PRIMING OF OBJECT CATEGORIZATION WITHIN AND ACROSS LEVELS OF SPECIFICITY

Identification of objects can occur at different levels of specificity. Depending on task and context, an object can be classified at the superordinate level (as an animal), at the basic level (a bird) or at the subordinate level (a sparrow). What are the interactions between these representational levels and do they rely on the same sequential processes that lead to successful object identification? In this electroencephalogram study, a task-switching paradigm (covert naming or living/non-living judgment) was used. Images of objects were repeated either within the same task, or with a switch from a covert naming task to a living or non-living judgment and vice versa. While covert naming accesses entrylevel (basic or subordinate), living/non-living judgments rely on superordinate classification. Our beha-vioural results demonstrated clear priming effects within both tasks. However, asymmetries were found when task-switching had occurred, with facilitation for covert naming but not for categorization. We also found lower accuracy and early-starting and persistent enhancements of event-related potentials (ERPs) for covert naming, indicating that this task was more difficult and involved more intense perceptual and semantic processing. Perceptual priming was marked by consistent reductions of the ERP component L1 for repeated presentations, both with and without task switching. Additional repetition effects were found in early event-related activity between 150-190 ms (N1) when a repeated image had been named at initial presentation. We conclude


INTRODUCTION
Identification of objects can occur at different levels of specificity.Each object can be classified at the superordinate level, at the basic level or at the subordinate level.For example, the same image could be classified as an animal, a bird or a sparrow.In everyday vision, the majority of objects are identified at the basic level (apple, bird, car, etc.) except for some very specific exemplars from relatively diverse categories, which are identified at the subordinate level (penguin, ostrich, ambulance, etc.).This combination of subordinate and basic-level classification generally occurs when objects are named in everyday situations (Jolicoeur,et al.,984).Thus, based on the requirements of the task at hand and the context in which the object has been presented, representation of an object can occur at different levels within the category hierarchy.But what are the interactions between these different levels of representation and do they rely on the same sequential processes that lead to successful object identification?
Priming is an ideal paradigm to explore such relations because it provides an indirect measure of representational overlap.In the priming paradigm, developed by Bartram (974), categorizing an object for the second time is faster and more accurate due to the fact that the representation of the object has already been accessed (for an overview, see Palmer,999).In priming paradigms, repetition suppression of neural markers of representational processing can be observed in the event-related potentials (ERPs; R. N. Henson, et al., 2004;Penney, et al., 200) and evoked gammaband activity (GBA; Gruber & Muller, 2006; but for an opposite finding see Gruber & Müller, 2002) of the electroencephalogram (EEG).Such repetition suppression, which also impacts on the blood-oxygen level dependent (BOLD) response in functional magnetic resonance imaging (fMRI; Fiebach, et al., 2005;R. Henson, et al., 2000), has been described as a general phenomenon of neural coding specificity (for an overview, see Grill-Spector, et al., 2006).However, Van Turennout et al. (2003) demonstrated in their fMRI study that priming of object-naming involves two distinct neural mechanisms: ) repetition suppression in bilateral occipito-temporal and left inferior frontal structures and 2) repetition enhancement in the left insula and basal ganglia.Thus, monitoring the markers of priming in the EEG can demonstrate the stage of neural computations at which representations and processes employed by naming and categorisation start interacting with each other.
A recent study by Gruber et al (2006) has examined object-related priming using task-switching between a categorization (living/non-living) and a perceptual judgment task (bigger/smaller than a shoebox).It found that task-switching eliminates the behavioural benefits of priming.Meanwhile, ERP suppression effects persisted even with a task-switch, indicating that neuronal priming had occurred in structures that process perceptual stimulus features.However, Gruber et al. (2006) used two very different tasks -a perceptual judgment about object size (bigger/ smaller than a shoebox) was contrasted to superordinate classification.One could assume that a judgment about an object feature such as size would not require an activation of superordinate category knowledge, as it would be unnecessary for successful task performance.More precisely -knowing that something is an animal would not be helpful in accessing its size, and vice versa, as size tends to vary extensively within the same superordinate or basic-level category.For example, while a chihuahua or a trout are smaller than a shoebox, a great dane or a tuna would be considerably bigger.Additionally, task-switching conditions were collapsed.Thus, it was not possible to assess if there was a priming effect of size judgment on categorization, or vice versa.Instead, the authors made a conclusion about the impact of task-switching in general without looking at possible differences between the two tasks.
A recent behavioural study by Francis et al. (2008) has addressed these questions by directly comparing semantic classification (natural/manufactured judgment) with overt picture naming.They found priming within and between tasks, with between-task priming being symmetric and significantly smaller than the asymmetric within-task priming.The authors also found that these priming effects were smaller than crosslanguage priming in bilinguals.Thus, the overlap of identification processes occurring in the two tasks is likely to be incomplete, with cross-task priming seemingly not involving the highest level components of object identification.Indeed, while picture classification seems to proceed more easily than word classification, naming is slower and more attentionally demanding for pictures than for words (Carr,et al.,982).This discrepancy between classification and naming is supported by Tyler et al.'s (2004) findings with fMRI that basic-level naming of objects recruits a larger segment of the ventral stream than superordinate naming.
In this study, we undertook to further examine interactions between picture categorization and naming, using EEG to identify potential changes in the eventrelated indicators of identification and implicit memory processing.In our recent studies (Martinovic, et al., 2008a;Martinovic, et al., 2007) we have introduced covert naming tasks that are suitable for use with EEG.This has allowed us to examine event-related EEG activity elicited by objects whose identity was accessed at entrylevel.However, we have not yet directly contrasted such tasks with categorization, which has been used in the majority of previous EEG studies and provided crucial findings on event-related phenomena in object recognition.In this study, we assessed two early (P, 00-40 ms; N, 50-90 ms) and two intermediate ERP components (L, 220-300 ms; L2, 360-420 ms).It was expected that both covert naming and living/ non-living categorization would elicit similar early EEG activity (P component), which could be ascribed to low-level perceptual processing of the pictures.Possible differences between the two tasks were expected at the level of N1, as it reflects processes that utilize low-level representations to perform visual discriminations (e.g., N is enhanced for subordinate as opposed to basic-level categorisations; Tanaka,et al.,999).Further task-related enhancements for covert naming were expected at the level of the L2, which reflects the depth of semantic processing (Gruber & Müller, 2005).A replication of previously found L reductions (Gruber, et al., 2006;Gruber, et al., 2004;Gruber & Müller, 2002) for perceptual priming was also expected.

