name the person. Only those faces that were named within the 5 s by at least 19 subjects were included in the PET study. The same procedure was used to choose the famous buildings. Twenty normal male subjects (age range 20–35 years) viewed 120 black-and-white pictures of famous buildings and landmarks and only stimuli that were named within 5 s by at least 19 subjects were used during the scanning sessions. The names of the famous stimuli are listed in Appendix I.

The non-famous faces were matched to the famous faces for mean age, sex and facial expression. The non-famous buildings were chosen to match their famous counterparts for general category (e.g. churches and arches). For instance, the non-famous stimulus corresponding to the Leaning Tower of Pisa was an unknown tower (Fig. 1).

Fig. 1

Examples of the stimuli used in Experiments 1 and 2 for the face conditions and in Experiment 2 for the building conditions.

The scrambled control stimuli were obtained by scrambling each face and building stimulus in the same way. To maintain a constant spatial frequency power density spectrum in the scrambled stimuli, the manipulation was on the phases of each spatial frequency in the image. The phase of each lower frequency component, starting from the lowest frequency, was swapped with the phase of a corresponding higher-frequency component, starting with the highest. A pattern was obtained that was no longer recognizable as a face.

Faces were framed with a black oval mask to avoid differences in the picture background. The same masking procedure eliminated as much as possible of the surrounding scene from the building stimuli. The framed faces and buildings were then paired to obtain a single stimulus that measured 9.4 × 13.6 cm. The pairs comprised two stimuli displayed either one next to the other on a black background (all faces and half of the building pairs) or one above the other (half of the building pairs). The corresponding scrambled stimuli were framed and paired similarly. Since the building and face conditions were never compared directly without also being compared with their respective scrambled stimuli, the different rotation was matched across type of stimulus. The buildings, faces and scrambled conditions were also matched for mean luminance.

PET and MRI scanning

Each subject underwent 12 PET relative perfusion scans over a period of 2 h. Scans were obtained using a Siemens/CPS ECAT Exact HR+ (model 962) PET scanner (Siemens/CTI, Knoxville, Tenn., USA) with collimating septa retracted. Participants received a 20 s intravenous bolus of H2150 at a concentration of 55 MBq/ml and a flow rate of 10 ml/min through a forearm cannula. For each subject, a T1-weighted structural MRI was obtained with a 2 T Magnetom Vision scanner (Siemens, Erlangen, Germany).

Data analysis

Data were analysed by statistical parametric mapping, using SPM99 software from the Wellcome Department of Cognitive Neurology, London, UK (http//www.fil.ion.ucl.ac.uk/spm) implemented in Matlab (Mathworks. Sherborn, Mass., USA) using standardized procedures (Friston et al.1995a, b), including realignment for head movements, spatial normalization to the Montreal Neurological Institute template brain (Cocosco et al.1997) in the space of Talairach and Tournoux (Talairach and Tournoux, 1988) and smoothing. The smoothing kernel was a 3D Gaussian filter of 16 mm. Condition and subject effects were estimated according to the general linear model at each voxel. Our design matrix specified two groups of subjects. One comprised nine subjects who performed the same–different matching paradigm described above on face (FF, NFF and SF) and building (FB, NFB and SB) stimuli. The other group comprised six subjects who performed the same–different matching task on the face but not the building stimuli. The data from the group of six subjects have been reported previously (Gorno-Tempini et al.1998) and are referred to here as Experiment 1. The new data (group of nine subjects) are referred to as Experiment 2. Both were included in the present study to (i) increase sensitivity to the effect of fame; and (ii) allow comparisons to be made between studies. However, differential effects of category and category × fame interactions were estimated using data from Experiment 2 only.

Three main comparisons were performed in order to identify regions that were commonly or differentially activated by category (faces or buildings) or fame (famous or non-famous).

