Naming of newly learned objects: A PET activation study

Cognitive Brain Research 25 (2005) 359 – 371 www.elsevier.com/locate/cogbrainres

Research Report

Naming of newly learned objects: A PET activation study Petra Gro¨nholma,*, Juha O. Rinneb, Victor Vorobyevc, Matti Lainea,c a

Department of Psychology, A˚bo Akademi University, FIN-20500 A˚bo, Finland b Turku PET-Centre, PO Box 52, FIN-20521 Turku, Finland c Centre for Cognitive Neuroscience, University of Turku, FIN-20014 Turku, Finland Accepted 29 June 2005 Available online 10 August 2005

Abstract The present study tracked the naming-related brain activity by positron emission tomography (PET) when successfully learned unfamiliar objects were named. Ten Finnish-speaking subjects participated in the study. Prior to the PET scan, each subject underwent a 4-day long training period in which 40 names of rare unfamiliar objects were taught. The stimulus categories were as follows: unfamiliar but real objects for which both the name and the definition were given during training, only the name was given, no information was given. In addition, familiar objects and visual noise patterns were used. The unfamiliar items mainly represented ancient domestic tools unknown to modern-day people. As semantic support did not affect the PET results, all trained items were pooled together. The trained objects vs. familiar objects contrast revealed rCBF increases in the left inferior frontal cortex (Broca’s area), the left anterior temporal area, and the cerebellum. Likewise, the trained objects vs. unfamiliar objects (for which no information was given) contrast revealed more extensive left frontal (roughly Broca’s area) and cerebellar rCBF increases, while anterior temporal activation was bilateral. Familiar objects, contrasted with both visual noise patterns and a rest condition, elicited activation increases in expected areas, i.e., bilateral occipital regions and the fusiform gyrus. Our results indicate that the naming of newly learned objects recruits more extensive brain areas than the naming of familiar items, namely a network that includes left-dominant frontotemporal areas and cerebellum. Its activity is tentatively related to enhanced lexical – semantic and lexical – phonological retrieval, as well as associative memory processes. D 2005 Elsevier B.V. All rights reserved. Theme: Neural basis of behavior Topic: Cognition Keywords: Word learning; Lexical retrieval; Object naming; Positron emission tomography

1. Introduction A crucial feature of the human brain is its immense learning ability. One important aspect of this learning ability is the capacity to acquire, maintain, and update a massive storage of words. In the present study, we examined the neural correlates of word acquisition as measured by picture naming. While the neural substrates of fully developed language systems have received intensive attention in both lesion studies and functional neuroimaging experiments [4,46], much less is known about the neural dynamics related

to language learning. In the present study, we explored naming-related brain activity when subjects retrieved newly learned names of unfamiliar real objects. In the following, we will firstly consider the process of naming familiar objects and its neural correlates. Secondly, we will briefly consider the effects of repeated exposure on object processing, and thirdly, review the few studies that have been conducted in the area of language learning that are directly related to the present study. Finally, we will consider associative learning and its brain correlates at a more general level. 1.1. Naming familiar objects

* Corresponding author. Fax: +358 2 2154833. E-mail address: [email protected] (P. Gro¨nholm). 0926-6410/$ - see front matter D 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.cogbrainres.2005.06.010

Oral naming of already familiar objects is a multistage process that includes early visual processing and pattern

360

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371

recognition, retrieval of semantic representations, word form retrieval, and motor speech processes [8]. Previous functional imaging studies on naming of single familiar objects have consistently revealed increased activation particularly in the occipital, the posterior inferior temporal (fusiform gyrus), and the left inferior frontal regions, as well as in the cerebellum [6,27,30,49]. Depending on the baseline task that the naming of objects is contrasted with, the brain activation patterns vary (for a review, see [19]). Semantic representations are thought to be automatically activated in object naming, even when simply viewing objects (e.g., [16]). Even though semantic memory may be widely distributed in the brain, retrieval of object meaning in picture naming has been linked particularly to activation of the fusiform gyrus (for a review, see [25]). However, it has recently been suggested that the fusiform gyrus serves as the neural substrate for the processing of an object’s structure, whereas retrieval of an object’s meaning would be related to the activity of the left posterior middle temporal gyrus [47]. In addition, certain left frontal regions appear to be involved in semantic processing, especially left inferior frontal areas (e.g., [29,35]). For example, in a PET study by Perani et al. [34] where a picture matching task (different pictures of the same object vs. pictures of a different objects) was used to study semantic processing, the discrimination of non-living entities vs. discrimination of living entities showed activation in left Brodmann area (BA) 44/45, suggesting that Broca’s area may be important for the semantic access of tools from pictures. Word form retrieval has been consistently related with activation in the left posterior superior and middle temporal gyri, whereas phonological encoding has been related to Broca’s area (for a review, see [19]). The final stage in object naming, phonetic encoding and articulation, has been found to activate several regions in studies on overt vs. covert naming, including premotor and motor areas in the frontal cortex [19]. 1.2. Processing of novel objects compared with repeated objects Given the fact that the newly learned objects in our study were repeated several times, studies of the neural correlates of repetition effects on visual object processing are of relevance here. In an fMRI study by van Turennout et al. [44], activity in bilateral occipitotemporal regions was greatly reduced for nameable objects that were repeated after a 30-s delay. After 3 days, activation was still significantly reduced within these regions. In addition, a decrease of the left inferior frontal activity was found. Several other studies on visual priming have also found a reduction of activation in posterior cortical regions when comparing repeated items with novel items (for a review, see [39]). Conceptual priming, which also relates to the present study since newly learned objects engage semantic processing, was studied by Raichle et al. [37]. In their study,