Participants
Twenty participants took part (6 male, aged 9-29 years).They were all healthy, right-handed university students and received class credit or a small honorarium for participation.Participants reported normal or corrected-to-normal vision and all were native speakers of German.None had participated in object recognition studies in preceding six months.Individual written informed consent was obtained and the study conformed to the Code of Ethics of the World Medical Association.

Materials and Procedure
Stimuli consisted of 380 coloured images of concrete objects taken from a standard photographic-quality picture library (Hemera Photo-Objects Volume , Hemera Technologies, Gatineau, Canada).The images depicted living or non-living entities whose German-language name was either of masculine or of feminine grammatical gender.The 380 images fully combined, thus providing us with 95 pictures for each combination of categories: living/masculine, living/feminine, nonliving/masculine and non-living/feminine.Stimulus presentation occurred in a random order, which was different for each of the participants.
Stimuli were presented centrally on a 9-inch computer screen with a 70 Hz refresh rate that was positioned meter in front of the participant in a dimly lit soundproof testing chamber.Participants first performed a practice block of 70 trials using stimulus images from a different set (Rossion & Pourtois, 2004).The practice block was repeated until at least 80% accuracy was reached, which usually required one repetition.Each trial started with a cue (500 ms), which instructed the participant as to the upcoming task.The cue (approximately 2° x 0.4°) was either the word NATUR (German for 'nature') or the word NAME (German for 'name').The 'nature' cue indicated that the participants had to make a 'living/non-living' decision regarding the depicted object; the 'name' cue required the grammatical gender of the object's name based on the definitive determiner ('der' or 'die') of the first correct name to come to their mind after seeing the picture.This particular covert naming task has been used in two of our previous studies (Martinovic, et al., 2008a;Martinovic, et al., 2008b).The response times obtained by such a grammatical gender decision task were found to correlate well with overt naming times.The participants were instructed to perform a grammatical gender judgment on the object's name, based on the German-language definitive determiner ("der" for masculine or "die" for feminine) of the first correct name that came to their mind after seeing the picture.Grammatical gender is a lexical property of words and is often used in implicit naming paradigms in psycholinguistics (e.g., Schmitt,et al.,200).This syntactic task is well suited to the German language, which contains three genders: masculine, feminine and neutral.
The cue was followed by a randomized 500-800 ms baseline period, during which a fixation cross (0.3° x 0.3°) was presented.
This was followed by a stimulus picture, which was displayed for 650 ms.Images of objects subtended a visual angle of approx.3.7° x 3.7°.Stimulus onset was synchronized to the vertical retrace of the monitor.
After 650 ms, the stimulus picture was replaced by the fixation cross, which remained on the screen for a period of 550ms.
Finally, an 'X' was shown for 900 ms, indicating to the participants that the trial was over and that they could blink.
Participants were instructed to react as quickly and as accurately as possible and to press one key for a 'living' or a 'masculine' stimulus and another key for a 'nonliving' and a 'feminine' stimulus.Key-to-task allocations were not counter-balanced due to the over-learned order of grammatical gender ('der' coming before 'die'); indeed, our pilots had shown that counterbalancing grammatical gender interferes with task performance.Half way through the experiment, participants were asked to change the responding hand.Participants were instructed to minimize eye movements and blinking during the display of a stimulus or the fixation cross.From the stimulus pool of 380 images, 304 were presented twice, with one or two intervening items (the number of intervening items was randomized), while the remaining 76 were used as fillers.Filler items were introduced to avoid expectations regarding the second presentation and were not further analyzed.Recurring pictures were either repeated with the same task requirements (no-switch) or with a different task (switch).This resulted in a 2x3 design comprising the factors of task (covert naming or living/non-living) and presentation (first presentation; second presentation with switching; second presentation without task switching).