Main effect of category irrespective of fame (Experiment 2 only)

The effect of faces relative to buildings was identified with a conjunction of FF–FB and NFF–NFB. Using the inclusive masking option in SPM99, we also eliminated any areas that were not active for FF–SF or NFF–SF at P < 0.001. The effect of buildings relative to faces was based on a conjunction of the reversed contrasts (FB–FF and NFB–NFF), inclusively masked with FB–SB and NFB–SF at P < 0.001.

Once the face- and building-specific areas had been identified, we also investigated the effect of fame within these areas, i.e. FF–NFF and NFF–FF in the face area and FB–NFB and NFB–FB in the building area.

Main effect of fame irrespective of category (Experiment 1 and 2)

This was identified with a conjunction of three contrasts: (i) FF–NFF (Experiment 1); (ii) FF–NFF (Experiment 2); and (iii) FB–NFB (Experiment 2). Inclusive masking, with FF–SF (Experiment 1 and 2) and FB–SB (Experiment 2) at P < 0.001, ensured that activation was also present above controls in both experiments.

The same procedure was performed in order to identify regions specific to non-famous stimuli.

Activations specific to famous faces or famous buildings (Experiment 2 only)

These were identified using the inclusive masking procedure to find areas that were activated for (i) FF–NFF; (ii) FF–SF; and (iii) the interaction of faces and fame (FF–NFF) – (FB–NFB).

Likewise, regions more involved in processing FB were those more active for (i) FB–NFB; FB–SB; and (ii) the interaction of fame with buildings (FB–NFB) – (FF–NFF).

Thresholds for both experiments

We accepted a level of significance of P < 0.05 corrected for multiple comparisons for all our main contrasts, except for the interaction, which we report at P < 0.001 uncorrected. When claiming the absence of an effect, we lowered the threshold to P > 0.1 uncorrected.

Results

Behavioural data

All subjects performed the experimental task at ceiling level, making fewer than three errors. Reaction time data were collated for each experimental condition and any score that deviated >2.5 SD from the mean was replaced with the mean of that condition. A factorial ANOVA (analysis of variance) was used to identify main effects of category (faces and buildings), fame (famous and non-famous) and the interaction between these variables (Fig. 2). No significant effect was found, indicating no difference in difficulty and attentional demands between the four experimental conditions. However, when a one-way ANOVA including the scrambled condition as a variable was performed, a significant difference in RTs was found (P < 0.05). A post hoc Scheffé test indicated that same–different matching for the scrambled conditions took significantly longer than for the faces or buildings. Reaction times for FF, NFF and SF in Experiment 1 have been reported elsewhere (Gorno-Tempini et al., 1998). There was no significant difference between the FF and NFF, but matching took significantly longer for scrambled stimuli than for both the face conditions. Thus, the behavioural data were consistent across experiments.

Fig. 2

Corrected means and standard errors of the RTs for each of the six experimental conditions (Experiment 2)

After scanning, subjects were presented with each face or building from the famous set and asked to name it within 5 s. No subject named fewer then 26 out of 28 faces and 25 out of 28 buildings. This confirmed that subjects were able to establish semantic and lexical associations for the famous stimuli.

Scanning data

Main effect of category irrespective of fame

Faces versus buildings.

When famous and non-famous faces were compared with the corresponding building conditions, a region in the right fusiform gyrus [Brodmann area (BA) 37/20], corresponding to the FFA, was activated at a corrected level of significance (P < 0.05) (Table 1). Both FF and NFF evoked activation in the FFA and there was no significant effect of FF–NFF or NFF–FF (P > 0.1). Figure 3A illustrates that the FFA response to buildings (relative to controls) was very small (P < 0.05 uncorrected).