subjects were to produce appropriate verbs for visually presented nouns that had been presented and repeated beforehand. They found blood flow reductions in the left prefrontal cortex, showing that normal subjects can alter task-related brain activation patterns following less than 15 min of practice. 1.3. Naming of newly learned objects The cognitive mechanisms underlying word learning have been much debated during the past years [3,17, 24,26,45]. At issue is as to which memory and learning mechanisms are involved in vocabulary acquisition and to which degree language learning is either domain specific or non-specific. While some researchers argue strongly that working memory [3] or short-term memory [26] is essential in word learning, there are also indications of the importance of declarative memory processes [43], and associative incidental learning [38]. Research on the neural mechanisms of word learning (more specifically the retrieval of newly learned words that is the topic of the present study) has been scant in functional neuroimaging literature. McCandliss et al. [28] are among the few who have studied brain correlates of word learning over a longer period of time by employing event-related potentials (ERP). Their subjects spent 5 weeks learning an artificial language of 68 words corresponding to meaningful objects. As the lessons progressed the subjects learned not only the names of the simple objects in isolation, but also more complex sentences. ERP measurements took place before, during, and after the 5 weeks of training. McCandliss et al. [28] found learning effects in the responses at approximately 300 ms when different types of written stimuli were compared across the three ERP measurements within the semantic task (the task was to judge whether a letter string represented something tangible or not). Before training, the responses at approximately 300 ms to artificial language strings were more positive than responses to English. However, after 20 h of training, artificial language stimuli showed a shift in the direction of English stimuli, suggesting that language learning is related to processes associated with semantic access. Furthermore, an interaction between training and hemisphere was found, namely after training at about 200 ms, the amplitude was more negative over left hemisphere channels. In a recent study by Raboyeau et al. [36], Frenchspeaking subjects learned the English names of 50 familiar objects (half were animals, half were tools). Naming-related PET scanning was conducted before and after the training period. English naming after training was not contrasted directly with French naming, but instead a compound contrast between pre- and post-training activation changes in English naming vs. French naming was computed. This contrast showed increases for English and decreases for French naming that were located in the anterior cingulate cortex and frontal motor cortex bilaterally, cerebellar areas,

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371

left anterior insular cortex, and in the right collateral sulcus. Furthermore, positive learning outcome in English naming correlated with increased activity in the posterior part of the left superior temporal sulcus and middle temporal cortex (BA 21), being interpreted as efficient semantic storage of the learned words. Furthermore, Breitenstein et al. [7] studied initial acquisition of a novel lexicon with an associative, implicit learning principle by event-related fMRI. They found that vocabulary acquisition was associated with modulations of activity within the left hippocampus and the left fusiform gyrus, and with an increasing activation in the left inferior parietal cortex. The medial temporal activity changes were interpreted as formation of episodic and semantic memory traces, while the left inferior parietal activity was related to phonological storage of the newly learned items. In their study, learning success correlated with less suppression of hippocampal activity. All three abovementioned studies coupled new words with already familiar objects, and thus did not rule out the possible effects of the original name of the familiar object that needs to be inhibited during naming of newly learned words. In a recent fMRI study by James and Gauthier [20], another type of learning paradigm was used. The subjects were presented with novel visual objects that were associated with either three semantic features (e.g., sticky, loud, nocturnal) and a name (e.g., John) or only a name (e.g., Carl or David Joseph Lamont). There were two groups of subjects that received the training in a slightly different manner. After training, subjects conducted a visual matching task with the objects during brain imaging. The results showed that the processing of novel objects associated with semantic features produced more activation in the left inferior frontal cortex and the parietal cortex than either novel objects associated with a name only or untrained novel objects did. There were also brain areas that were activated in one group of subjects, but not the other (fusiform gyrus, temporal pole), being attributed to slightly different training procedures by the authors. The study that is closest to ours is a magnetoencephalography (MEG) experiment conducted by Cornelissen et al. [9], employing five healthy Finnish-speaking subjects who successfully learned real names and/or definitions of very rare objects by computerized training. Stimulus categories were as follows: unfamiliar items for which (a) both the name and definition were taught during training (SemPhon), (b) only the name was taught (Phon), (c) only the definition was taught (Sem), (d) no information was given (UnFam), and familiar items (Fam). As the to-be-learned stimuli consisted mainly of ancient farming equipment, both the objects and their names were new to the subjects. Being real but unfamiliar Finnish words, the names followed the phonotactic rules of Finnish. Training consisted of a computerized daily training sequence until the preset criterion level (98%) for naming success was reached. Learning success was unrelated to the provision of semantic information. Naming-related MEG measurements were performed before,

361

during (at the point where at least 50% of the SemPhon and Phon targets were named correctly), and after the training (98% correct), and all subjects’ data were analyzed individually. The clearest differences between the categories of items were observed when 98% of the objects were named correctly. The naming of SemPhon and Phon objects resulted in late cortical responses that grouped together. The responses differed significantly from those of the UnFam objects (where a generic name ‘‘object’’ was given during the MEG measurement), approaching the responses to familiar items. The responses of the Sem objects (where subjects also said ‘‘object’’) did not differ from the UnFam condition. The cortical learning effect occurred in the inferior parietal cortex in four out of five subjects, being left-sided in three of the subjects. Cornelissen et al. [9] suggested that the activation in the inferior parietal cortex plays a role in the acquisition of novel word forms, presumably as a part of working memory systems, reflecting phonological storage of the newly learned words. Although our paradigm was adopted from the Cornelissen et al. [9] study, there are differences that should be mentioned. Firstly, the number of subjects in their study was limited. Secondly, the number of trained objects was larger, the training was computerized, and a 98% correct criterion level was used. Thirdly, they employed a delayed naming paradigm with MEG that might enhance working memory-related processes in naming. Finally, MEG is different from PET in that it has excellent temporal resolution but reflects mainly cortical activation changes. In summary, the few available neuroimaging experiments on word learning have all suggested predominantly left hemisphere mechanisms. As regards the functional components of word learning, the phonological component has been related to inferior parietal lobe function [7,9]. With regard to the semantic component (object meaning), the results have varied between and even within experiments. The prime candidates have been the left temporal lobe (posterior midsuperior temporal regions [36], left medial temporal structures [7]), and the left inferior frontal cortex [20]. Both of these cortical regions have been related to semantic processing also in other studies (for a review, see [25]). 1.4. Associative learning mechanisms At a more general level, learning to name new objects can be seen as an associative learning task. Firstly, it requires the subjects to associate an object to a name, i.e., to link a visual pattern to a phonological representation. Secondly, learning is likely to be accompanied by self-made associations in order to facilitate memory for the names. Associative learning has been related to the ventral prefrontal cortex in a number of studies, and the ventral prefrontal cortex has therefore been suggested to be a part of a system where associations are made between visual cues and the choices that they represent (for a review, see [33]). For example, in a visuomotor learning study by Toni and Passingham [41],

362

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371

normal subjects were required to associate four visual patterns with different finger movements. The PET results revealed that learning was related to a distributed network in the ventral extrastriate and prefrontal cortex. Associative learning has also been related to cerebellar function. Lesion studies of both animals and humans have shown that a simple form of associative learning, namely eyeblink conditioning, is dependent on the cerebellum (for a review, see [11]). It is not surprising that both the prefrontal cortex and cerebellum have been related to associative learning, as they have reciprocal connections. Furthermore, the cerebellum also appears to contribute to some other cognitive functions than associative learning, such as verbal working memory, implicit and explicit learning, and language [11]. The aim of the present study was to track the namingrelated brain activity by means of PET when healthy elderly adults retrieved the names of successfully learned unfamiliar objects. The subjects trained object naming with or without semantic support, as we also wanted to explore whether provision of semantic information would show an effect on novel word learning at the behavioral and neural level.