In short, 76 objects were named in both presentations; 76 objects were categorized in both presentations; 76 objects were first named and then categorized; 76 objects were first categorized and then named.Therefore, the experiment itself consisted of 684 trials (608 experimental trials and 76 filler trials).These were distributed over 6 blocks, each one lasting approximately eight minutes.The assignment of stimuli to conditions was randomized and counterbalanced across the sample in order to ensure that each image was displayed in each priming condition an equal number of times.Thus, in our sample of 20 participants, each picture was assigned to be named twice by a group of 5 participants, categorized twice by another group of 5 participants, first named and then categorized by another 5 participants, and first categorized and then named by the last 5 participants.This was achieved through a script written in Matlab (Mathworks, Natick, Massachusetts) which created stimulus presentation lists for the entire sample while taking into account all the relevant factors regarding the assignment of stimuli to conditions and random presentation of stimuli.These presentation lists were then loaded at the beginning of the experiment by a script written using a MATLAB Toolbox, which also controlled the precise visual presentation and response-recording timings (Cogent; www.vislab.ucl.ac.uk/Cogent/;Mathworks, Natick, Massachusetts).Importantly, stimulus repetition was task-irrelevant and thus fulfilled the operational criteria of implicit mnemonic processing (see Gruber & Muller, 2006).

EEG Recording
EEG was recorded continuously from 28 locations using active Ag-AgCl electrodes (BioSemi Active-Two amplifier system; Biosemi, Amsterdam, The Netherlands) placed in an elastic cap.In this system the typically-used 'ground' electrodes in other EEG amplifiers are replaced through the use of two additional active electrodes, positioned in close proximity to the electrode Cz of the international 0-20 system (Jasper,958): Common Mode Sense (CMS) acts as a recording reference and Driven Right Leg (DRL) serves as ground (Metting Van Rijn,et al.,990,99).Horizontal and vertical electrooculograms were recorded in order to exclude trials with blinks and significant eye movements.
EEG signal was sampled at a rate of 52 Hz and was segmented into epochs starting 500 ms prior and lasting 500 ms following picture onset.EEG data processing was performed using the EEGlab toolbox (Delorme & Makeig, 2004) combined with in-house procedures running under the Matlab (Mathworks, Natick, Massachusetts) environment.Artifact correction was performed by means of 'statistical correction of artefacts in dense array studies' (SCADS; Junghoefer, et al., 2000).It is widely accepted in the field and has been applied and described in several publications (e.g., Gruber, et al., 999;Martinovic, et al., 2007;Muller & Keil, 2004).All incorrectly answered trials were excluded prior to data analysis.The average rejection rate was 22.8 %, resulting in approx.53 remaining trials per condition.Further analyses were performed using the average reference.

Behavioural Data Analysis
Response times (RTs) between 400 and 2300 ms, the maximum time allowed for responses, for trials with correct responses were taken into further analysis.
Median RTs for correct items were computed for each participant.Means across participants were then computed to obtain a measure of central tendency known as a mean of median RT.
Differences in response times and accuracies between the conditions were analyzed with a 2x3 repeated measures ANOVA comprising the factors of task (covert naming or living/non-living) and presentation (first presentation; second presentation with switching; second presentation without task switching).Greenhouse-Geisser correction was used when necessary due to multiple comparisons.Post-hoc tests were performed using Bonferroni-corrected paired t-tests.