View this table:
View this table:
Table 1

Coordinates and Z values of the brain regions identified by our main analysis (see Results)

ContrastArea (BA)Talairach coordinatesZ
xyz
*P < 0.001 uncorrected.
Faces > buildingsRight fusiform (37/20) 48−48−224.9
Buildings > facesBilateral parahippocampus/−26−44−146.0
lingual/fusiform (36/19/37)−26−38−185.0
−34−48−124.5*
28−48−125.5
30−40−164.5*
FF–NFF + FB–NFBLeft MTG (21) 64 0−164.7
FF–NFF > FB–NFBRight MTG (21) 62 −2−143.2*
Fig. 3

From top to bottom, this figure illustrates areas of activation and parameter estimates for regions that were more activated for (A) faces than for buildings, (B) buildings than for faces, (C) famous than for non-famous faces and buildings, and (D) famous than for non-famous faces only. In the left column, all activations are superimposed on axial slices of the mean of the nine subjects' normalized structural MRIs and thresholded at P < 0.001 (uncorrected). In the right column, the plots indicate the value of the normalized regional cerebral blood flow at the indicated voxel (y-axis) for each of the experimental conditions in Experiments 1 and 2 (x-axis).

Buildings versus faces.

When FB and NFB were compared with the corresponding face conditions, an area in the medioventral occipitotemporal cortex was activated bilaterally (BA 36/19/37) at a corrected level of significance (P < 0.05). The peak of the activation was in the superior lingual gyrus but the cluster spread to the parahippocampus and, on the left side, also to the medial fusiform gyrus. Because of the resolution and smoothing of the PET data, we will refer to this activation as the parahippocampus/lingual region. This area was activated by both FB and NFB and there was no effect of FB–NFB or NFB–FB (P > 0.1 uncorrected). Figure 3B illustrates that the response in these regions was almost exclusive to buildings, with no effect of faces versus controls (P > 0.9 uncorrected).

Main effect of fame irrespective of category

Famous versus non-famous stimuli.

The conjunction analysis demonstrated that only one region showed a significant effect of fame (P < 0.05 corrected). This was the left anterior MTG (BA 21; P < 0.05 corrected; 240 voxels). The high level of significance of the conjunction analysis reflects the consistency of this activation across contrasts (FF–NFF and FB–NFB) and across experiments (Experiments 1 and 2). Figure 3C illustrates an equivalent effect of famous versus non-famous faces and buildings in Experiments 1 and 2 (P < 0.005 to P < 0.001). Furthermore, in this region there was no interaction between fame and category (P > 0.1 uncorrected).

Non-famous versus famous stimuli.

No significant effect was found for non-famous versus famous stimuli.

Areas specific to famous faces or famous buildings

A small region in the right anterior MTG (BA 21) showed a significant interaction (P < 0.001 uncorrected). In this area there was a small effect of FF–NFF (Z = 2.7; P < 0.01 uncorrected) but no effect of FB–NFB (P > 0.1 uncorrected) (Fig. 3D). Figure 3 illustrates that in this region there was also a small effect of FF versus NFF in Experiment 1 (Z = 2.4; P < 0.005 uncorrected) and the conjunction of FF–NFF in Experiments 1 and 2 raised the joint significance of this effect to P < 0.001 uncorrected (Z = 3.2).

Summary of results

As predicted, we demonstrated that (i) the FFA and the parahippocampal/lingual region were more active for faces and buildings, respectively; (ii) the anterior MTG was more active for famous faces relative to non-famous and scrambled faces; and (iii) there was no effect of fame on the FFA. In addition, we found that (i) there was no effect of fame on the parahippocampal/lingual region; and (ii) the left anterior MTG was equally active for famous faces and famous buildings. In the right anterior MTG, there was a small effect of fame for faces but not for buildings.

Discussion

In this study, we investigated brain responses to famous and non-famous faces and buildings in order to determine whether (i) the fusiform and parahippocampal/lingual regions, which have been associated previously with processing of unfamiliar faces and buildings, would respond differently to famous and non-famous stimuli when the cognitive task was kept constant; and (ii) the left and right anterior temporal regions, previously associated with famous face naming, would be commonly or differentially responsive to another category of unique items: famous buildings. The results show category-specific effects in the right fusiform and bilateral parahippocampal/lingual gyri for faces and buildings, respectively, but no effect of fame. In contrast, the left anterior MTG showed an effect of fame for both faces and buildings, but no effect of category. The responses of the right MTG were less clear-cut, with only a small effect of fame for faces and no effect for buildings. We first discuss the anterior temporal and then the fusiform and parahippocampal/lingual activations.