2. Materials and methods 2.1. Participants Ten healthy, right-handed subjects (7 females and 3 males) volunteered in the experiment. The handedness of the subjects was verified by a Finnish translation of the Edinburgh Inventory (mean +0.96, range +0.89 to +1.00) [31]. Their age range was 56– 77 years (mean = 65.5 years, SD = 6.9). The reason for the subjects being elderly is that they are a part of a larger research project in which verbal memory functions in both healthy controls, persons with mild cognitive impairment, and persons with Alzheimer’s disease are investigated. Educational level varied from 5 to 15 years of formal education with the mean of 11.3 years (SD 3.9 years). All of the subjects had normal or correctedto-normal vision and none of them reported any linguistic dysfunctions, neurological illnesses, or psychiatric problems. All were native speakers of Finnish. The study was approved by the local ethical committee and all participants gave their informed written consent for their participation according to the Declaration of Helsinki. A structural T1-weighted MRI image of the brain was taken from each participant for medical inspection. None of the subjects had any hippocampal atrophy or any other visible anatomical changes that might be considered abnormal for their age. In addition, cognitive functions were assessed with an extensive neuropsychological test battery including the following tests: CERAD, Wechsler Adult Intelligence Scale—Revised (Digit Span, Block Design, Digit Symbol, Similarities), Wechsler Memory Scale—Revised, Benton Visual Retention Test-C, Trail Making A + B, Stroop, and Boston Naming Test and Word

Fluency (semantic and phonological). None of the subjects showed any cognitive deficits compared with age-appropriate norms. 2.2. Stimuli The stimuli were black and white outline drawings of non-living objects. The stimulus categories were as follows: unfamiliar but real objects for which (a) both the name and the definition were given during training (SemPhon; n = 20, Fig. 1a), (b) only the name was given (Phon; n = 20, Fig. 1b), and (c) no information was given (UnFam; n = 20, Fig. 1c). In addition, (d) real familiar non-living objects (Fam; n = 20, Fig. 1d) and (e) visual noise patterns (VNP; n = 20, Fig. 1e) were used. The unfamiliar objects were selected from the pool of ancient and other rare objects employed by Cornelissen et al. [9]. They mainly represented ancient domestic tools unknown to modern-day people. The names of the objects in the SemPhon and Phon categories were matched by word length in letters (t(38) = 0), number of syllables (t(38) = 1), number of vowels (t(38) < 1), and number of consonants (t(38) < 1) (Table 1). In order to control for the visual complexity of the four groups of objects, 10 subjects that did not participate in the study were asked to rate the objects on a scale from 1 to 5 (1 being visually not at all complex and 5 being visually very complex). A one-way ANOVA showed that the mean ratings of visual complexity did not differ between the four groups of objects (F < 1) (Table 1). In addition, we asked the subjects to write down if they associated the object (SemPhon, Phon, UnFam) with one or several familiar objects. The three groups of objects did not differ in associative potential as shown by a one-way ANOVA (F < 1) (Table 1). 2.3. Procedure 2.3.1. Pretesting The experimental design is visualized in Fig. 2. Each subject received training during 4 sessions and underwent a naming test on a 5th session (in most cases 1 day after the last training session) in order to establish the performance level before the PET scan. Sessions 1– 5 were performed on separate days within a time period of maximally 2 weeks. To ensure that the SemPhon and Phon objects were in fact unfamiliar to the subjects before training took place, they were presented for naming on the first session prior to any training. Maximally 2 names of the objects were allowed to be familiar to the subjects. The subjects were shown all the 40 objects 4 times during one training session in a pseudorandomized order. The objects were shown as a slide show made with PowerPoint (Microsoft Inc., USA), one at a time for a period of 10 s. The subjects were asked to read aloud the name of the object and if the object’s definition was also given, they were to read aloud the semantic support as well. However, the subjects were

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371

363

Fig. 1. Examples of the stimuli used in the experiment: (a) the SemPhon category where both the name and the definition of the unfamiliar object were given during training; (b) the Phon category where only the name of the unfamiliar object was given during training; (c) the UnFam category where no information was given; (d) the Fam category consisting of familiar non-living objects; and (e) the VNP category comprising of visual noise patterns.

instructed to learn only the object names provided. The experimenter was present during the whole training procedure. The reason for reading the names and definitions aloud was that the present study is a part of a larger research project, where persons with memory disturbances are

investigated. In other words, it was necessary to control that the subjects with memory problems performed the training correctly, and in order to yield comparable results with the healthy controls, they also had to perform the training in the same manner.

364

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371

Table 1 Stimulus characteristics in the SemPhon, Phon, Unfam, and Fam conditions Stimulus characteristics

SemPhon

Phon

UnFam

Fam

Word length in letters, mean (SD) Number of syllables, mean (SD) Number of vowels, mean (SD) Number of consonants, mean (SD) Visual complexity, mean (SD) Associative potential, mean (SD)

5.90 (0.85)

5.90 (0.85)

n/a

n/a

2.25 (0.44)

2.40 (0.50)

n/a

n/a

2.40 (0.50)

2.50 (0.51)

n/a

n/a

2.90 (0.72)

3.00 (0.65)

n/a

n/a

2.59 (0.92)

2.40 (1.17)

2.85 (0.77)

2.84 (1.08)

0.69 (0.29)

0.67 (0.34)

0.64 (0.21)

n/a

Each training session was preceded by a naming test where the objects were presented on the computer screen, yielding thus 3 measurements during the training (sessions 2– 4) and one after the training (on the 5th session). In the naming test, the subjects were instructed to name the object as soon as possible. They were given 10 s to name each object, and the correct answer was not provided. Furthermore, each training session (sessions 1 –4) was followed by a pointing test, where all the objects were presented on a paper sheet and the examiner pointed at the objects one at a time in a pseudorandomized order and asked the subjects to name each one. If the subjects were not able to name the object in 10 s, the correct answer was given to them. During the 5th session, the subjects performed a recognition test presented on the computer that included the UnFam objects, and a semantic test, where the subjects were asked to recall the definitions of the objects (the results will be reported elsewhere) in addition to the naming test. The Fam objects were shown on paper to the subjects for oral naming on the 5th session to make sure that all the subjects perceived them in the same way. Furthermore, a follow-up was conducted 1 week, 4 weeks, and 8 weeks after the 5th session, including the recognition test, the naming test, and the semantic test (the results of the follow-up will be reported elsewhere).