Event Related Potentials Analysis
A 25 Hz low-pass filter was applied to the data before all ERP analyses.Based on previous findings on repetition priming (e.g., Gruber, et al., 2006;Gruber, et al., 2004;Gruber & Müller, 2002) we assessed four ERP components -P, N, L and L2.The analysis windows and electrode sites taken into the regional mean for each component are shown in Figure 2 (see Results).Mean amplitude within the respective time window was calculated for each component and the mean amplitude during the period 00 ms prior to stimulus onset (baseline) was subtracted.A 2x3 repeated measures ANOVA comprising the factors of task (covert naming or living/ non-living) and presentation (first presentation; second presentation with switching; second presentation without task switching).Greenhouse-Geisser correction was used when necessary due to multiple comparisons.Post-hoc tests were performed using Bonferroni-corrected paired t-tests.

Analysis of Evoked Spectral Changes
Spectral changes in oscillatory activity were analysed by means of Morlet wavelet analysis (Bertrand & Pantev,994), which provides a good compromise between time and frequency resolution 999).This approach provides a time-varying magnitude of the signal in each frequency band leading to a time by frequency (TF) representation of the signal.To that end, complex Morlet wavelets g can be generated in the time domain for different analysis frequencies f 0 according to with A' depending on the parameter σ f , specifying the width of the wavelet in the frequency domain, the analysis frequency f 0 and the user-selected ratio m: with and Thus, given a constant ratio m, the width of the wavelets in the frequency domain, σ f , and in the time domain, σ t , changes as a function of the analysis frequency f 0 .
In order to achieve good time and frequency resolution in the gamma frequency range, the wavelet family in this study was defined by a constant m= f0/σf = 7, with f0 ranging from 2.5 to 00 Hz in 0.5 Hz steps.This data was subsequently reduced to form 2.5 Hz-wide wavelets.
Evoked oscillatory activity is by definition time-and phase-locked to stimulus onset and was analyzed through a transformation of the unfiltered ERP into the frequency domain.Evoked GBA has low inter-individual variability and in object categorization studies it is usually observed at frequencies between 30 and 40 Hz, with maximal activity usually occurring in a narrow time interval around 50-50 ms post stimulus-onset (e.g., Gruber, et al., 2004;Gruber & Müller, 2005;Martinovic, et al., 2007).Therefore a ±5 Hz range was taken around a central wavelet of 35 Hz within a time window of 50-50 ms.For evoked GBA, differences between conditions () () in the amplitude after baseline subtraction were analyzed by means of a of a 2x3 repeated measures ANOVA comprising the factors of task (covert naming or living/ non-living) and presentation (first presentation; second presentation with switching; second presentation without task switching).Greenhouse-Geisser correction was used when necessary due to multiple comparisons.All post-hoc tests were conducted using Bonferroni-corrected paired t-tests.
Overall, the amount of priming with covert naming on both presentations was 253 ms.In cases where categorization was performed on the same image twice, the reduction in speed amounted to 67 ms.Finally, when covert naming was preceded by categorization, a reduction of 65 ms was obtained.Note: different voltage scales.Traditionally (although not always), EEG researchers plot negative voltages upwards and positive voltages downwards on the y-axis.Boxes indicate electrode sites included in the regional mean.The P component was maximal at occipital sites between 00 and 40 ms.There was no effect of repetition and no interaction of repetition and task.There was however a tendency for a higher P when a covert naming task had been cued.

Table 1: ERPs: outcome of statistical analyses of components
The N component was maximal at temporo-parietal sites, from 50 to 90 ms.It showed significant effects of both task and repetition.There was no interaction of these factors.N was more negative for covert naming; it also showed an enhancement when the objects had to be named on second presentation, both with and without task-switching.For categorization, a significant enhancement of the N1 for second presentations of the same objects was present for task-switching (i.e., when covert naming preceded it).There was no significant effect for categorization without a taskswitch.
The L component was maximal at occipital sites from 220 to 300 ms.It showed a tendency for increases for the covert naming task and a highly significant effect of repetition, without an interaction.
The L2 component was maximal at occipital sites from 360 to 420 ms.It was consistently higher for covert naming than for categorization.It did not show any effects of priming and no interaction between the factors.