Anterior temporal activations

Many previous functional neuroimaging studies have demonstrated that activation in the left anterolateral portions of the middle and inferior temporal gyri (BA 21/20) increases with the demands on semantic processing, particularly when specific information must be retrieved. For instance, this occurs when subjects categorize objects (Devlin et al.2000) and match words according to specific semantic features (Mummery et al.1998) or associations (Vandenberghe et al.1996). Interestingly, the left MTG is also activated when famous faces and proper names are matched (Gorno-Tempini et al.1998), with greater activation for naming famous faces than objects from other categories (Damasio et al.1996; Gorno-Tempini et al.2000). The naming and matching of objects are usually associated with activation in more posterior regions of the left inferior or middle temporal cortex (Bookheimer et al., 1995; Martin et al., 1995; Perani et al., 1995; Damasio et al., 1996; Moore and Price, 1999). Our question concerns why the left anterior temporal lobe is more active for the naming of famous faces than for object-naming. One explanation might be that there is a discrete region in the left anterior temporal lobe that is specific to person-specific semantic or lexical attributes. This would be consistent with patients having anterior temporal damage and loss of person-specific semantics, but not with the neuroimaging studies that have shown activation of the anterior MTG when retrieving specific semantic features related to objects. A more likely explanation is that naming or matching famous faces elicits more activation than object-naming in areas associated with the retrieval of semantic features. This might occur because the semantic associations evoked by famous faces are unique and are not shared by other items of the same category. We tested this hypothesis by measuring the response to famous buildings, which, to be correctly identified, also need to evoke specific semantic and lexical associations that are not shared with other category members. We found equivalent responses for famous faces and famous buildings in the left anterior MTG, thereby confirming that matching semantically unique items increased the demand on a semantic processing area, even when subjects were not explicitly required to identify them. However, when normal subjects were exposed to a very famous stimulus, identification and lexical retrieval occurred together automatically. Therefore, the anterior left MTG may also play a role in linking semantic and lexical information and we cannot exclude the possibility that this region is involved in pure lexical retrieval processes (Damasio et al.1996). Future studies, combining neuropsychological and neuroimaging, are necessary to solve this issue (Gorno-Tempini et al., 2001) and to investigate whether responses could be affected by familiarity or the age of acquisition with the object/face concepts.

We were also interested in how the right anterior temporal cortex responded to famous faces and buildings. Indeed, neuropsychological patients with person-specific semantic deficits often suffer from anterior temporal pathologies, such as epilepsy, neurodegenerative diseases and herpes simplex encephalitis, which are most commonly bilateral. Furthermore, patients have been described with mainly left, bilateral or mainly right damage (De Renzi et al.1987; Ellis et al.1989; Hanley et al.1989; Evans et al.1995; Kroll et al.1997; Hodges and Graham, 1998; Kitchener and Hodges, 1999), again suggesting that person-specific semantic knowledge is distributed bilaterally. However, although three functional neuroimaging studies have shown bilateral (Sergent et al.1992; Leveroni et al.2000) responses to famous faces, activations have been left-lateralized in others (Gorno-Tempini et al.1998, 2000; Henson et al.2000). The only right-lateralized anterior temporal activation was found for familiar faces and scenes (Nakamura et al.2000). In our study, responses in the anterior temporal cortex were much weaker in the right than the left hemisphere and were only observed for famous faces. It is possible that the differential anterior temporal lateralization in our study and that of Nakamura and colleagues (Nakamura et al.2000) is determined by the degree of familiarity with the stimuli, which influences the amount of semantic and lexical information that can be retrieved. The faces and buildings in our experiment were famous and well known to each subject, whereas in the study of Nakamura and colleagues the degree of familiarity was not controlled across subjects or stimuli. Another possibility is that semantic information can be retrieved by activation of either the left or the right anterior temporal cortex. This would explain why semantic impairments do not usually occur when anterior temporal damage is limited to one hemisphere. Further evidence, combining functional imaging and lesion studies, is therefore required to determine the different roles of the left and right anterior temporal lobes.