2.3.2. Stimulation parameters used in the PET session Each subject underwent 12 PET scans with 15O-water, including a rest condition (eyes open, blank screen) and 5 experimental conditions. All the conditions were presented twice, yielding two separate blocks with 6 conditions in a pseudorandomized order, so that no condition was immediately repeated. The presentation order of the two blocks was counterbalanced across participants. The experimental conditions were as follows: (1) SemPhon. The subjects were shown 20 trained SemPhon objects. The subjects were instructed to name the object aloud. If they could not retrieve the name of the object, they were instructed to stay silent and concentrate on the next object. In this and the four following conditions, each object was shown twice within a condition/scan, with no objects being immediately repeated. The objects were projected onto a screen located approximately 2 m in front of the subjects. The size of the screen was 130 cm in width and 100 cm in height. On the average, the objects (SemPhon, Phon, UnFam and Fam) subtended a visual angle of approximately 7-. All items appeared on the screen for 5 s. (2) Phon. The subjects were shown 20 trained Phon objects and were instructed to name the objects aloud. (3) UnFam. The subjects were shown 20 untrained UnFam objects. The subjects had seen them only once before the PET scanning, on the 5th session as a part of a recognition test. The subjects’ task was to say ‘‘picture’’ every time an object appeared on the screen. (4) Fam. The subjects were shown 20 familiar objects and the instruction was to name each of them. The subjects had seen and named them once before the PET scanning (on the 5th session) in order to make sure that all subjects perceived the objects in the same way. (5) VNP. The subjects were shown 20 black and white visual noise patterns and the task was to say ‘‘picture’’ every time a new random pattern appeared. In all the experimental conditions, an auditory warning signal was presented every time the object changed on the computer screen, and for the conditions 1, 2, and 4 the subjects’ responses were tape recorded.

Fig. 2. Experimental design of the study.

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371

2.3.3. PET data acquisition and processing In order to register relative changes in the regional cerebral blood flow (rCBF) between the experimental conditions, 12 emission PET scans were obtained for each subject using a GE Advance PET scanner (General Electric Medical Systems, Milwaukee, WI, USA), providing 35 transverse slices covering the entire brain and spaced 4.25 mm apart (center to center). The transmission scan for the attenuation correction was performed with a 68 Ge/68Ga source. The task that lasted approximately 3 min was initiated 15 s prior to the intravenous bolus (10 ml in 10– 15 s) administration of 300 MBq 15O-water, for each scan. Emission data were acquired in 3D mode for 90 s starting when the true coincidence rate exceeded 15 kcps. The PET data from each scan were integrated into a single frame of 90 s. Scans were separated by approximately 10 min. The images were reconstructed using a filtered back-projection algorithm into a series of 35 slices including 128  128 voxels each, yielding an in-plane pixel dimension of 2.34  2.34 mm. The PET image pre-processing and statistical analysis were performed using the Statistical Parametric Mapping software (SPM99, Wellcome Department of Cognitive Neurology, London, UK) [14] implemented in Matlab version 6.1 (Mathworks Inc., USA). In order to compensate for inter-scan head motion, a two-step image realignment procedure was performed. Each reconstructed PET image was realigned to the first image in the series and a mean of the realignment images was created. Then, all images were realigned to the mean one. After this, the realigned images were spatially normalized into a coordinate space defined by the SPM99 PET template that approximates the standard stereotactic space of the Talairach and Tournoux brain atlas [40]. An isotropic Gaussian filter of 16 mm full width at half maximum was applied to smooth each image to compensate for residual inter-individual differences in brain shape and to suppress high frequency noise in the images. An inter-scan difference in global signal was removed by proportional scaling of gray matter voxels’ values to their mean value. In the present study, the number of participants (10 subjects) was not enough for the random-effect statistical analysis. It has been shown with fMRI data that a minimum of 12 subjects is required to reach even liberal (uncorrected) threshold of P < .05 for random effect. Considering the necessity of correction for multiple comparisons, the sample size should be about twice larger [12]. On the other hand, in PET (in contrast to fMRI), the scan-to-scan variability within a PET session and the session-by-contrast interactions are about the same, and the difference between inferences based on fixed- and random-effect analyses is greatly attenuated [15]. Therefore, a fixed-effect model was used to estimate effects of conditions, which were compared to each other as linear contrasts using t statistics and the following threshold criteria: height threshold T = 4.67 corresponding to P < .05,

365

corrected for multiple non-independent comparisons [48] along with the cluster extent threshold of 50 contiguous voxels. Anatomical location of the activated foci was found by direct transferring the MNI coordinates of the rCBF peaks into the atlas of Talairach and Tournoux [40].

3. Results 3.1. Behavioral results All subjects learned the names of the objects effectively; mean percentage of correctly named objects on the 5th session was 94%. During the PET scanning, the mean percentage of correctly named trained items was 88%. A repeated measures analysis of variance (ANOVA) on the naming tests 1– 4 (sessions 2 –5) indicated that the subjects learned the names of the Phon objects better than those of the SemPhon items, yielding a significant main effect for stimulus type (F(1,9) = 14.31, P < .005; Fig. 3). As expected, the main effect for test session was also statistically significant, reflecting successful learning of the new names during the training period (F(3,27) = 104.71, P < .005, Fig. 3). The interaction term for stimulus type and test session was non-significant (F < 1). 3.2. PET results The results of the naming-related contrasts relevant to the aim of the present study are shown in Table 2. As the SemPhon condition did not differ significantly from the Phon condition (even when the cluster extent threshold was set to 0) in terms of rCBF patterns, these two conditions were pooled together forming the ‘‘Trained’’ condition.

Fig. 3. Average naming success on the SemPhon and Phon objects as measured by naming tests immediately preceding each training session (sessions 2 – 5).