Evoked Spectral Changes
Figure 3a shows evoked GBA grand mean baseline-corrected time-by-frequency plots (TF-plots) across 20 participants for each of the six experimental conditions at occipital sites.Figure 3b shows topographies and box plots of evoked GBA amplitudes for each condition.Evoked GBA showed no main effect of task (F/1,19/=2.02,n.s.) although it did show a tendency towards a repetition effect (F/1,19/=3.03, p=0.07).There was no interaction (F /1,19/=0.75,n.s.).

DISCUSSION
The present study investigated repetition priming effects in the human EEG, focusing on repetitions within the same task (covert naming or living/non-living categorization) as opposed to repetitions with a task switch.Covert naming and living/non-living categorization were chosen because they would elicit representational processing at different levels of specificity: entry-level for covert naming as opposed to superordinate level for categorization.Thus, the study would allow us to examine interactions between different levels of representation and to look at different ERP markers of processes (low-level image analysis, structural description stage and concept access) elicited by the two identification tasks.
Priming effects on accuracy and RTs were obtained for both covert naming and categorization if image repetition had occurred within the same task.Priming effects were larger for covert naming than for categorisation, in accordance with previous findings (for an overview, see Francis, et al., 2008).Consistent priming effects were not found when tasks were switched replicating the findings of Gruber et al. (2006) but contrary to Francis et al. (2008).In the ERPs, we found that covert naming, assessed through a grammatical gender decision task, elicited stronger activity than categorization.We also replicated previous findings that perceptual priming effects are marked by an amplitude decrease in the L component for all image repetitions regardless of task (Gruber, et al., 2004;Gruber & Müller, 2002;Penney, et al., 200, etc.).However, our study also found repetition effects at an earlier level: consistent N enhancements were observed when the second task to be performed was covert naming.N enhancements were also observed when categorization was preceded by covert naming.
A reduction in grammatical gender decision RTs which followed the categorization of the same object was observed, without a symmetrical facilitation of categorisation by covert naming.This is contrary to the findings of Francis et al. (2008), who found small (approx.30 ms) and symmetrical cross-task priming effects.Differences between this study and Francis et al. (2008) are multiple: their behavioural study had a much larger number of participants, used overt naming, employed a design with separate blocks for different tasks and repetitions and used a different stimulus set (Snodgrass & Vanderwart,980).The locus of repetition priming in picture naming is twofold: the first locus is in object recognition and the second is in name retrieval (Barry, et al., 2006).In our study, it is possible that facilitated RTs reflected only the object category prime, without the linguistic component -this is supported by a similar size of RT priming when switching from categorization to covert naming as for categorization without task-switching (approx 70 ms).However, it is also possible that covert verbal encoding had occurred on classification trials and resulted in the observed priming asymmetry.Further studies contrasting blocked and event-related designs would be necessary in order to determine the universality of RT priming effects across levels of specificity.
Accuracy differences (approximately 85% correct for covert naming, 95% for categorization) demonstrated that the covert naming task was much more difficult to perform on pictures than the categorization task, which is in accordance with Carr et al. (982).Furthermore, consistently enhanced early visual ERPs (tendency for P, N1) during covert naming are indicative of the fact that this increased difficulty was related to visual-perceptual processing.Effects of additional language processing that would need to occur in order to perform the covert naming task would not be expected to influence ERPs whose latency is shorter than approx.200 ms, as that is when effects delineating conceptual to syntactic processing are usually observed (Schmitt,et al.,200).One should not discard the role of attention either -it is possible that the tendency for enhanced processing at P1 level, reflecting the earliest perceptual stages of spatial frequency analysis and edge segmentation, was brought about by greater attentional effort when cued to perform the covert naming task.
Implicit memory effects at the level of the L component (220-300 ms) can be considered a very robust marker of perceptual priming, as they were again found to occur irrespective of task demands.However, we also found earlier priming effects -they occurred when the repeated object had previously been covertly named or was to be subsequently covertly named.These effects are not present when categorization was performed both times, replicating Gruber et al. (2006).N repetition enhancement effects have been obtained in a previous study that utilized an orientation task or an object utilization task which encouraged the analysis of 3D structure of an object (Soldan, et al., 2006).Visual discrimination processes are considered to be reflected by the N1 component (Vogel & Luck, 2000); in object identification, it is linked to lower order image classification such as perceptual grouping (Schendan & Kutas, 2007).Its sources reside in posterior ventral areas, most notably the lateral inferior occipital cortex and the posterior fusiform gyrus (Rossion, et al., 2003).Performance of a covert naming task which requires detailed analysis of the object's part-structure in order to successfully access its entry-level category has very likely been facilitated by the initial perceptual processing of the object image.Similarly, performance of a covert naming task entails a more extensive analysis of structural features of objects and thus leads to an enhanced processing at N level when the same object is presented for the second time irrespective of the task.Gruber et al. (2006) also observed a suppression of the induced GBA, but only if the repetition had occurred within the same task.Induced GBA is a high-frequency signal that can be observed in single cell recordings, EEG and magnetoencephalogram (MEG); it has been related to cortical object representation in numerous studies (e.g., Busch, et al., 2006;Tallon-Baudry & Betrand, 999) and shows reliable repetition suppression effects (Gruber, et al., 2004;Gruber & Müller, 2005).However, a recent study by Yuval-Greenberg et al. (2008) has indicated that microsaccadic or small saccadic eye movements can contaminate induced GBA in scalp recordings.According to these authors, eye movement-related activity is in fact a significant contributor to the often observed parieto-occipital peak in object identification studies.Yuval-Greenberg et al. (2008) suggest that the majority of experimental findings on induced GBA can be explained through ocular activity.However, it is doubtful that modulations of eye movement patterns can explain all of the findings on induced GBA so far (for a discussion on this topic, see comments to the Yuval-Greenberg et al. article at http:// www.cell.com/neuron/viewComment/S0896-6273(08)0030-2).A recent study by Leek and Johnston (2008) found that same regions of objects are fixated between learning and test phases, so eye movement changes are unlikely to lie behind repetition priming effects in induced GBA.Currently, several approaches are being developed for obtaining artifact-free induced GBA.However, reliable methods for the removal of potentially artifactual activity due to eye movements are not yet fully tested.