Fusiform and parahippocampal/lingual activations

Converging evidence has indicated that unfamiliar faces and scenes or buildings evoke differential responses in the fusiform and parahippocampal/lingual gyri when compared with each other and with other categories of objects (Aguirre and D'Esposito, 1997; Kanwisher et al.1997; Aguirre et al.1998; Chao et al.1999; Epstein et al.1999; Epstein and Kanwisher, 1998, 2000; Ishai et al.1999; Kanwisher et al.1999; Nakamura et al.2000; Maguire et al., 2001). We confirmed these results by showing a preferential response to faces in a region in the right fusiform gyrus and an almost exclusive response to buildings in bilateral parahippocampal/lingual areas. The results of the present study also demonstrate that, when cognitive task and attentional load were controlled (no difference in RT between conditions), both these category-specific regions were unaffected by fame. Previous studies that found a differential effect of fame in the FFA did not control for task (Sergent et al.1992) or used passive viewing (Henson et al.2000) or familiarity decision (Leveroni et al.2000) paradigms, in which it is likely that subjects would have been more engaged in the task when the stimuli were familiar. Indeed, it has been established clearly that the FFA is modulated by attentional demands (Wojciulik et al.1998). In contrast, we demonstrate that it is unaffected by fame when subjects attend to the perceptual features of unfamiliar as well as famous faces.

Our results are consistent with a pre-semantic perceptual role for the FFA in processing faces and for the parahippocampal/lingual region in processing buildings. Although the identification of the specific mechanism responsible for these activations is beyond the aim of our study (for discussion, see Gauthier, 2000; Haxby et al.2000; Kanwisher, 2000), the role of the FFA in the perceptual stage of the face-recognition process is consistent with lesion studies. In fact, temporo-occipital lesions, which include the FFA, usually cause the `apperceptive' type of prosopagnosia, where the defective recognition of familiar faces is accompanied by defective processing of unfamiliar faces (De Renzi et al.1991). On the other hand, more anterior lesions in the temporal lobes are more likely to cause the `associative' type of prosopagnosia, in which the deficit seems to be selective for familiar faces (Gentileschi et al.1999) and can also progress to the loss of person-specific semantics (Evans et al.1995; Kitchener and Hodges, 1999). A similar argument could apply to the parahippocampal region. This area and the surrounding fusiform/lingual regions have been shown previously to respond preferentially to stimuli that depict spatial and topographical information about the environment, e.g. rooms, buildings, houses, scenes and landscapes. Our results support the hypothesis that, within this region, the more posterior lingual focus is the most responsive to single buildings (Haxby et al.2000). The fact that the parahippocampal/lingual area did not respond differently to famous and non-famous buildings suggests a presemantic perceptual role in the analysis. This would then be followed by further analysis in more anterior regions, depending on the task required, for instance, in the right hippocampus for active space navigation (Maguire et al.1998) and the left anterior temporal cortex for retrieval of semantic information.

In conclusion, we demonstrate that distinct regions within the ventral occipitotemporal cortex are involved in the category-specific perceptual processing of face and building stimuli. In contrast, we found that a semantic area in the left anterior MTG was engaged by both famous faces and famous buildings. Future studies are necessary to establish unequivocally the role of this region in semantic and/or lexical retrieval processes.

Appendix I

Acknowledgments

This work was funded by The Wellcome Trust.

References

View Abstract