366

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371

3.2.1. Trained vs. Fam and Fam vs. Trained When contrasted with the Fam condition, the Trained condition elicited rCBF increases in the left superior temporal cortex (junction of BA 22 and 38), frontal regions (Broca’s area, BA 44), and in the cerebellum. The opposite Fam vs. Trained contrast revealed rCBF increases in the primary visual area, BA 17, with emphasis on the right hemisphere (see Table 2 and Fig. 4a). 3.2.2. Trained vs. UnFam and UnFam vs. Trained When the Trained objects were contrasted with the UnFam objects, rCBF increases were found in the left prefrontal (BA9, 46/10, 47) and precentral (BA 44/6) cortices, as well as in the anterior superior temporal gyrus (BA 22). On the right hemisphere, rCBF increases were found in the anterior part of the superior temporal cortex (BA 22 and BA 38) and in the superior temporal

sulcus. Furthermore, activation was found in the cerebellum. In the UnFam vs. Trained contrast, rCBF increases were observed in the left temporoparietal (BA 39) and right orbitofrontal (BA 11) cortices (see Table 2 and Fig. 4b). 3.2.3. Fam vs. VNP The Fam objects, when compared with the visual noise patterns, elicited bilateral activation in the ventrolateral occipitotemporal cortex. The peaks were located in the left middle occipital gyrus (BA 18) and the right fusiform gyrus (BA 19) (see Table 2 and Fig. 4c, red). 3.2.4. Fam vs. Rest When compared to rest, the Fam objects elicited bilateral increases in the posterior temporal cortex (peaking in the left BA 22 and right BA 21), the cerebellum, in the right premotor (BA 6) and occipital (BA 18) regions, as well as in

Table 2 Significant increases in rCBF during naming of objects (BA = the approximation of the Brodmann’s cytoarchitectonic area (BA) according to Talairach coordinates [40]) Contrast

Brain region

SemPhon – Phon Phon – SemPhon Trained – Fam

No significant peaks No significant peaks L. Ant. Sup. Temporal L. Ant. Sup. Temporal L. Precentral L. Ant. Sup. Temporal R. Ant. Cerebellum R. Cuneus L. Sup. Temporal L. Precentral L. Post. Inf. Frontal L. Post. Mid. Frontal L. Post. Inf. Frontal R. Ant. Temporal/Inf. Frontal R. Ant. Sup. Temporal R. Ant. Mid./Sup. Temporal R. Ant. Sup. Temporal R. Ant. Sup. Temporal R. Post. Cerebellum L. Ant. Cerebellum R. Ant. Cerebellum R. Post. Cerebellum R. Post. Cerebellum L. Post. Mid. Temporal R. Rectal/Frontal L. Mid. Occipital R. Fusiform L. Ant. Sup. Temporal L. Post. Sup. Temporal L. Post. Sup. Temporal R. Mid. Temporal R. Precentral L. Post. Cerebellum L. Fusiform R. Post. Cerebellum R. Inf. Occipital

Fam – Trained Trained – UnFam

UnFam – Trained Fam – VNP Fam – Rest

BA

k

BA BA BA BA

22/38 22 44 22

BA BA BA BA BA BA BA BA BA BA BA

17 22 44/6 9 47 46/10 38/47 38 21/22/38 38 22

T

127

195 76 438

173

711

BA BA BA BA BA BA BA BA BA

39 11 18 19 22 22 22 21 6

136 190 793 1050 181 89 453 211 1383

BA 37 1319 BA 18

7.06 6.87 6.75 6.17 5.78 5.37 8.26 7.46 6.66 5.88 5.36 6.40 6.05 5.98 5.28 5.15 6.07 5.45 5.30 5.06 5.06 6.38 6.05 8.19 7.82 6.25 5.55 5.16 6.41 6.34 8.79 7.57 8.40 7.04

x

y

62 62 60 64 4 4 62 64 58 54 52 62 64 66 52 68 12 2 4 4 4 48 2 44 42 54 50 60 58 62 38 48 42 36

z

8 14 12 4 50 98 14 10 12 40 42 24 14 8 32 12 62 48 52 72 76 60 44 78 74 2 42 44 30 4 74 60 72 90

8 0 8 4 20 0 0 8 32 4 0 12 12 12 28 0 44 8 28 28 32 16 24 8 12 4 12 8 0 40 16 16 16 4

Cluster sizes (k), T values, and coordinates (x, y, and z) for maxima (bold font) and submaxima are shown. L. = the left hemisphere; R. = the right hemisphere; Inf. = inferior; Sup. = superior; Ant. = anterior; Post. = posterior; Mid. = middle.

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371

367

Fig. 4. Areas of a relative rCBF increase ( P < .05 corrected, k > 50 voxels) in the (a) Trained vs. Fam (red) and Fam vs. Trained (green) contrasts; (b) Trained vs. UnFam (red) and UnFam vs. Trained (green) contrasts; (c) Fam vs. VNP (red) and Fam vs. Rest (green) contrasts; and their overlapping areas (yellow) shown as projections onto lateral and medial reconstructed surfaces of the anatomical MRI/SPM99 brain template.

the left anterior temporal cortex (BA 22) (see Table 2 and Fig. 4c, green).

4. Discussion We set out to study naming-related brain activity when the subjects retrieved the names of successfully learned unfamiliar objects, which they had trained with or without semantic support. To our knowledge, the present study is the first one studying the retrieval of previously unfamiliar names of real rare objects by means of PET. In what follows, we will first discuss the behavioral results. Thereafter, we will focus on the brain correlates of retrieving newly learned unfamiliar objects. Finally, we will consider how our results of naming familiar objects relate to previous studies, and also briefly consider the processing of unfamiliar pictures. 4.1. The role of semantic information in retrieving the names of newly learned objects While the naming of the SemPhon and the Phon objects did not differ at the neural level, at the behavioral level, the subjects could recall somewhat better the names of the Phon than the SemPhon objects. However, in absolute terms, the difference was not great (in the last naming test before the PET scanning, subjects retrieved correctly 92% of the SemPhon object names and 96% of the Phon object names). The fact that semantic support did not facilitate word learning is somewhat surprising, as one could expect that ‘‘deep’’ or semantic processing leads to a better learning (e.g., [10]). However, learning both the name and the definition in the SemPhon condition naturally took more time than reading the name only in the Phon condition, and this might have been somewhat distracting. Furthermore, the subjects were asked to learn only the names, not the definitions, and it is always preferable to focus at the essential aspects in a task (see also [9]). Moreover, one should consider another interpretation. Behavioral evidence

suggests that in word learning, indirectly available information may even be learned more effectively than observable object features [5]. It might be that the Phon items prompted a more effortful exploration of their features, facilitating the retrieval of their names. Even though semantic support did not have a facilitating effect on learning at the behavioral level, we tested it also on the neural level. There is some evidence suggesting that provision of semantic information may have an impact on brain activation patterns during word learning. James and Gauthier [20] found that the processing of novel objects that had been trained with semantic features and a name activated different brain areas than the processing of objects that were trained with a name only. We did not find any difference in activation patterns between the SemPhon and Phon objects, which could be explained by differences in the research paradigms between the present study and the one by James and Gauthier [20]. Firstly, all novel objects in their study were highly similar and did not resemble any well-known objects, and secondly their subjects performed a matching task (they indicated by a button press if two objects were the same or different), whereas our subjects performed a language production task, i.e., retrieved the names of the objects. Training was also conducted in a different manner, and the results of the two subject groups employed by James and Gauthier [20] indicate that even slight differences in the training protocol and stimuli can yield partly different brain activation patterns. One explanation to the lack of difference between the SemPhon and Phon condition in our PET results could be the fact that, according to the subjects’ reports, they often used ‘‘self-made’’ semantic and phonological associations in both the SemPhon and the Phon condition, e.g., by associating the name of the object either to a familiar name that phonologically resembled it or to some well-known object in order to use that as a memory cue. Given the difficulty of the task, it seems plausible that the subjects used any means they had to learn the objects and their names, including self-generated associations. Finally, our