Figure
Figure depicts an excerpt of the stimulus sequence.Each trial started with a cue (500 ms), which instructed the participant as to the upcoming task.The cue (approximately 2° x 0.4°) was either the word NATUR (German for 'nature') or the word NAME (German for 'name').The 'nature' cue indicated that the participants had to make a 'living/non-living' decision regarding the depicted object; the 'name' cue required the grammatical gender of the object's name based on the definitive determiner ('der' or 'die') of the first correct name to come to their mind after seeing the picture.This particular covert naming task has been used in two of our previous studies(Martinovic, et al., 2008a;Martinovic, et al., 2008b).The response times obtained by such a grammatical gender decision task were found to correlate well with overt naming times.The participants were instructed to perform a grammatical gender judgment on the object's name, based on the German-language definitive determiner ("der" for masculine or "die" for feminine) of the first correct name that came to their mind after seeing the picture.Grammatical gender is a lexical property of words and is often used in implicit naming paradigms in psycholinguistics (e.g.,Schmitt, et al., 200).This syntactic task is well suited to the German language, which contains three genders: masculine, feminine and neutral.The cue was followed by a randomized 500-800 ms baseline period, during which a fixation cross (0.3° x 0.3°) was presented.This was followed by a stimulus picture, which was displayed for 650 ms.Images of objects subtended a visual angle of approx.3.7° x 3.7°.Stimulus onset was synchronized to the vertical retrace of the monitor.After 650 ms, the stimulus picture was replaced by the fixation cross, which remained on the screen for a period of 550ms.Finally, an 'X' was shown for 900 ms, indicating to the participants that the trial was over and that they could blink.Participants were instructed to react as quickly and as accurately as possible and to press one key for a 'living' or a 'masculine' stimulus and another key for a 'nonliving' and a 'feminine' stimulus.Key-to-task allocations were not counter-balanced due to the over-learned order of grammatical gender ('der' coming before 'die'); indeed, our pilots had shown that counterbalancing grammatical gender interferes with task performance.Half way through the experiment, participants were asked to change the responding hand.Participants were instructed to minimize eye movements and blinking during the display of a stimulus or the fixation cross.

Figure
Figure 1: Trial outlook

Figure 2
Figure 2 depicts the ERPs.Table summarises the outcome of statistical analyses.

Figure 2 :
Figure 2: a) Grand mean baseline-corrected ERP waveform at P1/L1/L2 sites averaged across electrodes.Shaded areas indicate components of interest.b) Scalp topographies of P1, N1, L1 and L2 components reflecting grand mean data averaged across all conditions.