368

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371

results replicate those of Cornelissen et al. [9] who employed a similar paradigm and failed to find differences in the learning of SemPhon and Phon item names either at the behavioral or at the neural level. 4.2. Brain correlates of naming newly learned objects The naming of the Trained objects activated Broca’s area (BA 44), left anterior temporal areas (BA 22 and BA 38), and the cerebellum, when compared with the naming of the Fam objects. Broca’s area has been found to be activated in a number of neuroimaging studies on object naming (for a review, see [25]), and its activity is thought to reflect both semantic and phonological processes [35]. The activation of Broca’s area that we found when naming Trained objects can reflect one or both of the abovementioned processes. The subjects reported of self-generated meaning and formrelated associations, which suggests that they had incorporated the Trained objects and their names in pre-existing semantic and lexical –phonological networks by the time of the PET scanning. Furthermore, one could assume that the semantic retrieval associated with naming of the Fam objects is different from that of the Trained objects, as the Fam objects have been learned early on in childhood. The same argument can be made for retrieval of phonological, i.e., word form representations. In our case, it is plausible that the Trained objects require enhanced phonological retrieval, as the phonological representations of the newly learned words are far more novel compared with word forms of the familiar items that have a high word form frequency. As mentioned earlier, a crucial functional element in word learning is the phonological working memory system [3] that consists of a phonological storage and a rehearsal component (e.g., [2]). Phonological storage has been associated with left posterior parietal activity [1]. Cornelissen et al. [9], who employed a training paradigm similar to ours, found a cortical learning effect in the inferior parietal region in four out of five subjects. They suggested that this reflected the functions of the phonological storage component that was especially recruited when naming newly learned objects (see also [7]). Given the similarities between the study of Cornelissen et al. [9] and that of ours, it is somewhat unexpected that we did not find any parietal activation when our subjects named Trained objects. However, the research paradigm in these two studies was not identical. A potentially important difference was the fact that Cornelissen et al. [9] employed a delayed naming paradigm with MEG that might enhance the role of the phonological storage component in the naming task. In other words, the subjects had to keep the response in the working memory for a longer time compared with our study that did not require working memory processes to such an extent. Moreover, they had only five subjects, three of which showed learning-related activation changes in the same (left inferior parietal) region. One should also keep in

mind that MEG reflects synchronized cortical activity that does not have to map fully with regional blood flow patterns. Since MEG as a method is not directly comparable with PET, multi-method studies with the same subjects are needed to establish the correspondence between blood flow and synchronized neuromagnetic activity in verbal retrieval tasks. The temporal activation found in the Trained – Fam contrast was anterior (BA 22 and 38). Although temporal activation tends to be posterior – inferior (fusiform gyrus) in naming studies [6,27,30,49], left anterior temporal regions are occasionally found to be activated in picture naming [13,30]. Based on case studies, Holdstock et al. [18] have showed that the anterolateral temporal cortex underlies the slow acquisition of semantic information through repeated exposure, whereas the initial encoding of information depends on the hippocampus. Furthermore, based on single case reports and PET studies, Markowitsch [23] has argued for a memory retrieval system that encompasses both the anterolateral temporal and the ventrolateral frontal regions. In short, the frontal part would be related to effortful initiation of retrieval while the temporal component would serve as an important connection to more posterior areas responsible for long-term memory storage. Markowitsch [23] argues that this system is left lateralized for retrieval of semantic knowledge. The left anterior temporal activation that we observed during naming of Trained objects could thus reflect enhanced functioning of such a retrieval system. The Trained items would thus have become at least partly integrated with the existing lexical – semantic networks in the brain, rather than being represented as episodic memory traces. In other words, the retrieval of newly learned names of objects would be slowly approaching the naming of familiar objects. This fits with the findings of McCandliss et al. [28], who found a shift in processing artificial language stimuli in the direction of English stimuli after 20 h of training, as measured by ERPs. Nevertheless, it is plausible that the naming of newly learned objects requires much more retrieval effort than naming familiar objects. Retrieval effort incorporates semantic and phonological processes discussed above, as well as associative memory processes, related to the ventral prefrontal cortex (for a review, see [33]), and to the cerebellum (for a review, see [11]), i.e., areas that showed increased activation during naming of the Trained objects in our study. There are a few previous studies on neural correlates of word learning [7,20,28,36]. All of them have employed different research paradigms in comparison with our study. Nevertheless, all of the previous studies have found predominantly left-lateralized results, as was the case with our results as well. Breitenstein et al. [7] found modulations of activation in the left hippocampus in relation to vocabulary acquisition. Given their research paradigm, the results seem to reflect novel, implicit language learning, whereas our results tap slower, more explicit acquisition processes in word learning. This interpretation would also

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371

be in line with the findings of Holdstock et al. [18]. In turn, James and Gauthier [20] found activation in the inferior frontal cortex in relation to matching objects that had been learned with name and semantic features compared with objects that had been trained with only a name. On the basis of their results, one might speculate whether semantic processing in word learning is more dependent on inferior frontal regions than on middle temporal areas that are usually found to be activated in other tasks tapping semantic processing (for a review, see [25]). When the naming of the Trained objects was contrasted with the UnFam objects, we observed a more extensive bilateral activation pattern than in the Trained vs. Fam contrast. This was expected since the contrast is less specific in the way that the UnFam objects do not match with any lexical – semantic representations. As the subjects merely produced a generic response (‘‘picture’’) to the UnFam objects, it is also evident that the naming of Trained objects required more phonological processing and more complex articulations. Therefore, it seems natural to find a more extensive left frontotemporal activation, with emphasis on prefrontal areas as well as additional cerebellar activation, than in the Trained vs. Fam contrast. This probably reflects involvement of both semantic and phonological processes that were not seen to such an extent in the Trained vs. Fam contrast, in which both conditions required retrieval of unique lexical entities. The fact that we found right temporal activation in BAs 22 and 38 might reflect recruitment of more general resources for memory retrieval. 4.3. Brain correlates of naming familiar objects The Fam vs. Trained contrast revealed activation in area BA 17. The subjects had seen the Fam objects only once on paper before the scan, whereas they had seen the Trained objects repeatedly during training. In other words, the Fam objects were visually more novel to the subjects and might have required more visual processing, hence increased blood flow in area BA 17 that corresponds to the primary visual area. It is also possible that the activation reflects perceptual priming, i.e., activation decreases for the repeated Trained objects. Perceptual priming has been related to reduced activation in posterior brain areas in several studies (for a review, see [39]). When familiar objects were contrasted with visual noise patterns, large and nearly symmetric activations were found in the lateral occipital cortex peaking in the left middle occipital gyrus (BA 18) and the right fusiform gyrus (BA 19). The findings are in line with previous neuroimaging studies on naming familiar objects [6,27,30]. The results of the Fam vs. Rest contrast, showing activations in the temporal (left BA 22 and BA 37, right BA 21), right frontal (BA 6), and occipital (BA 18) lobes, as well as the cerebellum, fit well with some previous studies of naming familiar objects [6,27,30,49]. These results give further

369

support to the claim that the retrieval of newly learned names recruits specific brain areas. 4.4. Brain correlates of observing unfamiliar objects The opposite UnFam vs. Trained object contrast elicited activation in the left temporoparietal (BA 39) cortex. Activation in left posterior temporal areas (BA 39, middle temporal gyrus) has been found in previous studies on semantic memory retrieval [21] and visual word recognition [32] as a part of a larger network. It is thus possible that the activation we found reflects spontaneous semantic associations that the UnFam objects give rise to. We also found rCBF increases in BA 11 for the UnFam objects. Activation increases in either right or bilateral area BA 11 and nearby regions have been found in previous studies employing maintenance of attention [42], analogical reasoning [22], and picture naming [30]. The orbitofrontal activation may therefore reflect maintenance of attention when a routine task calls for a generic response. Another relevant issue is that for both the Fam and UnFam conditions, the subjects had seen the objects only once before the PET scan. Repetition priming has been found to reduce activity greatly in occipitotemporal regions [44]. The posterior temporal activation we found might thus reflect priming-related activation decreases for the repeated Trained objects.

5. Conclusions The results of our study suggest that the naming of newly learned unfamiliar objects entails neural processes that are partly different from the naming of familiar objects. More specifically, we observed the recruitment of a left-lateralized network that includes left-dominant frontotemporal areas and the cerebellum, when elderly subjects were engaged in naming newly learned vs. familiar objects. The semantic support given for part of the targets showed no effects on the PET results. However, the subjects’ reports indicated a common use of self-generated semantic and phonological associations as retrieval cues. Accordingly, the frontotemporo-cerebellar activation increases may reflect a number of cognitive processes that are enhanced during naming of newly learned objects: lexical – semantic and lexical – phonological retrieval, and more general associative memory mechanisms.

Acknowledgments This work was supported by a grant from the Research ˚ bo Akademi University Foundation, the Institute of the A Paulo Foundation, The Foundation for Swedish Culture in Finland, and clinical grants (EVO) of the Turku University Central Hospital.

370

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371

References [1] E. Awh, J. Jonides, E.E. Smith, E.H. Schumacher, R.A. Koeppe, S. Katz, Dissociation of storage and rehearsal in verbal working memory: evidence from positron emission tomography, Psychol. Sci. 7 (1996) 25 – 31. [2] A. Baddeley, The episodic buffer: a new component of working memory? Trends Cogn. Sci. 4 (2000) 417 – 423. [3] A. Baddeley, S. Gathercole, C. Papagno, The phonological loop as a language learning device, Psychol. Rev. 105 (1998) 158 – 173. [4] J. Binder, C.J. Price, Functional neuroimaging of language, in: R. Cabeza, A. Kingstone (Eds.), Handbook of Functional Neuroimaging of Cognition, MIT Press, Cambridge, MA, 2001, pp. 187 – 251. [5] P. Bloom, How Children Learn the Meanings of Words, The MIT Press, Cambridge, MA, 2000, pp. 25 – 53. [6] S.Y. Bookheimer, T.A. Zeffiro, T. Blaxton, W. Gaillard, W. Theodore, Regional cerebral blood flow during object naming and word reading, Hum. Brain Mapp. 3 (1995) 93 – 106. [7] C. Breitenstein, A. Jansen, M. Deppe, A-F. Foerster, J. Sommer, T. Wolbers, S. Knecht, Hippocampus activity differentiates good from poor learners of a novel lexicon, NeuroImage 25 (2005) 958 – 968. [8] D. Chialant, A. Costa, A. Caramazza, Models of naming, in: A.E. Hillis (Ed.), The Handbook of Adult Language Disorders, Psychology Press, East Sussex, Hove, UK, 2002, pp. 123 – 142. [9] K. Cornelissen, M. Laine, K. Renvall, T. Saarinen, N. Martin, R. Salmelin, Learning new names for new objects: cortical effects as measured by magnetoencephalography, Brain Lang. 89 (2004) 617 – 622. [10] F.I.M. Craik, E. Tulving, Depth of processing and the retention of words in episodic memory, J. Exp. Psychol. Gen. 194 (1975) 268 – 294. [11] J.E. Desmond, A.F. Fiez, Neuroimaging studies of the cerebellum: language, learning and memory, Trends Cogn. Sci. 2 (1998) 355 – 362. [12] J.E. Desmond, A.P. Glover, Estimating sample size in functional MRI (fMRI) neuroimaging studies: statistical power analyses, J. Neurosci. Methods 118 (2002) 115 – 128. [13] O. Etard, E. Mellet, D. Papathanassiou, K. Benali, O. Houde´, B. Mazoyer, N. Tzourio-Mazoyer, Picture naming without Broca’s and Wernicke’s area, NeuroReport 11 (1999) 617 – 622. [14] K.J. Friston, A.P. Holmes, K.J. Worsley, J.B. Poline, C.D. Frith, R.S.J. Frackowiak, Statistical parametric maps in functional imaging: a general linear approach, Hum. Brain Mapp. 2 (1995) 189 – 210. [15] H.J. Friston, A.P. Holmes, C.J. Price, C. Buchel, K.J. Worsley, Multisubject fMRI studies and conjunction analyses, NeuroImage 10 (1999) 385 – 396. [16] W.R. Glaser, Picture naming, Cognition 42 (1992) 61 – 105. [17] P. Gupta, B. MacWhinney, Vocabulary acquisition and verbal shortterm memory: computational and neural bases, Brain Lang. 59 (1997) 267 – 333. [18] J.S. Holdstock, A.R. Mayes, C.L. Isaac, Q. Gong, N. Roberts, Differential involvement of the hippocampus and temporal lobe cortices in rapid and slow learning of new semantic information, Neuropsychologia 40 (2002) 748 – 768. [19] P. Indefrey, W.J.M. Levelt, The spatial and temporal signatures of word production components, Cognition 92 (2004) 101 – 144. [20] T.W. James, I. Gauthier, Brain areas engaged during visual judgements by involuntary access to novel semantic information, Vision Res. 44 (2004) 429 – 439. [21] A.C.H. Lee, T.W. Robbins, K.S. Graham, A.M. Owen, ‘‘Pray or prey?’’ Dissociation of semantic memory retrieval from episodic memory processes using positron emission tomography and a novel homophone task, NeuroImage 16 (2002) 724 – 735. [22] Q. Luo, C. Perry, D. Peng, Z. Jin, D. Xu, G. Ding, S. Xu, The neural substrate of analogical reasoning: an fMRI study, Cogn. Brain Res. 17 (2003) 527 – 534. [23] H.J. Markowitsch, Which brain regions are critically involved in

[24] [25]

[26]

[27]

[28] [29]

[30] [31] [32]

[33]

[34]

[35]

[36]

[37]

[38]

[39] [40] [41]

[42]

[43] [44]

[45] [46]

the retrieval of old episodic memory? Brain Res. Rev. 21 (1995) 117 – 127. L. Markson, P. Bloom, Evidence against a dedicated system for word learning in children, Nature 385 (1997) 813 – 815. A. Martin, Functional neuroimaging of semantic memory, in: R. Cabeza, A. Kingstone (Eds.), Handbook of Functional Neuroimaging of Cognition, MIT Press, Cambridge, MA, 2001, pp. 153 – 186. N. Martin, P. Gupta, Exploring the relationship between word processing and verbal short-term memory: evidence from associations and dissociations, Cogn. Neuropsychol. 21 (2004) 213 – 228. A. Martin, C.L. Wiggs, L.G. Ungerleider, J.V. Haxby, Neural correlates of category-specific knowledge, Nature 379 (1996) 649 – 652. B.D. McCandliss, M.I. Posner, T. Givon, Brain plasticity in learning visual words, Cogn. Psychol. 33 (1997) 88 – 110. R.-A. Mu¨ller, N. Kleinhans, E. Courchesne, Linguistic theory and neuroimaging evidence: an fMRI study of Broca’s area in lexical sematics, Neuropsychologia 41 (2002) 1199 – 1207. S. Murtha, H. Chertkow, M. Beaugard, A. Evans, The neural substrate of picture naming, J. Cogn. Neurosci. 11 (1999) 399 – 423. R.C. Oldfield, The assessment and analysis of handedness: the Edinburgh inventory, Neuropsychologia 9 (1971) 97 – 113. K. Pammer, P.C. Hansen, M.L. Kringelbach, I. Holliday, G. Barnes, A. Hillebrand, K.D. Singh, P.L. Cornelissen, Visual word recognition: the first half second, NeuroImage 22 (2004) 1819 – 1825. R.E. Passingham, I. Toni, M.F.S. Rushworth, Specialisation within the prefrontal cortex: the ventral prefrontal cortex and associative learning, Exp. Brain Res. 133 (2000) 103 – 113. D. Perani, T. Schnur, M. Tettamanti, M. Gorlo-Tempini, S.F. Cappa, F. Fazio, Word and picture matching: a PET study of semantic category effects, Neuropsychologia 37 (1999) 293 – 306. R.A. Poldrack, A.D. Wagner, M.W. Prull, J.E. Desmond, G.H. Glover, J.D.E. Gabrieli, Functional specialization for semantic and phonological processing in the left inferior prefrontal cortex, NeuroImage 10 (1999) 15 – 35. G. Raboyeau, N. Marie, S. Balduyck, H. Gros, J.-F. De´monet, D. Cardebat, Lexical learning of the English language: a PET study in healthy French subjects, NeuroImage 22 (2004) 1808 – 1818. M.E. Raichle, J.A. Fiez, T.O. Videen, A-M.K. MacLeod, J.V. Pardo, P.T. Fox, S.E. Petersen, Practice-related changes in human brain functional anatomy during nonmotor learning, Cereb. Cortex 4 (1994) 8 – 26. J.R. Saffran, E.L. Newport, R.N. Aslin, R.A. Tunick, S. Barrueco, Incidental language learning: listening (and learning) out of the corner of your ear, Psychol. Sci. 8 (1997) 101 – 105. D.L. Schacter, R.L. Buckner, Priming and the brain, Neuron 20 (1998) 185 – 195. J. Talairach, P. Tournoux, Co-planar Stereotaxic Atlas of the Human Brain, Thieme Medical Publishers, New York, 1988, pp. 84 – 110. I. Toni, R.E. Passingham, Prefrontal – basal ganglia pathways are involved in the learning of arbitrary visuomotor associations: a PET study, Exp. Brain Res. 127 (1999) 19 – 32. T. Tsukiura, T. Fujii, T. Takahashi, R. Xiao, M. Inase, T. Iijima, A. Yamadori, J. Okuda, Neuroanatomical discrimination between manipulating and maintaining processes involved in verbal working memory: a functional MRI study, Cogn. Brain Res. 11 (2001) 13 – 21. M.T. Ullman, Contributions of memory circuits to language: the declarative/procedural model, Cognition 92 (2004) 231 – 270. M. Van Turennout, T. Ellmore, A. Martin, Long-lasting cortical plasticity in the object naming system, Nat. Neurosci. 3 (2000) 1329 – 1334. S.R. Waxman, A.E. Booth, Principles that are invoked in the acquisition of words, but not facts, Cognition 77 (2000) B33 – B43. C. Whatmough, H. Chertkow, Neuroanatomical aspects of naming, in: A.E. Hillis (Ed.), The Handbook of Adult Language Disorders, Psychology Press, New York, 2002, pp. 143 – 161.

P. Gro¨nholm et al. / Cognitive Brain Research 25 (2005) 359 – 371 [47] C. Whatmough, H. Chertkow, S. Murtha, K. Hanratty, Dissociable brain regions process object meaning and object structure during picture naming, Neuropsychologia 40 (2002) 174 – 186. [48] K.J. Worsley, S. Marrett, P. Neelin, A.C. Vandal, K.J. Friston, A.C. Evans, A unified statistical approach for determining significant

371

signals in images of cerebral activation, Hum. Brain Mapp. 4 (1996) 58 – 73. [49] B.J. Zelkowicz, A.N. Herbster, R.D. Nebes, M.A. Mintun, J.T. Becker, An examination of regional cerebral blood flow during object naming tasks, J. Int. Neuropsychol. Soc. 4 (1998) 160 – 166.